ArneBinder/sam-pointer-bart-base-v0.3
Updated
•
65
Datasets:
document_id
stringlengths 3
3
| text
stringlengths 27.4k
72.7k
| text_bound_annotations
list | relations
list |
|---|---|---|---|
A10 | <?xml version="1.0" encoding="UTF-8" standalone="no"?>
<Document xmlns:gate="http://www.gate.ac.uk" name="A10_S12_Real-TimeWeighted_Pose-Space_Deformation_on_the_GPU_CITATION_PURPOSE_M_v1.xml">
0f19576cc37e43e4757200e19f6a45f9584905abee6902833a23266c931cee03
3wxl
http://dx.doi.org/10.1111/j.1467-6346.2008.01649.x
Volume 25 (2006), Number 3
EUROGRAPHICS 2006 / E. Gröller and L. Szirmay-Kalos (Guest Editors)
<Title>Real-Time Weighted Pose-Space Deformation on the GPU</Title>
Taehyun Rhee †1 J.P. Lewis 1 University of Southern California, U.S.A. 2 Stanford University, U.S.A.
<Abstract>WPSD (Weighted Pose Space Deformation) is an example based skinning method for articulated body animation. The per-vertex computation required in WPSD can be parallelized in a SIMD (Single Instruction Multiple Data) manner and implemented on a GPU. While such vertex-parallel computation is often done on the GPU vertex processors, further parallelism can potentially be obtained by using the fragment processors. In this paper, we develop a parallel deformation method using the GPU fragment processors. Joint weights for each vertex are automatically calculated from sample poses, thereby reducing manual effort and enhancing the quality of WPSD as well as SSD (Skeletal Subspace Deformation). We show sufficient speed-up of SSD, PSD (Pose Space Deformation) and WPSD to make them suitable for real-time applications.</Abstract>
Categories and Subject Descriptors (according to ACM CCS): I.3.1 [Computer Graphics]: Hardware ArchitectureParallel processing, I.3.5 [Computer Graphics]: Computational Geometry and Object Modeling-Curve, surface, solid and object modeling, I.3.7 [Computer Graphics]: Three-Dimensional Graphics and Realism-Animation.
c The Eurographics Association
Blackwell Publishing 2006. Published by Blackwell Publishing
Garsington Road
Oxford OX4 2DQ
UK
Main Street
Malden
MA
USA.
T. Rhee, J.P. Lewis, and U. Neumann / EG 2006
† trhee@usc.edu ‡ zilla@computer.org § uneumann@graphics.usc.edu
‡2 and Ulrich Neumann §1
<H1>1. Introduction</H1>
Skinning is an important part of realistic articulated body animation and is an important topic of computer graphics and animation. Generally, skinning can be categorized into algorithmic, physically-based, and example-based methods. Although widely used, simple algorithmic skinning schemes cannot capture the complexity and subtlety of real skin deformation, and revised approaches will be required to increase character animation realism. Physically-based skinning is based on the biomechanics of skin deformation arising from the motions of muscles and tendons. Although this approach can generate physically accurate simulations of each layer, it is not at present suitable for real time applications such as gaming due to the large computation required. Example-based methods capture some of the complexity of real skin deformation by interpolating scanned or sculpted examples of the desired skin shape in various poses. Al-
though this requires gathering a sufficient number of samples and some pre-calculation, example-based methods can potentially be used in real-time applications due to their relatively simple real-time computation. Weighted pose space deformation (WPSD) is an example based skinning method that generates high quality skinning with a limited number of sample poses
[KM04].
Although it can generate an accurate skinning, it requires more computation than the original pose space deformation (PSD)
[LCF00], since joint distances are computed independently for each vertex. As such, this method has not been suitable for real-time applications. Furthermore, both WPSD and SSD require joint weights for each vertex, and accurate joint weights are required to achieve good results. However, the weights are usually manually generated by artists, which requires effort and great skill in the case of a complex skeletal system such as the human hand. In this paper, we present a parallel WPSD algorithm (including automatic determination of joint weights) suitable for SIMD architectures such as current GPUs. The joint
weights for each vertex are automatically computed from the sample poses. This can enhance the skinning quality not only of SSD but also WPSD, since both methods require accurate joint weight values. The deformation required in WPSD and SSD is independent for each vertex and this per-vertex computation can be parallelized in a SIMD architecture. The GPU is a general SIMD architecture having one-sided (unidirectional) communication to texture memory. We demonstrate our parallel WPSD method using GPU fragment processors. In our experiments, we can speed up SSD, PSD, as well as WPSD to around 20 times faster than on the CPU (from 1.2FPS to 25FPS speed-up of WPSD on a detailed model having 22836 triangles with 11574 vertices) using a modern graphics card, thus making WPSD a feasible real-time skinning solution for various applications including games, virtual reality, and other real-time simulations.
<H1>2. Related work</H1>
Many commercial software packages generate skin deformation arising from joint movement using a method known as (linear blend) skinning, Skeletal Subspace Deformation (SSD), enveloping, etc., based in part on work published by Thalmann et al. [MTLT88]. SSD is based on the weighted blending of affine transformations of each joint and used in many real-time applications due to its simple and fast computation. However, it also exhibits some well known artifacts such as skin that collapses around the joints at increasing bend angles, and a variety of solutions for these problems have been published
[Web00, WP02, MTG03, KZ05]. Recently, example-based methods [LCF00, SRC01, ACP02, KJP02, KM04]
have permitted more complex skinning effects such as muscle bulges and major wrinkles, while also addressing the artifacts of simple algorithmic schemes. In these methods, a number of provided (scanned or sculpted) samples of the desired skin shape are simply interpolated based on the creature’s pose (and possibly additional abstract control “dimensions”). These example-based methods can also be considered as a non-parametric approach to skin deformation. In common with non-parametric sampling methods in texture synthesis (and more generally in statistical regression), the amount of memory for these methods grows with the number of training samples, but arbitrary distributions can be approximated. Some of the most impressive example-based results to date are those of Kurihara and Miyata’s hand model derived from medical images
[KM04]. Since acquiring 3D medical images is relatively expensive, they developed weighted pose space deformation (WPSD) to generate proper skinning from a limited number of pose samples. They modify the distance between poses using the joint weights of each vertex to provide a more appropriate distance measure for skinning. Although the joint weights for each vertex are important data for SSD and WPSD calculations, they have traditionally been manually generated by skilled artists. Least-squares based vertex weight estimation was shown in the skinning methods
[WP02, MTG03]. James et al. describe mesh based skinning including estimation of bone parameters and vertex weights for each bone
[JT05]. In their paper, the vertex weights of each joint are calculated by NNLS (non-negative least squares) and we derive a similar approach to calculate weights for SSD and WPSD. In recent years, since the performance of GPUs has been improving more rapidly than that of CPUs, and GPUs have many processing units serving as a SIMD parallel architecture, many algorithms have been accelerated by GPU programming [LHK ∗ 04, PF05, GPG]. Deformation and skinning algorithms can also be enhanced by GPUs and several papers have profited from this
[JP02, KJP02, BK05, JT05]. However, in previous research, since vertex information cannot be accessed in the fragment program, GPU-based vertex deformation is usually performed by vertex programs. In this paper, we develop a parallel WPSD method using the fragment processors to gain greater parallelism and performance. Person-specific data modeling and its deformation is also an interesting topic in realistic articulated body simulation. Rhee et al. described human hand modeling from surface anatomy of the person
[RNL06]. Anguelov et al. developed shape completion and animation of people, derived from the set of range scan data and example based deformation in pose and shape space [ASK ∗ 05]. Physically inspired skinning should be also recognized as another important area of articulated body animation. However, we entrust the review of the subject to the recent related papers [AHS03, CBC ∗ 05, PCLS05, SNF05].
<H1>3. Skin deformation</H1>
Example-based skinning problems can be described by the following general equation,
1
v(p a ) = S(v 0 + D(p a ))
where p a is an arbitrary pose, v(p a ) is a vertex of a deformed target surface of the arbitrary pose, v 0 is an undeformed (rest pose) vertex, S is the SSD function, and D(p a ) is a displacement as a function of the arbitrary pose. In skeletal subspace deformation the displacement D(p a ) is omitted and the target surface is calculated by SSD as a blend of affine transforms of v 0 [section 3.1]. Skinning methods related to PSD use the displacement of an arbitrary pose D(p a ), calculated by interpolation in pose space [section 3.2].
c The Eurographics Association and Blackwell Publishing 2006.
T. Rhee, J.P. Lewis, and U. Neumann / EG 2006
(a)
(b)
(d)
(e)
Figure 1: Skinning result of each algorithm: (a) SSD, (b) PSD, (c) WPSD, (d) Difference between SSD and PSD (blue dotted area), (e) Difference between SSD and WPSD (blue dotted area), (f) Difference between PSD and WPSD (blue dotted area); areas around blue and red arrows represent unexpected results of SSD and PSD respectively.
<H2>3.1. Skeletal subspace deformation (SSD)</H2>
SSD [MTLT88] is based on the weighted blending of an affine transformation of each joint by equation 2.
2
n joint v a = S(v 0 ) = ( ∑ w j T j )v 0 j=1
where n joint is the number of joints, v a is a vertex in an arbitrary pose p a , v 0 is a vertex in the rest pose, and w j is a joint weight that defines the contribution of joint j’s transformations to the current vertex. The weight w j can be assigned by the artist to control deformation and usually ∑ n j=1 joint (w j ) = 1.0. This simple algorithm is used in many commercial graphics packages and real-time rendering applications but shows several limitations, because the deformation of this method is restricted to the subspace of the affine transformation of the joints
[LCF00].
<H2>3.2. Pose space deformation (PSD)</H2>
If we have a sufficient set of examples to describe the movement of an articulated object, we can interpolate displacement in “pose space” [LCF00]. Each sample pose consists of sample skin geometry and the related joint skeleton, and a vector containing the joint angles represents the pose. If we translate each skinning sample k to the rest coordinate frame using inverse SSD, the displacement between the
(c)
(f)
sample vertex v k and the rest pose vertex can be calculated: n joint
3
d k = ( ∑ w j T j ) −1 v k − v 0 j=1
where v k is a vertex in sample pose p k and d k is the displacement of this vertex relative to v 0 in the sample pose p k ; the other variables are defined as in equation 2. Note that the inverse here is of the weighted sum of affine transforms.
After defining the displacement of each pose, the displacement at an arbitrary pose can be calculated by RBF (Radial Basis Function)
[LCF00] or normalized radial basis function
[KM04] interpolation of the example poses’ displacements. The displacement d a of a vertex in an arbitrary pose p a can be calculated as n pose
4
d a = D(p a ) = ∑ r k (p a )d k k=1
where n pose is the number of sample poses, d a is a displacement of the vertex in an arbitrary pose p a , and the weight r k (p a ) defines the contribution of each sample pose. Normalized Radial Basis Functions can smoothly interpolate pose space using: n pose
5
f t (p a ) = ∑ λ t,k φ k (γ k ) k=1
where f t (p a ) is the radial basis weight function for example t evaluated at an arbitrary pose p a , n pose is the number of sample poses, λ t,k are real valued weights between pose t and k, φ k are the radial basis functions, and γ k is the distance between the pose p k and the arbitrary pose p a (defined as the Euclidian distance between the joint vectors of each pose). The weight r k (p a ) is calculated using normalized RBFs and is used in equation 4 to calculate the displacement d a of a vertex in an arbitrary pose p a :
c The Eurographics Association and Blackwell Publishing 2006.
T. Rhee, J.P. Lewis, and U. Neumann / EG 2006
6
r k (p a ) = ∑ n pose f k (p f a (p ) )
t=1 Gaussian radial basis functions φ k (γ k ) = exp( −γ 2σ 2 k 2 ) are one possible choice of the basis and the constant σ can be specified experimentally [LCF00].
<H2>3.3. Weighted pose space deformation (WPSD)</H2>
WPSD is developed by Kurihara et al.
[KM04] to deform their example-based human hand model derived from medical images. In the original PSD, the distance between two poses p a and p k having n joint number of joints for each pose is defined as n joint
7
γ k (p a , p k ) = ∑ (p a, j − p k, j ) 2 j=1
In equation 7, since the γ k is the difference of n joint dimensional joint vectors of related poses, every vertex in the pose p k has same distance γ k resulting in the same weight r k (p a ) in every vertex of the pose p k . Furthermore, because each element of the joint vector equally contributes to the distance calculation, two vectors having a same value but different order generate same pose distance. For example, three different joint vectors p 1 = (θ, 0, 0), p 2 = (0, θ, 0), p 3 = (0, 0, θ) have same distance between them and it can cause unexpected results in PSD. In WPSD [KM04], Kurihara et al. modify the distance definition between poses using joint weight of each vertex i to give proper weight to each element of a joint vector,
8
n joint γ i,k (p a , p k ) = ∑ w i, j (p a, j − p k, j ) 2 j=1
where γ i,k (p a , p k ) is the distance between pose p a and p k of vertex i, and w i, j is weight of joint j of vertex i used in equation 2. From this definition, a more accurate pose distance is obtained and it generates better skinning in arbitrary poses, especially when the poses are far from the examples. Figure 1 shows result of three different skinning methods, but we entrust the detail comparison between quality of each algorithm to their original papers [MTLT88,LCF00,KM04].
<H1>4. Computing joint weights from samples</H1>
The joint weights of each vertex are important to generate accurate skinning in SSD (equation 2) as well as in WPSD (equation 8). In many applications, the weights are manually generated by skilled artists and it is hard to generate accurate values when a number of joints are involved in deforming a region. In this paper, we automatically calculate the joint weights of each vertex from the sample poses to enhance the accuracy of the weight value. This results in better skinning and reduces the elaborate manual work required to create weight maps. In each sample pose p k , we have following equation based on SSD: n joint
9
v ̃ k − e k = ( ∑ w j T j )v 0 j=1
where v ̃ k is a particular vertex from skin sample k, the right hand side is the SSD deformation of vertex v 0 from the rest pose, e k is a displacement between the SSD deformation and v ̃ k , and the other variables are as in equation 2. If we have sufficient examples involving the same set of n joint joints, we have n pose equations of the form: n joint
10
v ̃ k − e k = ( ∑ v j w j ) j=1
where v j is v 0 transformed by T j . Although the e k is unknown, we can solve for weights that minimize the e k in a least squares sense by stacking the equations 10 (with e k omitted) into a linear matrix system
11
2 v − Aw
where w is a n joint -dimensional weight vector, v is a 3n pose dimensional vector containing the vertex v ̃ i from every sample, and A is a 3n pose ×n joint matrix . >From equation 11, we can calculate w from the given value of v and A to reduce the error of this equation. We use the non-negative least square (NNLS) method to solve this problem and it determines positive weight values minimizing error in equation 10. The calculated weight vector w is normalized to satisfy ∑ n j=1 joint w j = 1.0. In order to avoid a singular matrix A, the number of poses should be greater or equal to the number of overall DOF (Degree Of Freedom) of the joint vector (each joint has 3 DOF), and the sample poses should be sufficiently different. James et al. used a similar approach to estimate vertex weights in each joint [JT05] and we demonstrate their efforts in our skinning method.
<H1>5. Parallel deformation on GPU</H1>
Skinning deformations vary across vertices. In SSD and WPSD, this per-vertex computation is independent for each vertex and can be parallelized by a SIMD parallel architecture. We developed a parallel skinning algorithm for SSD and WPSD that is suitable to GPUs having a SIMD architecture with one-side communication to texture memory.
c The Eurographics Association and Blackwell Publishing 2006.
T. Rhee, J.P. Lewis, and U. Neumann / EG 2006
<H2>5.1. Parallel WPSD</H2>
The computation cost of the SSD skinning algorithm is O(n vertex × n joint ) from equations 1, 2, PSD is O(n vertex × n joint × n pose ) from equations 1, 2, 4, and WPSD is O(n vertex × n joint × n pose × n pose × n pose ) from equations 1, 2, 4, 5, 6. Where, computation cost of original PSD is defined by equation 1, 2, 4, since r i is same in all vertices and d i can be pre-calculated. The number of joints n joint and poses n pose can be reduced to the smaller numbers using the method developed by Kry et al.
[KJP02], as will be discussed in section 5.2.1 with efforts to reduce texture memory space. In previous research, the Eigenskin method based on PSD was developed using GPU vertex programming [KJP02]. The vertex program uses a relatively small number of slow processing units compared with the fragment processors, and the per-vertex computation cost of the original PSD is O(n joint × n pose ). Therefore WPSD, having higher pervertex computation cost O(n joint × n pose × n pose × n pose ), can clearly benefit from parallel computation on fragment processors.
<H2>5.2. Parallel WPSD on GPU</H2>
We developed parallel skinning using the GPU fragment processors and demonstrate our method using three rendering passes. In order to minimize real-time computation, we separate possible pre-calculation steps and save the results into texture memory using texture maps. Because the value in the texture memory is not changed in the successive deformation, it can be pre-computed and stored in the read-only texture memory. In the first and second pass, per-vertex deformation is calculated in the fragment program and the results are stored in texture maps using the FBO (Frame Buffer Object) extension [Gre05]. These texture maps are bound to the geometry of the rest pose with their texture coordinates. In the third pass, each vertex in the rest pose is changed by the deformed vertex stored in the output texture generated in the first and second passes using vertex texture fetch.
<H3>5.2.1. Packing data into textures</H3>
The fragment processors cannot access vertex information. Instead, we can use texture memory to send data to the fragment program. Information needed in the fragment program is packed into texture maps and stored into texture memory. Geometry information from the rest pose is stored into two RGB texture maps, a vertex texture T v and normal texture T n ; each has size n vertex × 3. These textures represent parameter v 0 in equation 2 and each 3D element (x, y, z) is stored into the (r, g, b) value of a texel [ Figure 2 ]. The joint weights calculated in section 4 are also stored
Figure 2: Packing data into textures: texture map T i can be
T v , T n , T w1 , T w2 , and T d j . V(vertex), N(normal), W(weight), J(joint index), and D(displacement) represent each texel (RGB(A)) value of the related texture. T d consist of eight T d j storing displacements of each pose j.
in the texture maps. In general, the distribution of skinning effects in an articulated body is local to several joints
[MMT97,KJP02], even in a region as complicated as a hand. For example, deformations arising from the PIP (Proximal Interphalangeal) joint of index finger do not propagate to the other fingers, and deformation on the middle phalanx of index finger is only affected by the movement of PIP and DIP(Distal phalanx) joints. From this observation, we can reduce joint weight storage from the actual number of joint n joint to a smaller number of “principal joints” n ̃ joint selected by sorting on the weight value. We threshold n ̃ joint at four in our tests with an additional four elements to hold the related joint index. As a result, we can save the joint weights of entire geometry in two RGBA textures T w1 , T w2 each with size n vertex × 4(rgba) and store the entire information required for SSD [equation 2] in four textures T v , T n , T w1 , and T w2 . The displacement values calculated by equation 3 can be stored in n pose displacement textures; n pose is the number of sample poses. In case of complex joint structures and a large DOF model, we need many sample poses to calculate accurate joint weights and PSD deformation. However, since the joint weights can be pre-calculated, we can reduce the number of sample poses needed in real-time PSD computation. PCA (Principal Component Analysis) of pose space can yield an orthogonal basis called “ Eigendisplacement ” [KJP02]. If we reduce the size of pose space from n pose to n ̃ pose “principal poses” ( n ̃ pose < n pose ), we can reduce the number of displacement textures. In our paper, we set n ̃ pose as eight in our experiment and save displacements of all poses into a RGB texture T d having size n vertex × 8( n ̃ pose ) × 3(rgb). Therefore, from the two important observations of “principal joints” and “principal poses”, the original computation cost for SSD, PSD, and WPSD discussed in section 5.1 can be reduced using n ̃ joint and n ̃ pose rather than n joint and n pose . In the original PSD, since the weight r i in equation 4 is the same at every vertex, we do not need to calculate this value in the GPU. Since the size of this value is just n ̃ pose , we can simply pass them to the GPU as parameters without generating a texture map. Therefore, we store all the information needed to calculate the original PSD at this point. In order to reduce real-time computation, we pre-calculate T j in equation 2 and λ in equation 5 and store them into another one channel texture T x having size n ̃ pose × ( n ̃ pose + n ̃ joint × 3(x, y, z)). As a result, we store all the variables required to calculate WPSD, PSD, and SSD in six texture maps: T v , T n , T w1 , T w2 , T d , and T x . The values in the texture maps are stored in the texture memory at setup time, since they are not changed during the deformation process. In current graphic card architectures, data transfer from CPU to GPU is slow compared with memory access within the GPU. Since the only data changed in each deformation and passed from CPU to GPU is a joint vector p a (size = n joint ) representing the current arbitrary pose, the memory access rate in our method is very efficient; In the original PSD method, an additional r k value (size = n ̃ pose ) is required.
c The Eurographics Association and Blackwell Publishing 2006.
T. Rhee, J.P. Lewis, and U. Neumann / EG 2006
<H3>5.2.2. Configurations for fragment program</H3>
Variables: T out = output texture, T v = vertex texture 1 /* Set orthographic camera with same size of quad */; 2 gluOrtho2D(-1, 1, -1, 1); 3 bind(FBO); 4 /* Bind T out and set to FBO drawbuffer */; 5 bind(T out ), SetFBOdrawbuffer(T out ); 6 bind(T v ); 7 enable(fragment program); 8 /* Set viewport to the resolution of the texture */; 9 glViewport(0, 0, texWidth, texHeight); 10 /* Render a quad into T out using FBO */; 11 glBegin(GL_QUADS); 12 { glTexCoord2f(0, 0); glVertex3f(-1, -1, -0.5f); 13 glTexCoord2f(1, 0); glVertex3f( 1, -1, -0.5f); 14 glTexCoord2f(1, 1); glVertex3f( 1, 1, -0.5f); 15 glTexCoord2f(0, 1); glVertex3f(-1, 1, -0.5f); 16 }; 17 disable(fragment program); Algorithm 1: Configuration of fragment program for vertex refering and direct rendering into texture
Since vertex information cannot be accessed by the fragment program, vertex deformation on a GPU is usually performed by a vertex program
[KJP02, BK05]. Although, we cannot access vertex data in the fragment program, the efficiency of parallel computation on a fragment program is higher, since the fragment processor has more processing units and each of them has more computation power than a vertex processor. The fragment processing system is a general SIMD architecture using fragment streams as input data; each fragment is assigned to a fragment processor to calculate its final color value independently and in parallel. We developed a parallel WPSD algorithms using the fragment processors to enhance the extent of parallel computation. Geometry information like vertex positions and normals are stored in texture maps T v and T n as described in section 5.2.1 and the vertex information is referred in the fragment processors to calculate final color values. In order to assign each vertex value stored in a texture map to a fragment, we bind the geometry texture T v or T n to a quad and render it using an orthographic camera having the same width and height as the quad. Furthermore, since the viewport is set to the same resolution as the textures, each fragment is exactly matched with each texel holding the vertex information, and we can access each vertex using the texture coordinates of the fragment; vertex weights and displacements stored in the texture maps can also be accessed by similar methods. A similar idea was developed in [PBMH02] to calculate ray tracing in a fragment program and is used in GPGPU (General Purpose computation on GPUs) applications [GPG, LHK ∗ 04, PF05]. The FBO (Frame Buffer Object) extension
[Gre05]
supports rendering into an attached texture. This saves memory and time, since there is no copy operation from frame buffer to texture buffer. We implemented our WPSD algorithm using the fragment program with the FBO extension to store the result directly into texture maps accessed by vertex program in the next pass. A summary of this method is shown in Algorithm 1.
<H2>5.3. GPU implementation</H2>
We implemented GPU deformation using three rendering passes, and the basic architecture is described in figure 3 . In the first pass, we parallelize per-vertex deformation using GPU fragment processors. The data required to calculate this deformation is stored in the textures as described in section 5.2.1 and the deformation for each vertex is calculated in a fragment processor. In a given arbitrary pose defined by a joint vector, SSD is computed by equation 2 using texture maps T v , T w1 , T w2 and T x ; refer to the texture map notation in section 5.2.1. PSD is computed by equation 4 using T d , T x , after calculating r k (p a ) by equation 6. In the case of WPSD, the weighted distance is computed by equation 8 using T w1 , T w2 , and T x . In the first pass, the result of the deformation is rendered
c The Eurographics Association and Blackwell Publishing 2006.
T. Rhee, J.P. Lewis, and U. Neumann / EG 2006
Figure 3: Overview of WPSD on GPU: Each T i is the texture
map storing the required data for the calculation (refer to section 5.2.1 for their values) , T i ′ s are the texture maps to store the output of the 1st and 2nd passes, P 0 is the geometry in the rest pose, and P a is a deformed model in an arbitrary pose.
into a texture map T v ′ , using the FBO, and passed to the third pass. In the second pass we calculate and store normal deformation with a similar method as in the first pass, and the results are stored in the texture map T n ′ . In the third pass, using a vertex program, each vertex of the rest pose is transformed to the final deformed position using the information from the texture maps computed in the previous two passes. In order to access related texture information in each vertex, we created texture coordinates of each texel in pre-processing and used them in the vertex program. Specifically, the two texture maps, T v ′ and T n ′ that are generated in the first and second passes are accessed in the vertex program using the texture coordinate of the current vertex. Alternatively, multiple render targets (MRTs) can combine the first and second pass, and vertex buffer objects (VBOs) could be used to render the deformed results back to the vertex array [OPE, GPG, LHK ∗ 04].
<H1>6. Results</H1>
We tested our methods using upper arm models consisting of four joints (collar, shoulder, elbow, and wrist). Each has three DOF and the wrist is the end joint having no DOF. Three different resolution meshes are used to test the performance of GPU parallel computation: the high-resolution model has 91460 triangles with 46036 vertices, the midresolution model has 22836 triangles with 11574 vertices, and the low-resolution model has 5762 triangles with 2972 vertices [ Figure 4 ]. Note that these models are considerably more detailed than those used in current games, so the reported frame rates would be much higher if typical gameresolution models were used. On the other hand, with the expected growth of GPU processing power, models such as these will be in wide use in a few years, and algorithms such as WPSD will be required to produce realistic deformations at this level of resolution. Eight sample poses were created by Poser [Cur] and the joints weights and displacements of each sample were derived from these models [ Figure 5 ]. Our parallel algorithm is based on three pass GPU computation. The fragment program for the 1st and 2nd pass, and the vertex program for the 3rd pass are implemented in the Cg language
[FK03]. For accuracy the GPU computation is performed by 32bit floating point operations with 32bit floating point texture maps. Table 2 shows the total memory space to store texture maps required by the fragment program. Note that the maximum required memory space for the highest resolution model is just 6.8 Mbytes; the size of the output texture T v ′ and T n ′ is the same as the size of T v and T n . The results of GPU-based deformation for SSD, PSD, and WPSD are shown in Figure 1 and 6, and the experiment is performed in a GeForce 6800 Ultra GPU and a 3.4Ghz Pentium 4 CPU. The timing results of each algorithm on the CPU and GPU are summarized in table 1 . On average, our GPU-based deformation shows around 20 times speed-up compared with CPU-based deformation. GPU-based WPSD has roughly the same speed as CPUbased SSD. Therefore, real-time applications using SSD can substitute WPSD running on the GPU without loosing their real-time performance. Since our algorithm shows speed-up for SSD and PSD as well as WPSD, applications can choose the most appropriate skinning method according to the required deformation and detail.
Figure 4: Mesh of test data: the top row is a low-resolution mesh, the second row is a mid-resolution mesh, and the bottom row is a high-resolution mesh
c The Eurographics Association and Blackwell Publishing 2006.
T. Rhee, J.P. Lewis, and U. Neumann / EG 2006
Method
Mesh
CPU(FPS)
GPU(FPS)
SSD
low
150
1425
middle
39
630
high
5
164
PSD
low
98
1230
middle
23
530
high
4.5
140
WPSD
low
5
85
middle
1.2
25
high
0.29
7
Method Mesh CPU(FPS) GPU(FPS) SSD low 150 1425 middle 39 630 high 5 164 PSD low 98 1230 middle 23 530 high 4.5 140 WPSD low 5 85 middle 1.2 25 high 0.29 7
Table 1: Timing results (in FPS) of each algorithm: the lowresolution mesh has 5762 triangles with 2972 vertices, the mid-resolution mesh has 22836 triangles with 11574 vertices, and the high-resolution mesh has 91460 triangles with 46036 vertices.
Vertices
T v &T n
T w1 &T w2
T d
T x
Tot
2972 (low)
35×2
46×2
278
1
441
11574 (mid)
135×2
180×2
1080
1
1711
46036 (high)
539×2
719×2
4315
1
6832
Vertices T v &T n T w1 &T w2 T d T x Tot 2972 (low) 35×2 46×2 278 1 441 11574 (mid) 135×2 180×2 1080 1 1711 46036 (high) 539×2 719×2 4315 1 6832
Table 2: Texture memory to store data required in fragment program (in Kbytes); refer to section 5.2.1 for texture notation .
<H1>7. Conclusions</H1>
In this paper, we present a parallel skinning algorithm suitable for SIMD architectures such as GPUs. The joint weights of each vertex are automatically computed by NNLS and used in the skinning computation for SSD and WPSD. Independent per-vertex deformation is parallelized on the GPU using three rendering passes. In the first and second passes, per-vertex deformation is calculated by the fragment processors and the results are stored in texture maps using FBO. In the third pass, using vertex processors, each vertex of the rest pose is changed by the deformed vertex stored in the textures generated by the first and second passes. Articulated body skinning using SSD, PSD, and WPSD are efficiently parallelized by our GPU-based method, and on a detailed model, we obtain around 20 times speed-up compared with CPU-based computation. Principal component compression of the examples and careful analysis of joint distributions can reduce the domain of computation
[KJP02] and other algorithms based on the SSD, PSD, and shape interpolation may be parallelized on GPU using our approach.
Figure 6: Arbitrary poses deformed by WPSD on GPU
<H1>Acknowledgments</H1>
This research has been funded by the Integrated Media System Center/USC, and Samsung Electronics. We wish to thank KyungKook Park, Changki Min, and Tim Foley for discussions about GPUs, and the anonymous reviewers for their sincere comments.
<H1>References</H1>
[ACP02] ALLEN B., CURLESS B., POPOVI ́ c; Z.: Articulated body deformation from range scan data. In SIG-GRAPH ’02: Proceedings of the 29th annual conference on Computer graphics and interactive techniques (New York, NY, USA, 2002), ACM Press, pp.
612–619.
[AHS03] ALBRECHT I., HABER J., SEIDEL H. P.: Construction and animation of anatomically based human hand models. In Proceedings of the 2003 ACM SIGGRAPH/Eurographics Symposium on Computer Animation (SCA-03) (2003), pp. 98–109.
[ASK05] ANGUELOV D., SRINIVASAN P., KOLLER D., THRUN S., RODGERS J., DAVIS J.: Scape: shape completion and animation of people. ACM Trans. Graph. 24, 3 (2005), 408–416.
[BK05] BOTSCH M., KOBBELT L.: Real-time shape editing using radial basis functions. Computer Graphics Forum 24, 3 (2005), 611–621. (Proceedings of Eurographics 2005).
[CBC05] CAPELL S., BURKHART M., CURLESS B., DUCHAMP T., POPOVI ́ c; Z.: Physically based rigging for deformable characters. In SCA ’05: Proceedings of the 2005 ACM SIGGRAPH/Eurographics symposium on Computer animation (New York, NY, USA, 2005), ACM Press, pp. 301–310. [Cur] CURIOUSLAB: Poser 6. http://www.curiouslabs.com .
[FK03] FERNANDO R., KILGARD M. J.: The Cg Tutorial; The Definitive Guide to Programmable Real-Time Graphics. Addison-Wesley, 2003. [GPG] GPGPU: General-purpose computation using graphics hardware. http://gpgpu.org .
[Gre05] GREEN S.: The OpenGL Framebuffer Object Extension. Game Developers Conference, 2005. http://developer.nvidia.com/object/gdc_2005_ presentations.html.
[JP02] JAMES D. L., PAI D. K.: Dyrt: dynamic response textures for real time deformation simulation with graphics hardware. In SIGGRAPH ’02: Proceedings of the 29 th annual conference on Computer graphics and interactive techniques (New York, NY, USA, 2002), ACMPress, pp. 582–585.
[JT05] JAMES D. L., TWIGG C. D.: Skinning mesh animations. ACM Trans. Graph. 24, 3 (2005), 399–407.
[KJP02] KRY P. G., JAMES D. L., PAI D. K.: Eigen Skin: Real time large deformation character skinning in hardware. In Proceedings of the 2002 ACM SIGGRAPH Symposium on Computer Animation (SCA-02) (2002), pp. 153–160.
[KM04] KURIHARA T., MIYATA N.: Modeling deformable human hands from medical images. In Proceedings of the 2004 ACM SIGGRAPH Symposium on Computer Animation (SCA-04) (2004), pp. 357–366.
[KZ05] KAVAN L., ZARA J.: Spherical blend skinning: A real-time deformation of articulated models. In 2005 ACMSIGGRAPH Symposium on Interactive 3D Graphics and Games (April 2005), ACM Press, pp. 9–16.
[LCF00] LEWIS J. P., CORDNER M., FONG N.: Pose space deformation: a unified approach to shape interpolation and skeleton-driven deformation. In SIGGRAPH ’00: Proceedings of the 27th annual conference on Computer graphics and interactive techniques (New York, NY, USA, 2000), ACMPress/Addison-Wesley Publishing Co., pp. 165–172.
[LHK04] LUEBKE D., HARRIS M., KRUGER J., PURCELLT., GOVINDARAJU N., BUCK I., WOOLLEY C., LEFOHN A.: Gpgpu: general purpose computation ongraphics hardware. In GRAPH ’04: Proceedings of the conference on SIGGRAPH 2004 course
notes (New York, NY, USA, 2004), ACM Press, p. 33.
[MMT97] MOCCOZET L., MAGNENAT-THALMANN N.: Dirichlet free-form deformations and their application to hand simulation. In Computer Animation (1997).
[MTG03] MOHR A., TOKHEIM L., GLEICHER M.: Direct manipulation of interactive character skins. In SI3D ’03: Proceedings of the 2003 symposium on Interactive 3D graphics (New York, NY, USA, 2003), ACM Press, pp. 27–30.
[MTLT88] MAGNENAT-THALMANN N., LAPERRIÈRE R., THALMANN D.: Joint dependent local deformations for hand animation and object grasping. In Graphics In terface ’88 (June 1988), pp. 26–33.
[OPE] OPENGL: Opengl extension registry. http:// oss.sgi.com/projects/ogl sample/registry/.
[PBMH02] PURCELL T. J., BUCK I., MARK W. R., HANRAHAN P.: Ray tracing on programmable graphics hardware. ACM Transactions on Graphics 21, 3 (July 2002), 703–
712. ISSN 0730-0301 (Proceedings of ACM SIGGRAPH 2002).
[PCLS05] PRATSCHER M., COLEMAN P., LASZLO J., SINGH K.: Outside-in anatomy based character rigging. In SCA ’05: Proceedings of the 2005 ACM SIGGRAPH/Eurographics symposium on Computer animation (New York, NY, USA, 2005), ACM Press, pp. 329–338.
[PF05] PHARR M., FERNANDO R.: GPU Gems 2; Programming Techniques for HighPerformance Graphics and General-Purpose Computation. Addison-Wesley, 2005.
[RNL06] RHEE T., NEUMANN U., LEWIS J.: Human hand modeling from surface anatomy. In I3DG ’06: Proc. of ACM SIGGRAPH Symposium on Interactive 3D Graphics and Games (2006).
[SNF05] SIFAKIS E., NEVEROV I., FEDKIW R.: Automatic determination of facial muscle activations from sparse motion capture marker data. ACM Trans. Graph. 24, 3 (2005), 417–425.
[SRC01] SLOAN P.-P. J., ROSE C. F., COHEN M. F.: Shape by example. In SI3D ’01: Proceedings of the 2001 symposium on Interactive 3D graphics (New York, NY, USA, 2001), ACM Press, pp. 135–143.
[Web00] WEBER J.: Run-time skin deformation. In Proceedings of Game Developers Conference (2000).
[WP02] WANG X. C., PHILLIPS C.: Multi-weight enveloping: least-squares approximation techniques for skin animation. In SCA ’02: Proceedings of the 2002 ACM
SIGGRAPH/Eurographics symposium on Computer animation (New York, NY, USA, 2002),
ACM Press, pp. 129–138.
Figure 5: Samples poses and displacements: the first row shows each sample poses, the second row shows displacement of each sample pose with the rest pose in the first column of the third row. Please enlarge to see details.
c The Eurographics Association and Blackwell Publishing 2006.
</Document>
| [
{
"offsets": [
[
2452,
2582
]
],
"text": "['Skinning is an important part of realistic articulated body animation and is an important topic of computer graphics and animation']",
"type": "background_claim",
"id": "T1"
},
{
"offsets": [
[
2595,
2684
]
],
"text": "['skinning can be categorized into algorithmic, physically-based, and example-based methods']",
"type": "background_claim",
"id": "T2"
},
{
"offsets": [
[
2708,
2811
]
],
"text": "['simple algorithmic skinning schemes cannot capture the complexity and subtlety of real skin deformation']",
"type": "background_claim",
"id": "T3"
},
{
"offsets": [
[
2695,
2706
]
],
"text": "['widely used']",
"type": "background_claim",
"id": "T4"
},
{
"offsets": [
[
2817,
2892
]
],
"text": "['revised approaches will be required to increase character animation realism']",
"type": "background_claim",
"id": "T5"
},
{
"offsets": [
[
2894,
3016
]
],
"text": "['Physically-based skinning is based on the biomechanics of skin deformation arising from the motions of muscles and tendons']",
"type": "background_claim",
"id": "T6"
},
{
"offsets": [
[
3027,
3099
]
],
"text": "['this approach can generate physically accurate simulations of each layer']",
"type": "background_claim",
"id": "T7"
},
{
"offsets": [
[
3101,
3157
]
],
"text": "['it is not at present suitable for real time applications']",
"type": "background_claim",
"id": "T8"
},
{
"offsets": [
[
3180,
3210
]
],
"text": "['the large computation required']",
"type": "background_claim",
"id": "T9"
},
{
"offsets": [
[
3212,
3378
]
],
"text": "['Example-based methods capture some of the complexity of real skin deformation by interpolating scanned or sculpted examples of the desired skin shape in various poses']",
"type": "background_claim",
"id": "T10"
},
{
"offsets": [
[
3397,
3476
]
],
"text": "['this requires gathering a sufficient number of samples and some pre-calculation']",
"type": "background_claim",
"id": "T11"
},
{
"offsets": [
[
3478,
3549
]
],
"text": "['example-based methods can potentially be used in real-time applications']",
"type": "background_claim",
"id": "T12"
},
{
"offsets": [
[
3557,
3602
]
],
"text": "['their relatively simple real-time computation']",
"type": "background_claim",
"id": "T13"
},
{
"offsets": [
[
3604,
3635
]
],
"text": "['Weighted pose space deformation']",
"type": "background_claim",
"id": "T14"
},
{
"offsets": [
[
3643,
3753
]
],
"text": "['is an example based skinning method that generates high quality skinning with a limited number of sample poses']",
"type": "background_claim",
"id": "T15"
},
{
"offsets": [
[
3756,
3760
]
],
"text": "['KM04']",
"type": "data",
"id": "T16"
},
{
"offsets": [
[
3773,
3809
]
],
"text": "['it can generate an accurate skinning']",
"type": "background_claim",
"id": "T17"
},
{
"offsets": [
[
3811,
3880
]
],
"text": "['it requires more computation than the original pose space deformation']",
"type": "background_claim",
"id": "T18"
},
{
"offsets": [
[
3889,
3894
]
],
"text": "['LCF00']",
"type": "data",
"id": "T19"
},
{
"offsets": [
[
3903,
3961
]
],
"text": "['joint distances are computed independently for each vertex']",
"type": "background_claim",
"id": "T20"
},
{
"offsets": [
[
3972,
4032
]
],
"text": "['this method has not been suitable for real-time applications']",
"type": "background_claim",
"id": "T21"
},
{
"offsets": [
[
4047,
4102
]
],
"text": "['both WPSD and SSD require joint weights for each vertex']",
"type": "background_claim",
"id": "T22"
},
{
"offsets": [
[
4108,
4167
]
],
"text": "['accurate joint weights are required to achieve good results']",
"type": "background_claim",
"id": "T23"
},
{
"offsets": [
[
4178,
4231
]
],
"text": "['the weights are usually manually generated by artists']",
"type": "background_claim",
"id": "T24"
},
{
"offsets": [
[
4239,
4270
]
],
"text": "['requires effort and great skill']",
"type": "background_claim",
"id": "T25"
},
{
"offsets": [
[
4286,
4311
]
],
"text": "['a complex skeletal system']",
"type": "data",
"id": "T26"
},
{
"offsets": [
[
4351,
4387
]
],
"text": "['we present a parallel WPSD algorithm']",
"type": "own_claim",
"id": "T27"
},
{
"offsets": [
[
4441,
4472
]
],
"text": "['suitable for SIMD architectures']",
"type": "own_claim",
"id": "T28"
},
{
"offsets": [
[
4495,
4504
],
[
4505,
4580
]
],
"text": "['The joint', '']",
"type": "own_claim",
"id": "T29"
},
{
"offsets": [
[
4582,
4649
]
],
"text": "['This can enhance the skinning quality not only of SSD but also WPSD']",
"type": "own_claim",
"id": "T30"
},
{
"offsets": [
[
4657,
4706
]
],
"text": "['both methods require accurate joint weight values']",
"type": "background_claim",
"id": "T31"
},
{
"offsets": [
[
4708,
4779
]
],
"text": "['The deformation required in WPSD and SSD is independent for each vertex']",
"type": "background_claim",
"id": "T32"
},
{
"offsets": [
[
4784,
4854
]
],
"text": "['this per-vertex computation can be parallelized in a SIMD architecture']",
"type": "background_claim",
"id": "T33"
},
{
"offsets": [
[
5053,
5136
]
],
"text": "['we can speed up SSD, PSD, as well as WPSD to around 20 times faster than on the CPU']",
"type": "own_claim",
"id": "T34"
},
{
"offsets": [
[
5138,
5238
]
],
"text": "['from 1.2FPS to 25FPS speed-up of WPSD on a detailed model having 22836 triangles with 11574 vertices']",
"type": "data",
"id": "T35"
},
{
"offsets": [
[
5240,
5268
]
],
"text": "['using a modern graphics card']",
"type": "own_claim",
"id": "T36"
},
{
"offsets": [
[
5275,
5350
]
],
"text": "['making WPSD a feasible real-time skinning solution for various applications']",
"type": "own_claim",
"id": "T37"
},
{
"offsets": [
[
5361,
5366
]
],
"text": "['games']",
"type": "data",
"id": "T38"
},
{
"offsets": [
[
5368,
5383
]
],
"text": "['virtual reality']",
"type": "data",
"id": "T39"
},
{
"offsets": [
[
5389,
5416
]
],
"text": "['other real-time simulations']",
"type": "data",
"id": "T40"
},
{
"offsets": [
[
5466,
5650
]
],
"text": "['Many commercial software packages generate skin deformation arising from joint movement using a method known as (linear blend) skinning, Skeletal Subspace Deformation (SSD), enveloping']",
"type": "background_claim",
"id": "T41"
},
{
"offsets": [
[
5658,
5707
]
],
"text": "['based in part on work published by Thalmann et al']",
"type": "background_claim",
"id": "T42"
},
{
"offsets": [
[
5710,
5716
]
],
"text": "['MTLT88']",
"type": "data",
"id": "T43"
},
{
"offsets": [
[
5719,
5796
]
],
"text": "['SSD is based on the weighted blending of affine transformations of each joint']",
"type": "background_claim",
"id": "T44"
},
{
"offsets": [
[
5801,
5836
]
],
"text": "['used in many real-time applications']",
"type": "background_claim",
"id": "T45"
},
{
"offsets": [
[
5844,
5875
]
],
"text": "['its simple and fast computation']",
"type": "background_claim",
"id": "T46"
},
{
"offsets": [
[
5886,
5928
]
],
"text": "['it also exhibits some well known artifacts']",
"type": "background_claim",
"id": "T47"
},
{
"offsets": [
[
5937,
6000
]
],
"text": "['skin that collapses around the joints at increasing bend angles']",
"type": "data",
"id": "T48"
},
{
"offsets": [
[
6006,
6067
]
],
"text": "['a variety of solutions for these problems have been published']",
"type": "background_claim",
"id": "T49"
},
{
"offsets": [
[
6070,
6075
]
],
"text": "['Web00']",
"type": "data",
"id": "T50"
},
{
"offsets": [
[
6077,
6081
]
],
"text": "['WP02']",
"type": "data",
"id": "T51"
},
{
"offsets": [
[
6083,
6088
]
],
"text": "['MTG03']",
"type": "data",
"id": "T52"
},
{
"offsets": [
[
6090,
6094
]
],
"text": "['KZ05']",
"type": "data",
"id": "T53"
},
{
"offsets": [
[
6097,
6128
]
],
"text": "['Recently, example-based methods']",
"type": "background_claim",
"id": "T54"
},
{
"offsets": [
[
6165,
6209
]
],
"text": "['have permitted more complex skinning effects']",
"type": "background_claim",
"id": "T55"
},
{
"offsets": [
[
6130,
6135
]
],
"text": "['LCF00']",
"type": "data",
"id": "T56"
},
{
"offsets": [
[
6137,
6142
]
],
"text": "['SRC01']",
"type": "data",
"id": "T57"
},
{
"offsets": [
[
6144,
6149
]
],
"text": "['ACP02']",
"type": "data",
"id": "T58"
},
{
"offsets": [
[
6151,
6156
]
],
"text": "['KJP02']",
"type": "data",
"id": "T59"
},
{
"offsets": [
[
6158,
6162
]
],
"text": "['KM04']",
"type": "data",
"id": "T60"
},
{
"offsets": [
[
6218,
6231
]
],
"text": "['muscle bulges']",
"type": "data",
"id": "T61"
},
{
"offsets": [
[
6236,
6250
]
],
"text": "['major wrinkles']",
"type": "data",
"id": "T62"
},
{
"offsets": [
[
6252,
6317
]
],
"text": "['while also addressing the artifacts of simple algorithmic schemes']",
"type": "background_claim",
"id": "T63"
},
{
"offsets": [
[
6741,
6821
]
],
"text": "['the amount of memory for these methods grows with the number of training samples']",
"type": "background_claim",
"id": "T64"
},
{
"offsets": [
[
6827,
6870
]
],
"text": "['arbitrary distributions can be approximated']",
"type": "background_claim",
"id": "T65"
},
{
"offsets": [
[
6872,
7003
]
],
"text": "['Some of the most impressive example-based results to date are those of Kurihara and Miyata’s hand model derived from medical images']",
"type": "background_claim",
"id": "T66"
},
{
"offsets": [
[
7006,
7010
]
],
"text": "['KM04']",
"type": "data",
"id": "T67"
},
{
"offsets": [
[
7019,
7070
]
],
"text": "['acquiring 3D medical images is relatively expensive']",
"type": "background_claim",
"id": "T68"
},
{
"offsets": [
[
7072,
7118
]
],
"text": "['they developed weighted pose space deformation']",
"type": "background_claim",
"id": "T69"
},
{
"offsets": [
[
7126,
7191
]
],
"text": "['to generate proper skinning from a limited number of pose samples']",
"type": "background_claim",
"id": "T70"
},
{
"offsets": [
[
7193,
7330
]
],
"text": "['They modify the distance between poses using the joint weights of each vertex to provide a more appropriate distance measure for skinning']",
"type": "background_claim",
"id": "T71"
},
{
"offsets": [
[
7341,
7423
]
],
"text": "['the joint weights for each vertex are important data for SSD and WPSD calculations']",
"type": "background_claim",
"id": "T72"
},
{
"offsets": [
[
7425,
7491
]
],
"text": "['they have traditionally been manually generated by skilled artists']",
"type": "background_claim",
"id": "T73"
},
{
"offsets": [
[
7493,
7571
]
],
"text": "['Least-squares based vertex weight estimation was shown in the skinning methods']",
"type": "background_claim",
"id": "T74"
},
{
"offsets": [
[
7574,
7578
]
],
"text": "['WP02']",
"type": "data",
"id": "T75"
},
{
"offsets": [
[
7580,
7585
]
],
"text": "['MTG03']",
"type": "data",
"id": "T76"
},
{
"offsets": [
[
7588,
7702
]
],
"text": "['James et al. describe mesh based skinning including estimation of bone parameters and vertex weights for each bone']",
"type": "background_claim",
"id": "T77"
},
{
"offsets": [
[
7705,
7709
]
],
"text": "['JT05']",
"type": "data",
"id": "T78"
},
{
"offsets": [
[
7712,
7783
]
],
"text": "['In their paper, the vertex weights of each joint are calculated by NNLS']",
"type": "background_claim",
"id": "T79"
},
{
"offsets": [
[
7817,
7883
]
],
"text": "['we derive a similar approach to calculate weights for SSD and WPSD']",
"type": "own_claim",
"id": "T80"
},
{
"offsets": [
[
7908,
7981
]
],
"text": "['the performance of GPUs has been improving more rapidly than that of CPUs']",
"type": "background_claim",
"id": "T81"
},
{
"offsets": [
[
7987,
8058
]
],
"text": "['GPUs have many processing units serving as a SIMD parallel architecture']",
"type": "background_claim",
"id": "T82"
},
{
"offsets": [
[
8060,
8116
]
],
"text": "['many algorithms have been accelerated by GPU programming']",
"type": "background_claim",
"id": "T83"
},
{
"offsets": [
[
8118,
8126
]
],
"text": "['LHK ∗ 04']",
"type": "data",
"id": "T84"
},
{
"offsets": [
[
8128,
8132
]
],
"text": "['PF05']",
"type": "data",
"id": "T85"
},
{
"offsets": [
[
8134,
8137
]
],
"text": "['GPG']",
"type": "data",
"id": "T86"
},
{
"offsets": [
[
8140,
8204
]
],
"text": "['Deformation and skinning algorithms can also be enhanced by GPUs']",
"type": "background_claim",
"id": "T87"
},
{
"offsets": [
[
8250,
8254
]
],
"text": "['JP02']",
"type": "data",
"id": "T88"
},
{
"offsets": [
[
8256,
8261
]
],
"text": "['KJP02']",
"type": "data",
"id": "T89"
},
{
"offsets": [
[
8263,
8267
]
],
"text": "['BK05']",
"type": "data",
"id": "T90"
},
{
"offsets": [
[
8269,
8273
]
],
"text": "['JT05']",
"type": "data",
"id": "T91"
},
{
"offsets": [
[
8313,
8374
]
],
"text": "['vertex information cannot be accessed in the fragment program']",
"type": "background_claim",
"id": "T92"
},
{
"offsets": [
[
8376,
8444
]
],
"text": "['GPU-based vertex deformation is usually performed by vertex programs']",
"type": "background_claim",
"id": "T93"
},
{
"offsets": [
[
8285,
8305
]
],
"text": "['in previous research']",
"type": "background_claim",
"id": "T94"
},
{
"offsets": [
[
8461,
8568
]
],
"text": "['we develop a parallel WPSD method using the fragment processors to gain greater parallelism and performance']",
"type": "own_claim",
"id": "T95"
},
{
"offsets": [
[
8570,
8689
]
],
"text": "['Person-specific data modeling and its deformation is also an interesting topic in realistic articulated body simulation']",
"type": "background_claim",
"id": "T96"
},
{
"offsets": [
[
8691,
8767
]
],
"text": "['Rhee et al. described human hand modeling from surface anatomy of the person']",
"type": "background_claim",
"id": "T97"
},
{
"offsets": [
[
8770,
8775
]
],
"text": "['RNL06']",
"type": "data",
"id": "T98"
},
{
"offsets": [
[
8778,
8939
]
],
"text": "['Anguelov et al. developed shape completion and animation of people, derived from the set of range scan data and example based deformation in pose and shape space']",
"type": "background_claim",
"id": "T99"
},
{
"offsets": [
[
8941,
8949
]
],
"text": "['ASK ∗ 05']",
"type": "data",
"id": "T100"
},
{
"offsets": [
[
8952,
9062
]
],
"text": "['Physically inspired skinning should be also recognized as another important area of articulated body animation']",
"type": "background_claim",
"id": "T101"
},
{
"offsets": [
[
9140,
9145
]
],
"text": "['AHS03']",
"type": "data",
"id": "T102"
},
{
"offsets": [
[
9147,
9155
]
],
"text": "['CBC ∗ 05']",
"type": "data",
"id": "T103"
},
{
"offsets": [
[
9157,
9163
]
],
"text": "['PCLS05']",
"type": "data",
"id": "T104"
},
{
"offsets": [
[
9165,
9170
]
],
"text": "['SNF05']",
"type": "data",
"id": "T105"
},
{
"offsets": [
[
9232,
9314
]
],
"text": "['Example-based skinning problems can be described by the following general equation']",
"type": "background_claim",
"id": "T106"
},
{
"offsets": [
[
9347,
9373
]
],
"text": "['v(p a ) = S(v 0 + D(p a ))']",
"type": "background_claim",
"id": "T107"
},
{
"offsets": [
[
9626,
9777
]
],
"text": "['In skeletal subspace deformation the displacement D(p a ) is omitted and the target surface is calculated by SSD as a blend of affine transforms of v 0']",
"type": "background_claim",
"id": "T108"
},
{
"offsets": [
[
9779,
9790
]
],
"text": "['section 3.1']",
"type": "data",
"id": "T109"
},
{
"offsets": [
[
9793,
9917
]
],
"text": "['Skinning methods related to PSD use the displacement of an arbitrary pose D(p a ), calculated by interpolation in pose space']",
"type": "background_claim",
"id": "T110"
},
{
"offsets": [
[
9919,
9930
]
],
"text": "['section 3.2']",
"type": "data",
"id": "T111"
},
{
"offsets": [
[
10602,
10605
]
],
"text": "['SSD']",
"type": "background_claim",
"id": "T112"
},
{
"offsets": [
[
10607,
10613
]
],
"text": "['MTLT88']",
"type": "data",
"id": "T113"
},
{
"offsets": [
[
10615,
10702
]
],
"text": "['is based on the weighted blending of an affine transformation of each joint by equation']",
"type": "background_claim",
"id": "T114"
},
{
"offsets": [
[
10737,
10781
]
],
"text": "['n joint v a = S(v 0 ) = ( ∑ w j T j )v 0 j=1']",
"type": "background_claim",
"id": "T115"
},
{
"offsets": [
[
11095,
11129
]
],
"text": "['usually ∑ n j=1 joint (w j ) = 1.0']",
"type": "background_claim",
"id": "T116"
},
{
"offsets": [
[
11023,
11090
]
],
"text": "['The weight w j can be assigned by the artist to control deformation']",
"type": "background_claim",
"id": "T117"
},
{
"offsets": [
[
11131,
11234
]
],
"text": "['This simple algorithm is used in many commercial graphics packages and real-time rendering applications']",
"type": "background_claim",
"id": "T118"
},
{
"offsets": [
[
11239,
11264
]
],
"text": "['shows several limitations']",
"type": "background_claim",
"id": "T119"
},
{
"offsets": [
[
11274,
11377
]
],
"text": "['the deformation of this method is restricted to the subspace of the affine transformation of the joints']",
"type": "background_claim",
"id": "T120"
},
{
"offsets": [
[
11380,
11385
]
],
"text": "['LCF00']",
"type": "data",
"id": "T121"
},
{
"offsets": [
[
11464,
11550
]
],
"text": "['we have a sufficient set of examples to describe the movement of an articulated object']",
"type": "data",
"id": "T122"
},
{
"offsets": [
[
11552,
11599
]
],
"text": "['we can interpolate displacement in “pose space”']",
"type": "background_claim",
"id": "T123"
},
{
"offsets": [
[
11601,
11606
]
],
"text": "['LCF00']",
"type": "data",
"id": "T124"
},
{
"offsets": [
[
11756,
11838
]
],
"text": "['we translate each skinning sample k to the rest coordinate frame using inverse SSD']",
"type": "data",
"id": "T125"
},
{
"offsets": [
[
11840,
11868
]
],
"text": "['the displacement between the']",
"type": "background_claim",
"id": "T126"
},
{
"offsets": [
[
11901,
11970
]
],
"text": "['sample vertex v k and the rest pose vertex can be calculated: n joint']",
"type": "background_claim",
"id": "T127"
},
{
"offsets": [
[
12002,
12038
]
],
"text": "['d k = ( ∑ w j T j ) −1 v k − v 0 j=1']",
"type": "background_claim",
"id": "T128"
},
{
"offsets": [
[
12304,
12412
]
],
"text": "['After defining the displacement of each pose, the displacement at an arbitrary pose can be calculated by RBF']",
"type": "background_claim",
"id": "T129"
},
{
"offsets": [
[
12439,
12444
]
],
"text": "['LCF00']",
"type": "data",
"id": "T130"
},
{
"offsets": [
[
12446,
12481
]
],
"text": "['or normalized radial basis function']",
"type": "background_claim",
"id": "T131"
},
{
"offsets": [
[
12484,
12488
]
],
"text": "['KM04']",
"type": "data",
"id": "T132"
},
{
"offsets": [
[
12490,
12539
]
],
"text": "['interpolation of the example poses’ displacements']",
"type": "background_claim",
"id": "T133"
},
{
"offsets": [
[
12541,
12626
]
],
"text": "['The displacement d a of a vertex in an arbitrary pose p a can be calculated as n pose']",
"type": "background_claim",
"id": "T134"
},
{
"offsets": [
[
12658,
12693
]
],
"text": "['d a = D(p a ) = ∑ r k (p a )d k k=1']",
"type": "background_claim",
"id": "T135"
},
{
"offsets": [
[
12890,
12973
]
],
"text": "['Normalized Radial Basis Functions can smoothly interpolate pose space using: n pose']",
"type": "background_claim",
"id": "T136"
},
{
"offsets": [
[
13005,
13040
]
],
"text": "['f t (p a ) = ∑ λ t,k φ k (γ k ) k=1']",
"type": "background_claim",
"id": "T137"
},
{
"offsets": [
[
13436,
13590
]
],
"text": "['The weight r k (p a ) is calculated using normalized RBFs and is used in equation 4 to calculate the displacement d a of a vertex in an arbitrary pose p a']",
"type": "background_claim",
"id": "T138"
},
{
"offsets": [
[
13748,
13787
]
],
"text": "['r k (p a ) = ∑ n pose f k (p f a (p ) )']",
"type": "background_claim",
"id": "T139"
},
{
"offsets": [
[
14053,
14089
]
],
"text": "['WPSD is developed by Kurihara et al.']",
"type": "background_claim",
"id": "T140"
},
{
"offsets": [
[
14092,
14096
]
],
"text": "['KM04']",
"type": "data",
"id": "T141"
},
{
"offsets": [
[
14098,
14172
]
],
"text": "['to deform their example-based human hand model derived from medical images']",
"type": "background_claim",
"id": "T142"
},
{
"offsets": [
[
14421,
14500
]
],
"text": "['the γ k is the difference of n joint dimensional joint vectors of related poses']",
"type": "data",
"id": "T143"
},
{
"offsets": [
[
14502,
14624
]
],
"text": "['every vertex in the pose p k has same distance γ k resulting in the same weight r k (p a ) in every vertex of the pose p k']",
"type": "background_claim",
"id": "T144"
},
{
"offsets": [
[
14174,
14305
]
],
"text": "['In the original PSD, the distance between two poses p a and p k having n joint number of joints for each pose is defined as n joint']",
"type": "background_claim",
"id": "T145"
},
{
"offsets": [
[
14337,
14382
]
],
"text": "['γ k (p a , p k ) = ∑ (p a, j − p k, j ) 2 j=1']",
"type": "background_claim",
"id": "T146"
},
{
"offsets": [
[
14648,
14728
]
],
"text": "['each element of the joint vector equally contributes to the distance calculation']",
"type": "background_claim",
"id": "T147"
},
{
"offsets": [
[
14730,
14809
]
],
"text": "['two vectors having a same value but different order generate same pose distance']",
"type": "background_claim",
"id": "T148"
},
{
"offsets": [
[
14824,
14978
]
],
"text": "['three different joint vectors p 1 = (θ, 0, 0), p 2 = (0, θ, 0), p 3 = (0, 0, θ) have same distance between them and it can cause unexpected results in PSD']",
"type": "background_claim",
"id": "T149"
},
{
"offsets": [
[
14980,
14987
]
],
"text": "['In WPSD']",
"type": "background_claim",
"id": "T150"
},
{
"offsets": [
[
14989,
14993
]
],
"text": "['KM04']",
"type": "data",
"id": "T151"
},
{
"offsets": [
[
14996,
15148
]
],
"text": "['Kurihara et al. modify the distance definition between poses using joint weight of each vertex i to give proper weight to each element of a joint vector']",
"type": "background_claim",
"id": "T152"
},
{
"offsets": [
[
15181,
15243
]
],
"text": "['n joint γ i,k (p a , p k ) = ∑ w i, j (p a, j − p k, j ) 2 j=1']",
"type": "background_claim",
"id": "T153"
},
{
"offsets": [
[
15404,
15467
]
],
"text": "['From this definition, a more accurate pose distance is obtained']",
"type": "background_claim",
"id": "T154"
},
{
"offsets": [
[
15472,
15519
]
],
"text": "['it generates better skinning in arbitrary poses']",
"type": "background_claim",
"id": "T155"
},
{
"offsets": [
[
15829,
15912
]
],
"text": "['The joint weights of each vertex are important to generate accurate skinning in SSD']",
"type": "background_claim",
"id": "T156"
},
{
"offsets": [
[
15914,
15924
]
],
"text": "['equation 2']",
"type": "data",
"id": "T157"
},
{
"offsets": [
[
15926,
15944
]
],
"text": "['as well as in WPSD']",
"type": "background_claim",
"id": "T158"
},
{
"offsets": [
[
15946,
15956
]
],
"text": "['equation 8']",
"type": "data",
"id": "T159"
},
{
"offsets": [
[
15959,
16034
]
],
"text": "['In many applications, the weights are manually generated by skilled artists']",
"type": "background_claim",
"id": "T160"
},
{
"offsets": [
[
16039,
16077
]
],
"text": "['it is hard to generate accurate values']",
"type": "background_claim",
"id": "T161"
},
{
"offsets": [
[
16083,
16136
]
],
"text": "['a number of joints are involved in deforming a region']",
"type": "data",
"id": "T162"
},
{
"offsets": [
[
16153,
16278
]
],
"text": "['we automatically calculate the joint weights of each vertex from the sample poses to enhance the accuracy of the weight value']",
"type": "own_claim",
"id": "T163"
},
{
"offsets": [
[
16280,
16380
]
],
"text": "['This results in better skinning and reduces the elaborate manual work required to create weight maps']",
"type": "own_claim",
"id": "T164"
},
{
"offsets": [
[
16382,
16456
]
],
"text": "['In each sample pose p k , we have following equation based on SSD: n joint']",
"type": "own_claim",
"id": "T165"
},
{
"offsets": [
[
16488,
16523
]
],
"text": "['v ̃ k − e k = ( ∑ w j T j )v 0 j=1']",
"type": "own_claim",
"id": "T166"
},
{
"offsets": [
[
16786,
16854
]
],
"text": "['we have sufficient examples involving the same set of n joint joints']",
"type": "data",
"id": "T167"
},
{
"offsets": [
[
16856,
16901
]
],
"text": "['we have n pose equations of the form: n joint']",
"type": "own_claim",
"id": "T168"
},
{
"offsets": [
[
16934,
16966
]
],
"text": "['v ̃ k − e k = ( ∑ v j w j ) j=1']",
"type": "own_claim",
"id": "T169"
},
{
"offsets": [
[
17031,
17049
]
],
"text": "['the e k is unknown']",
"type": "own_claim",
"id": "T170"
},
{
"offsets": [
[
17051,
17198
]
],
"text": "['we can solve for weights that minimize the e k in a least squares sense by stacking the equations 10 (with e k omitted) into a linear matrix system']",
"type": "own_claim",
"id": "T171"
},
{
"offsets": [
[
17231,
17239
]
],
"text": "['2 v − Aw']",
"type": "own_claim",
"id": "T172"
},
{
"offsets": [
[
17429,
17534
]
],
"text": "['From equation 11, we can calculate w from the given value of v and A to reduce the error of this equation']",
"type": "own_claim",
"id": "T173"
},
{
"offsets": [
[
17536,
17608
]
],
"text": "['We use the non-negative least square (NNLS) method to solve this problem']",
"type": "own_claim",
"id": "T174"
},
{
"offsets": [
[
17613,
17681
]
],
"text": "['it determines positive weight values minimizing error in equation 10']",
"type": "own_claim",
"id": "T175"
},
{
"offsets": [
[
17683,
17762
]
],
"text": "['The calculated weight vector w is normalized to satisfy ∑ n j=1 joint w j = 1.0']",
"type": "own_claim",
"id": "T176"
},
{
"offsets": [
[
17764,
17878
]
],
"text": "['In order to avoid a singular matrix A, the number of poses should be greater or equal to the number of overall DOF']",
"type": "own_claim",
"id": "T177"
},
{
"offsets": [
[
17899,
17918
]
],
"text": "['of the joint vector']",
"type": "own_claim",
"id": "T178"
},
{
"offsets": [
[
17947,
17996
]
],
"text": "['the sample poses should be sufficiently different']",
"type": "own_claim",
"id": "T179"
},
{
"offsets": [
[
17998,
18075
]
],
"text": "['James et al. used a similar approach to estimate vertex weights in each joint']",
"type": "background_claim",
"id": "T180"
},
{
"offsets": [
[
18077,
18081
]
],
"text": "['JT05']",
"type": "data",
"id": "T181"
},
{
"offsets": [
[
18210,
18252
]
],
"text": "['Skinning deformations vary across vertices']",
"type": "background_claim",
"id": "T182"
},
{
"offsets": [
[
18254,
18385
]
],
"text": "['In SSD and WPSD, this per-vertex computation is independent for each vertex and can be parallelized by a SIMD parallel architecture']",
"type": "background_claim",
"id": "T183"
},
{
"offsets": [
[
18387,
18544
]
],
"text": "['We developed a parallel skinning algorithm for SSD and WPSD that is suitable to GPUs having a SIMD architecture with one-side communication to texture memory']",
"type": "own_claim",
"id": "T184"
},
{
"offsets": [
[
18728,
18979
]
],
"text": "['The computation cost of the SSD skinning algorithm is O(n vertex × n joint ) from equations 1, 2, PSD is O(n vertex × n joint × n pose ) from equations 1, 2, 4, and WPSD is O(n vertex × n joint × n pose × n pose × n pose ) from equations 1, 2, 4, 5, 6']",
"type": "own_claim",
"id": "T185"
},
{
"offsets": [
[
19059,
19116
]
],
"text": "['r i is same in all vertices and d i can be pre-calculated']",
"type": "own_claim",
"id": "T186"
},
{
"offsets": [
[
18988,
19051
]
],
"text": "['computation cost of original PSD is defined by equation 1, 2, 4']",
"type": "own_claim",
"id": "T187"
},
{
"offsets": [
[
19118,
19242
]
],
"text": "['The number of joints n joint and poses n pose can be reduced to the smaller numbers using the method developed by Kry et al.']",
"type": "own_claim",
"id": "T188"
},
{
"offsets": [
[
19245,
19250
]
],
"text": "['KJP02']",
"type": "data",
"id": "T189"
},
{
"offsets": [
[
19277,
19290
]
],
"text": "['section 5.2.1']",
"type": "data",
"id": "T190"
},
{
"offsets": [
[
19358,
19434
]
],
"text": "['the Eigenskin method based on PSD was developed using GPU vertex programming']",
"type": "background_claim",
"id": "T191"
},
{
"offsets": [
[
19436,
19441
]
],
"text": "['KJP02']",
"type": "data",
"id": "T192"
},
{
"offsets": [
[
19444,
19556
]
],
"text": "['The vertex program uses a relatively small number of slow processing units compared with the fragment processors']",
"type": "background_claim",
"id": "T193"
},
{
"offsets": [
[
19562,
19637
]
],
"text": "['the per-vertex computation cost of the original PSD is O(n joint × n pose )']",
"type": "background_claim",
"id": "T194"
},
{
"offsets": [
[
19649,
19653
]
],
"text": "['WPSD']",
"type": "own_claim",
"id": "T195"
},
{
"offsets": [
[
19736,
19804
]
],
"text": "['can clearly benefit from parallel computation on fragment processors']",
"type": "own_claim",
"id": "T196"
},
{
"offsets": [
[
19655,
19734
]
],
"text": "['having higher pervertex computation cost O(n joint × n pose × n pose × n pose )']",
"type": "own_claim",
"id": "T197"
},
{
"offsets": [
[
19993,
20139
]
],
"text": "['In order to minimize real-time computation, we separate possible pre-calculation steps and save the results into texture memory using texture maps']",
"type": "own_claim",
"id": "T198"
},
{
"offsets": [
[
20149,
20225
]
],
"text": "['the value in the texture memory is not changed in the successive deformation']",
"type": "own_claim",
"id": "T199"
},
{
"offsets": [
[
20227,
20292
]
],
"text": "['it can be pre-computed and stored in the read-only texture memory']",
"type": "own_claim",
"id": "T200"
},
{
"offsets": [
[
20830,
20886
]
],
"text": "['The fragment processors cannot access vertex information']",
"type": "own_claim",
"id": "T201"
},
{
"offsets": [
[
20897,
20959
]
],
"text": "['we can use texture memory to send data to the fragment program']",
"type": "own_claim",
"id": "T202"
},
{
"offsets": [
[
21215,
21267
]
],
"text": "['These textures represent parameter v 0 in equation 2']",
"type": "own_claim",
"id": "T203"
},
{
"offsets": [
[
21272,
21343
]
],
"text": "['each 3D element (x, y, z) is stored into the (r, g, b) value of a texel']",
"type": "own_claim",
"id": "T204"
},
{
"offsets": [
[
21346,
21354
]
],
"text": "['Figure 2']",
"type": "data",
"id": "T205"
},
{
"offsets": [
[
21797,
21883
]
],
"text": "['the distribution of skinning effects in an articulated body is local to several joints']",
"type": "background_claim",
"id": "T206"
},
{
"offsets": [
[
21886,
21891
]
],
"text": "['MMT97']",
"type": "data",
"id": "T207"
},
{
"offsets": [
[
21892,
21897
]
],
"text": "['KJP02']",
"type": "data",
"id": "T208"
},
{
"offsets": [
[
21956,
21989
]
],
"text": "['deformations arising from the PIP']",
"type": "background_claim",
"id": "T209"
},
{
"offsets": [
[
22017,
22076
]
],
"text": "['joint of index finger do not propagate to the other fingers']",
"type": "background_claim",
"id": "T210"
},
{
"offsets": [
[
22082,
22179
]
],
"text": "['deformation on the middle phalanx of index finger is only affected by the movement of PIP and DIP']",
"type": "background_claim",
"id": "T211"
},
{
"offsets": [
[
22196,
22202
]
],
"text": "['joints']",
"type": "background_claim",
"id": "T212"
},
{
"offsets": [
[
22227,
22394
]
],
"text": "['we can reduce joint weight storage from the actual number of joint n joint to a smaller number of “principal joints” n ̃ joint selected by sorting on the weight value']",
"type": "own_claim",
"id": "T213"
},
{
"offsets": [
[
22520,
22685
]
],
"text": "['we can save the joint weights of entire geometry in two RGBA textures T w1 , T w2 each with size n vertex × 4(rgba) and store the entire information required for SSD']",
"type": "own_claim",
"id": "T214"
},
{
"offsets": [
[
22699,
22743
]
],
"text": "['in four textures T v , T n , T w1 , and T w2']",
"type": "own_claim",
"id": "T215"
},
{
"offsets": [
[
22396,
22505
]
],
"text": "['We threshold n ̃ joint at four in our tests with an additional four elements to hold the related joint index']",
"type": "own_claim",
"id": "T216"
},
{
"offsets": [
[
22687,
22697
]
],
"text": "['equation 2']",
"type": "data",
"id": "T217"
},
{
"offsets": [
[
22746,
22840
]
],
"text": "['The displacement values calculated by equation 3 can be stored in n pose displacement textures']",
"type": "own_claim",
"id": "T218"
},
{
"offsets": [
[
22939,
23020
]
],
"text": "['we need many sample poses to calculate accurate joint weights and PSD deformation']",
"type": "own_claim",
"id": "T219"
},
{
"offsets": [
[
22891,
22915
]
],
"text": "['complex joint structures']",
"type": "data",
"id": "T220"
},
{
"offsets": [
[
22920,
22937
]
],
"text": "['a large DOF model']",
"type": "data",
"id": "T221"
},
{
"offsets": [
[
23037,
23076
]
],
"text": "['the joint weights can be pre-calculated']",
"type": "own_claim",
"id": "T222"
},
{
"offsets": [
[
23078,
23154
]
],
"text": "['we can reduce the number of sample poses needed in real-time PSD computation']",
"type": "own_claim",
"id": "T223"
},
{
"offsets": [
[
23191,
23263
]
],
"text": "['of pose space can yield an orthogonal basis called “ Eigendisplacement ”']",
"type": "background_claim",
"id": "T224"
},
{
"offsets": [
[
23156,
23159
]
],
"text": "['PCA']",
"type": "background_claim",
"id": "T225"
},
{
"offsets": [
[
23265,
23270
]
],
"text": "['KJP02']",
"type": "data",
"id": "T226"
},
{
"offsets": [
[
23276,
23351
]
],
"text": "['we reduce the size of pose space from n pose to n ̃ pose “principal poses”']",
"type": "data",
"id": "T227"
},
{
"offsets": [
[
23379,
23428
]
],
"text": "['we can reduce the number of displacement textures']",
"type": "own_claim",
"id": "T228"
},
{
"offsets": [
[
23444,
23593
]
],
"text": "['we set n ̃ pose as eight in our experiment and save displacements of all poses into a RGB texture T d having size n vertex × 8( n ̃ pose ) × 3(rgb)']",
"type": "own_claim",
"id": "T229"
},
{
"offsets": [
[
23606,
23841
]
],
"text": "['from the two important observations of “principal joints” and “principal poses”, the original computation cost for SSD, PSD, and WPSD discussed in section 5.1 can be reduced using n ̃ joint and n ̃ pose rather than n joint and n pose']",
"type": "own_claim",
"id": "T230"
},
{
"offsets": [
[
23871,
23927
]
],
"text": "['the weight r i in equation 4 is the same at every vertex']",
"type": "background_claim",
"id": "T231"
},
{
"offsets": [
[
23844,
23863
]
],
"text": "['In the original PSD']",
"type": "background_claim",
"id": "T232"
},
{
"offsets": [
[
23929,
23978
]
],
"text": "['we do not need to calculate this value in the GPU']",
"type": "background_claim",
"id": "T233"
},
{
"offsets": [
[
23986,
24026
]
],
"text": "['the size of this value is just n ̃ pose']",
"type": "own_claim",
"id": "T234"
},
{
"offsets": [
[
24029,
24110
]
],
"text": "['we can simply pass them to the GPU as parameters without generating a texture map']",
"type": "own_claim",
"id": "T235"
},
{
"offsets": [
[
24123,
24202
]
],
"text": "['we store all the information needed to calculate the original PSD at this point']",
"type": "own_claim",
"id": "T236"
},
{
"offsets": [
[
24646,
24697
]
],
"text": "['they are not changed during the deformation process']",
"type": "own_claim",
"id": "T237"
},
{
"offsets": [
[
24561,
24638
]
],
"text": "['The values in the texture maps are stored in the texture memory at setup time']",
"type": "own_claim",
"id": "T238"
},
{
"offsets": [
[
24204,
24415
]
],
"text": "['In order to reduce real-time computation, we pre-calculate T j in equation 2 and λ in equation 5 and store them into another one channel texture T x having size n ̃ pose × ( n ̃ pose + n ̃ joint × 3(x, y, z))']",
"type": "own_claim",
"id": "T239"
},
{
"offsets": [
[
24430,
24558
]
],
"text": "['we store all the variables required to calculate WPSD, PSD, and SSD in six texture maps: T v , T n , T w1 , T w2 , T d , and T x']",
"type": "own_claim",
"id": "T240"
},
{
"offsets": [
[
24699,
24818
]
],
"text": "['In current graphic card architectures, data transfer from CPU to GPU is slow compared with memory access within the GPU']",
"type": "background_claim",
"id": "T241"
},
{
"offsets": [
[
24826,
24974
]
],
"text": "['the only data changed in each deformation and passed from CPU to GPU is a joint vector p a (size = n joint ) representing the current arbitrary pose']",
"type": "own_claim",
"id": "T242"
},
{
"offsets": [
[
24976,
25030
]
],
"text": "['the memory access rate in our method is very efficient']",
"type": "own_claim",
"id": "T243"
},
{
"offsets": [
[
25032,
25115
]
],
"text": "['In the original PSD method, an additional r k value (size = n ̃ pose ) is required']",
"type": "background_claim",
"id": "T244"
},
{
"offsets": [
[
26120,
26181
]
],
"text": "['vertex information cannot be accessed by the fragment program']",
"type": "background_claim",
"id": "T245"
},
{
"offsets": [
[
26183,
26251
]
],
"text": "['vertex deformation on a GPU is usually performed by a vertex program']",
"type": "background_claim",
"id": "T246"
},
{
"offsets": [
[
26254,
26259
]
],
"text": "['KJP02']",
"type": "data",
"id": "T247"
},
{
"offsets": [
[
26261,
26265
]
],
"text": "['BK05']",
"type": "data",
"id": "T248"
},
{
"offsets": [
[
26278,
26330
]
],
"text": "['we cannot access vertex data in the fragment program']",
"type": "own_claim",
"id": "T249"
},
{
"offsets": [
[
26332,
26402
]
],
"text": "['the efficiency of parallel computation on a fragment program is higher']",
"type": "own_claim",
"id": "T250"
},
{
"offsets": [
[
26410,
26526
]
],
"text": "['the fragment processor has more processing units and each of them has more computation power than a vertex processor']",
"type": "own_claim",
"id": "T251"
},
{
"offsets": [
[
26744,
26859
]
],
"text": "['We developed a parallel WPSD algorithms using the fragment processors to enhance the extent of parallel computation']",
"type": "own_claim",
"id": "T252"
},
{
"offsets": [
[
27318,
27376
]
],
"text": "['the viewport is set to the same resolution as the textures']",
"type": "own_claim",
"id": "T253"
},
{
"offsets": [
[
27378,
27457
]
],
"text": "['each fragment is exactly matched with each texel holding the vertex information']",
"type": "own_claim",
"id": "T254"
},
{
"offsets": [
[
27463,
27534
]
],
"text": "['we can access each vertex using the texture coordinates of the fragment']",
"type": "own_claim",
"id": "T255"
},
{
"offsets": [
[
27536,
27635
]
],
"text": "['vertex weights and displacements stored in the texture maps can also be accessed by similar methods']",
"type": "own_claim",
"id": "T256"
},
{
"offsets": [
[
27637,
27665
]
],
"text": "['A similar idea was developed']",
"type": "background_claim",
"id": "T257"
},
{
"offsets": [
[
27678,
27745
]
],
"text": "['to calculate ray tracing in a fragment program and is used in GPGPU']",
"type": "background_claim",
"id": "T258"
},
{
"offsets": [
[
27784,
27796
]
],
"text": "['applications']",
"type": "background_claim",
"id": "T259"
},
{
"offsets": [
[
27670,
27676
]
],
"text": "['PBMH02']",
"type": "data",
"id": "T260"
},
{
"offsets": [
[
27798,
27801
]
],
"text": "['GPG']",
"type": "data",
"id": "T261"
},
{
"offsets": [
[
27803,
27811
]
],
"text": "['LHK ∗ 04']",
"type": "data",
"id": "T262"
},
{
"offsets": [
[
27813,
27817
]
],
"text": "['PF05']",
"type": "data",
"id": "T263"
},
{
"offsets": [
[
27820,
27827
]
],
"text": "['The FBO']",
"type": "background_claim",
"id": "T264"
},
{
"offsets": [
[
27850,
27859
]
],
"text": "['extension']",
"type": "background_claim",
"id": "T265"
},
{
"offsets": [
[
27870,
27913
]
],
"text": "['supports rendering into an attached texture']",
"type": "background_claim",
"id": "T266"
},
{
"offsets": [
[
27862,
27867
]
],
"text": "['Gre05']",
"type": "data",
"id": "T267"
},
{
"offsets": [
[
27915,
27941
]
],
"text": "['This saves memory and time']",
"type": "background_claim",
"id": "T268"
},
{
"offsets": [
[
27949,
28011
]
],
"text": "['there is no copy operation from frame buffer to texture buffer']",
"type": "background_claim",
"id": "T269"
},
{
"offsets": [
[
28013,
28187
]
],
"text": "['We implemented our WPSD algorithm using the fragment program with the FBO extension to store the result directly into texture maps accessed by vertex program in the next pass']",
"type": "own_claim",
"id": "T270"
},
{
"offsets": [
[
28226,
28237
]
],
"text": "['Algorithm 1']",
"type": "data",
"id": "T271"
},
{
"offsets": [
[
30411,
30434
]
],
"text": "['multiple render targets']",
"type": "own_claim",
"id": "T272"
},
{
"offsets": [
[
30442,
30479
]
],
"text": "['can combine the first and second pass']",
"type": "own_claim",
"id": "T273"
},
{
"offsets": [
[
30485,
30506
]
],
"text": "['vertex buffer objects']",
"type": "own_claim",
"id": "T274"
},
{
"offsets": [
[
30514,
30583
]
],
"text": "['could be used to render the deformed results back to the vertex array']",
"type": "own_claim",
"id": "T275"
},
{
"offsets": [
[
30585,
30588
]
],
"text": "['OPE']",
"type": "data",
"id": "T276"
},
{
"offsets": [
[
30590,
30593
]
],
"text": "['GPG']",
"type": "data",
"id": "T277"
},
{
"offsets": [
[
30595,
30603
]
],
"text": "['LHK ∗ 04']",
"type": "data",
"id": "T278"
},
{
"offsets": [
[
30656,
30726
]
],
"text": "['We tested our methods using upper arm models consisting of four joints']",
"type": "own_claim",
"id": "T279"
},
{
"offsets": [
[
30765,
30783
]
],
"text": "['Each has three DOF']",
"type": "own_claim",
"id": "T280"
},
{
"offsets": [
[
30788,
30828
]
],
"text": "['the wrist is the end joint having no DOF']",
"type": "own_claim",
"id": "T281"
},
{
"offsets": [
[
30830,
30924
]
],
"text": "['Three different resolution meshes are used to test the performance of GPU parallel computation']",
"type": "own_claim",
"id": "T282"
},
{
"offsets": [
[
30926,
30991
]
],
"text": "['the high-resolution model has 91460 triangles with 46036 vertices']",
"type": "own_claim",
"id": "T283"
},
{
"offsets": [
[
30993,
31056
]
],
"text": "['the midresolution model has 22836 triangles with 11574 vertices']",
"type": "own_claim",
"id": "T284"
},
{
"offsets": [
[
31062,
31124
]
],
"text": "['the low-resolution model has 5762 triangles with 2972 vertices']",
"type": "own_claim",
"id": "T285"
},
{
"offsets": [
[
31127,
31135
]
],
"text": "['Figure 4']",
"type": "data",
"id": "T286"
},
{
"offsets": [
[
31149,
31225
]
],
"text": "['these models are considerably more detailed than those used in current games']",
"type": "own_claim",
"id": "T287"
},
{
"offsets": [
[
31230,
31275
]
],
"text": "['the reported frame rates would be much higher']",
"type": "own_claim",
"id": "T288"
},
{
"offsets": [
[
31279,
31318
]
],
"text": "['typical gameresolution models were used']",
"type": "data",
"id": "T289"
},
{
"offsets": [
[
31389,
31444
]
],
"text": "['models such as these will be in wide use in a few years']",
"type": "own_claim",
"id": "T290"
},
{
"offsets": [
[
31344,
31387
]
],
"text": "['the expected growth of GPU processing power']",
"type": "own_claim",
"id": "T291"
},
{
"offsets": [
[
31450,
31552
]
],
"text": "['algorithms such as WPSD will be required to produce realistic deformations at this level of resolution']",
"type": "own_claim",
"id": "T292"
},
{
"offsets": [
[
31554,
31687
]
],
"text": "['Eight sample poses were created by Poser [Cur] and the joints weights and displacements of each sample were derived from these models']",
"type": "own_claim",
"id": "T293"
},
{
"offsets": [
[
31690,
31698
]
],
"text": "['Figure 5']",
"type": "data",
"id": "T294"
},
{
"offsets": [
[
31702,
31763
]
],
"text": "['Our parallel algorithm is based on three pass GPU computation']",
"type": "own_claim",
"id": "T295"
},
{
"offsets": [
[
31765,
31886
]
],
"text": "['The fragment program for the 1st and 2nd pass, and the vertex program for the 3rd pass are implemented in the Cg language']",
"type": "own_claim",
"id": "T296"
},
{
"offsets": [
[
31889,
31893
]
],
"text": "['FK03']",
"type": "data",
"id": "T297"
},
{
"offsets": [
[
32120,
32205
]
],
"text": "['the maximum required memory space for the highest resolution model is just 6.8 Mbytes']",
"type": "own_claim",
"id": "T298"
},
{
"offsets": [
[
31896,
32015
]
],
"text": "['For accuracy the GPU computation is performed by 32bit floating point operations with 32bit floating point texture maps']",
"type": "own_claim",
"id": "T299"
},
{
"offsets": [
[
32207,
32292
]
],
"text": "['the size of the output texture T v ′ and T n ′ is the same as the size of T v and T n']",
"type": "own_claim",
"id": "T300"
},
{
"offsets": [
[
32556,
32660
]
],
"text": "['On average, our GPU-based deformation shows around 20 times speed-up compared with CPU-based deformation']",
"type": "own_claim",
"id": "T301"
},
{
"offsets": [
[
32546,
32553
]
],
"text": "['table 1']",
"type": "data",
"id": "T302"
},
{
"offsets": [
[
32662,
32719
]
],
"text": "['GPU-based WPSD has roughly the same speed as CPUbased SSD']",
"type": "own_claim",
"id": "T303"
},
{
"offsets": [
[
32732,
32847
]
],
"text": "['real-time applications using SSD can substitute WPSD running on the GPU without loosing their real-time performance']",
"type": "own_claim",
"id": "T304"
},
{
"offsets": [
[
32855,
32915
]
],
"text": "['our algorithm shows speed-up for SSD and PSD as well as WPSD']",
"type": "own_claim",
"id": "T305"
},
{
"offsets": [
[
32917,
33026
]
],
"text": "['applications can choose the most appropriate skinning method according to the required deformation and detail']",
"type": "own_claim",
"id": "T306"
},
{
"offsets": [
[
36388,
36473
]
],
"text": "['we present a parallel skinning algorithm suitable for SIMD architectures such as GPUs']",
"type": "own_claim",
"id": "T307"
},
{
"offsets": [
[
37012,
37115
]
],
"text": "['Articulated body skinning using SSD, PSD, and WPSD are efficiently parallelized by our GPU-based method']",
"type": "own_claim",
"id": "T308"
},
{
"offsets": [
[
37121,
37212
]
],
"text": "['on a detailed model, we obtain around 20 times speed-up compared with CPU-based computation']",
"type": "own_claim",
"id": "T309"
},
{
"offsets": [
[
37214,
37342
]
],
"text": "['Principal component compression of the examples and careful analysis of joint distributions can reduce the domain of computation']",
"type": "background_claim",
"id": "T310"
},
{
"offsets": [
[
37345,
37350
]
],
"text": "['KJP02']",
"type": "data",
"id": "T311"
},
{
"offsets": [
[
37356,
37465
]
],
"text": "['other algorithms based on the SSD, PSD, and shape interpolation may be parallelized on GPU using our approach']",
"type": "own_claim",
"id": "T312"
}
] | [
{
"id": "R1",
"head": {
"ref_id": "T3",
"role": "Arg1"
},
"tail": {
"ref_id": "T4",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R2",
"head": {
"ref_id": "T3",
"role": "Arg1"
},
"tail": {
"ref_id": "T5",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R3",
"head": {
"ref_id": "T9",
"role": "Arg1"
},
"tail": {
"ref_id": "T8",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R4",
"head": {
"ref_id": "T8",
"role": "Arg1"
},
"tail": {
"ref_id": "T7",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R5",
"head": {
"ref_id": "T13",
"role": "Arg1"
},
"tail": {
"ref_id": "T12",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R6",
"head": {
"ref_id": "T12",
"role": "Arg1"
},
"tail": {
"ref_id": "T11",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R7",
"head": {
"ref_id": "T15",
"role": "Arg1"
},
"tail": {
"ref_id": "T14",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R8",
"head": {
"ref_id": "T16",
"role": "Arg1"
},
"tail": {
"ref_id": "T15",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R9",
"head": {
"ref_id": "T18",
"role": "Arg1"
},
"tail": {
"ref_id": "T17",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R10",
"head": {
"ref_id": "T19",
"role": "Arg1"
},
"tail": {
"ref_id": "T18",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R11",
"head": {
"ref_id": "T20",
"role": "Arg1"
},
"tail": {
"ref_id": "T18",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R12",
"head": {
"ref_id": "T18",
"role": "Arg1"
},
"tail": {
"ref_id": "T21",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R13",
"head": {
"ref_id": "T26",
"role": "Arg1"
},
"tail": {
"ref_id": "T25",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R14",
"head": {
"ref_id": "T24",
"role": "Arg1"
},
"tail": {
"ref_id": "T25",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R15",
"head": {
"ref_id": "T28",
"role": "Arg1"
},
"tail": {
"ref_id": "T27",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R16",
"head": {
"ref_id": "T31",
"role": "Arg1"
},
"tail": {
"ref_id": "T30",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R17",
"head": {
"ref_id": "T35",
"role": "Arg1"
},
"tail": {
"ref_id": "T34",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R18",
"head": {
"ref_id": "T36",
"role": "Arg1"
},
"tail": {
"ref_id": "T34",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R19",
"head": {
"ref_id": "T36",
"role": "Arg1"
},
"tail": {
"ref_id": "T37",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R20",
"head": {
"ref_id": "T38",
"role": "Arg1"
},
"tail": {
"ref_id": "T37",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R21",
"head": {
"ref_id": "T39",
"role": "Arg1"
},
"tail": {
"ref_id": "T37",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R22",
"head": {
"ref_id": "T40",
"role": "Arg1"
},
"tail": {
"ref_id": "T37",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R23",
"head": {
"ref_id": "T43",
"role": "Arg1"
},
"tail": {
"ref_id": "T42",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R24",
"head": {
"ref_id": "T46",
"role": "Arg1"
},
"tail": {
"ref_id": "T45",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R25",
"head": {
"ref_id": "T47",
"role": "Arg1"
},
"tail": {
"ref_id": "T45",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R26",
"head": {
"ref_id": "T48",
"role": "Arg1"
},
"tail": {
"ref_id": "T47",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R27",
"head": {
"ref_id": "T50",
"role": "Arg1"
},
"tail": {
"ref_id": "T49",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R28",
"head": {
"ref_id": "T51",
"role": "Arg1"
},
"tail": {
"ref_id": "T49",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R29",
"head": {
"ref_id": "T52",
"role": "Arg1"
},
"tail": {
"ref_id": "T49",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R30",
"head": {
"ref_id": "T53",
"role": "Arg1"
},
"tail": {
"ref_id": "T49",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R31",
"head": {
"ref_id": "T56",
"role": "Arg1"
},
"tail": {
"ref_id": "T54",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R32",
"head": {
"ref_id": "T57",
"role": "Arg1"
},
"tail": {
"ref_id": "T54",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R33",
"head": {
"ref_id": "T58",
"role": "Arg1"
},
"tail": {
"ref_id": "T54",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R34",
"head": {
"ref_id": "T59",
"role": "Arg1"
},
"tail": {
"ref_id": "T54",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R35",
"head": {
"ref_id": "T60",
"role": "Arg1"
},
"tail": {
"ref_id": "T54",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R36",
"head": {
"ref_id": "T55",
"role": "Arg1"
},
"tail": {
"ref_id": "T54",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R37",
"head": {
"ref_id": "T61",
"role": "Arg1"
},
"tail": {
"ref_id": "T55",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R38",
"head": {
"ref_id": "T62",
"role": "Arg1"
},
"tail": {
"ref_id": "T55",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R39",
"head": {
"ref_id": "T63",
"role": "Arg1"
},
"tail": {
"ref_id": "T55",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R40",
"head": {
"ref_id": "T67",
"role": "Arg1"
},
"tail": {
"ref_id": "T66",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R41",
"head": {
"ref_id": "T70",
"role": "Arg1"
},
"tail": {
"ref_id": "T69",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R42",
"head": {
"ref_id": "T68",
"role": "Arg1"
},
"tail": {
"ref_id": "T69",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R43",
"head": {
"ref_id": "T67",
"role": "Arg1"
},
"tail": {
"ref_id": "T69",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R44",
"head": {
"ref_id": "T67",
"role": "Arg1"
},
"tail": {
"ref_id": "T71",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R45",
"head": {
"ref_id": "T73",
"role": "Arg1"
},
"tail": {
"ref_id": "T72",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R46",
"head": {
"ref_id": "T75",
"role": "Arg1"
},
"tail": {
"ref_id": "T74",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R47",
"head": {
"ref_id": "T76",
"role": "Arg1"
},
"tail": {
"ref_id": "T74",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R48",
"head": {
"ref_id": "T78",
"role": "Arg1"
},
"tail": {
"ref_id": "T77",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R49",
"head": {
"ref_id": "T78",
"role": "Arg1"
},
"tail": {
"ref_id": "T79",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R50",
"head": {
"ref_id": "T82",
"role": "Arg1"
},
"tail": {
"ref_id": "T83",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R51",
"head": {
"ref_id": "T81",
"role": "Arg1"
},
"tail": {
"ref_id": "T83",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R52",
"head": {
"ref_id": "T84",
"role": "Arg1"
},
"tail": {
"ref_id": "T83",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R53",
"head": {
"ref_id": "T85",
"role": "Arg1"
},
"tail": {
"ref_id": "T83",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R54",
"head": {
"ref_id": "T86",
"role": "Arg1"
},
"tail": {
"ref_id": "T83",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R55",
"head": {
"ref_id": "T88",
"role": "Arg1"
},
"tail": {
"ref_id": "T87",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R56",
"head": {
"ref_id": "T89",
"role": "Arg1"
},
"tail": {
"ref_id": "T87",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R57",
"head": {
"ref_id": "T90",
"role": "Arg1"
},
"tail": {
"ref_id": "T87",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R58",
"head": {
"ref_id": "T91",
"role": "Arg1"
},
"tail": {
"ref_id": "T87",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R59",
"head": {
"ref_id": "T92",
"role": "Arg1"
},
"tail": {
"ref_id": "T93",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R60",
"head": {
"ref_id": "T93",
"role": "Arg1"
},
"tail": {
"ref_id": "T94",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R61",
"head": {
"ref_id": "T93",
"role": "Arg1"
},
"tail": {
"ref_id": "T87",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R62",
"head": {
"ref_id": "T98",
"role": "Arg1"
},
"tail": {
"ref_id": "T97",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R63",
"head": {
"ref_id": "T100",
"role": "Arg1"
},
"tail": {
"ref_id": "T99",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R64",
"head": {
"ref_id": "T102",
"role": "Arg1"
},
"tail": {
"ref_id": "T101",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R65",
"head": {
"ref_id": "T103",
"role": "Arg1"
},
"tail": {
"ref_id": "T101",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R66",
"head": {
"ref_id": "T104",
"role": "Arg1"
},
"tail": {
"ref_id": "T101",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R67",
"head": {
"ref_id": "T105",
"role": "Arg1"
},
"tail": {
"ref_id": "T101",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R68",
"head": {
"ref_id": "T107",
"role": "Arg1"
},
"tail": {
"ref_id": "T106",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R69",
"head": {
"ref_id": "T109",
"role": "Arg1"
},
"tail": {
"ref_id": "T108",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R70",
"head": {
"ref_id": "T111",
"role": "Arg1"
},
"tail": {
"ref_id": "T110",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R71",
"head": {
"ref_id": "T113",
"role": "Arg1"
},
"tail": {
"ref_id": "T112",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R72",
"head": {
"ref_id": "T114",
"role": "Arg1"
},
"tail": {
"ref_id": "T112",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R73",
"head": {
"ref_id": "T115",
"role": "Arg1"
},
"tail": {
"ref_id": "T114",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R74",
"head": {
"ref_id": "T119",
"role": "Arg1"
},
"tail": {
"ref_id": "T118",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R75",
"head": {
"ref_id": "T120",
"role": "Arg1"
},
"tail": {
"ref_id": "T119",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R76",
"head": {
"ref_id": "T121",
"role": "Arg1"
},
"tail": {
"ref_id": "T120",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R77",
"head": {
"ref_id": "T122",
"role": "Arg1"
},
"tail": {
"ref_id": "T123",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R78",
"head": {
"ref_id": "T124",
"role": "Arg1"
},
"tail": {
"ref_id": "T123",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R79",
"head": {
"ref_id": "T125",
"role": "Arg1"
},
"tail": {
"ref_id": "T126",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R80",
"head": {
"ref_id": "T127",
"role": "Arg1"
},
"tail": {
"ref_id": "T126",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R81",
"head": {
"ref_id": "T128",
"role": "Arg1"
},
"tail": {
"ref_id": "T127",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R82",
"head": {
"ref_id": "T130",
"role": "Arg1"
},
"tail": {
"ref_id": "T129",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R83",
"head": {
"ref_id": "T131",
"role": "Arg1"
},
"tail": {
"ref_id": "T129",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R84",
"head": {
"ref_id": "T132",
"role": "Arg1"
},
"tail": {
"ref_id": "T131",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R85",
"head": {
"ref_id": "T133",
"role": "Arg1"
},
"tail": {
"ref_id": "T131",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R86",
"head": {
"ref_id": "T135",
"role": "Arg1"
},
"tail": {
"ref_id": "T134",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R87",
"head": {
"ref_id": "T137",
"role": "Arg1"
},
"tail": {
"ref_id": "T136",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R88",
"head": {
"ref_id": "T139",
"role": "Arg1"
},
"tail": {
"ref_id": "T138",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R89",
"head": {
"ref_id": "T141",
"role": "Arg1"
},
"tail": {
"ref_id": "T140",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R90",
"head": {
"ref_id": "T142",
"role": "Arg1"
},
"tail": {
"ref_id": "T140",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R91",
"head": {
"ref_id": "T143",
"role": "Arg1"
},
"tail": {
"ref_id": "T144",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R92",
"head": {
"ref_id": "T146",
"role": "Arg1"
},
"tail": {
"ref_id": "T145",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R93",
"head": {
"ref_id": "T147",
"role": "Arg1"
},
"tail": {
"ref_id": "T148",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R94",
"head": {
"ref_id": "T149",
"role": "Arg1"
},
"tail": {
"ref_id": "T148",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R95",
"head": {
"ref_id": "T151",
"role": "Arg1"
},
"tail": {
"ref_id": "T150",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R96",
"head": {
"ref_id": "T153",
"role": "Arg1"
},
"tail": {
"ref_id": "T152",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R97",
"head": {
"ref_id": "T152",
"role": "Arg1"
},
"tail": {
"ref_id": "T150",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R98",
"head": {
"ref_id": "T157",
"role": "Arg1"
},
"tail": {
"ref_id": "T156",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R99",
"head": {
"ref_id": "T158",
"role": "Arg1"
},
"tail": {
"ref_id": "T156",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R100",
"head": {
"ref_id": "T159",
"role": "Arg1"
},
"tail": {
"ref_id": "T158",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R101",
"head": {
"ref_id": "T162",
"role": "Arg1"
},
"tail": {
"ref_id": "T161",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R102",
"head": {
"ref_id": "T166",
"role": "Arg1"
},
"tail": {
"ref_id": "T165",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R103",
"head": {
"ref_id": "T169",
"role": "Arg1"
},
"tail": {
"ref_id": "T168",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R104",
"head": {
"ref_id": "T167",
"role": "Arg1"
},
"tail": {
"ref_id": "T168",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R105",
"head": {
"ref_id": "T172",
"role": "Arg1"
},
"tail": {
"ref_id": "T171",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R106",
"head": {
"ref_id": "T171",
"role": "Arg1"
},
"tail": {
"ref_id": "T170",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R107",
"head": {
"ref_id": "T178",
"role": "Arg1"
},
"tail": {
"ref_id": "T177",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R108",
"head": {
"ref_id": "T181",
"role": "Arg1"
},
"tail": {
"ref_id": "T180",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R109",
"head": {
"ref_id": "T186",
"role": "Arg1"
},
"tail": {
"ref_id": "T187",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R110",
"head": {
"ref_id": "T189",
"role": "Arg1"
},
"tail": {
"ref_id": "T188",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R111",
"head": {
"ref_id": "T190",
"role": "Arg1"
},
"tail": {
"ref_id": "T188",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R112",
"head": {
"ref_id": "T192",
"role": "Arg1"
},
"tail": {
"ref_id": "T191",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R113",
"head": {
"ref_id": "T197",
"role": "Arg1"
},
"tail": {
"ref_id": "T195",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R114",
"head": {
"ref_id": "T196",
"role": "Arg1"
},
"tail": {
"ref_id": "T195",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R115",
"head": {
"ref_id": "T194",
"role": "Arg1"
},
"tail": {
"ref_id": "T195",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R116",
"head": {
"ref_id": "T199",
"role": "Arg1"
},
"tail": {
"ref_id": "T200",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R117",
"head": {
"ref_id": "T205",
"role": "Arg1"
},
"tail": {
"ref_id": "T204",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R118",
"head": {
"ref_id": "T207",
"role": "Arg1"
},
"tail": {
"ref_id": "T206",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R119",
"head": {
"ref_id": "T208",
"role": "Arg1"
},
"tail": {
"ref_id": "T206",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R120",
"head": {
"ref_id": "T210",
"role": "Arg1"
},
"tail": {
"ref_id": "T209",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R121",
"head": {
"ref_id": "T212",
"role": "Arg1"
},
"tail": {
"ref_id": "T211",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R122",
"head": {
"ref_id": "T209",
"role": "Arg1"
},
"tail": {
"ref_id": "T206",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R123",
"head": {
"ref_id": "T211",
"role": "Arg1"
},
"tail": {
"ref_id": "T206",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R124",
"head": {
"ref_id": "T216",
"role": "Arg1"
},
"tail": {
"ref_id": "T214",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R125",
"head": {
"ref_id": "T217",
"role": "Arg1"
},
"tail": {
"ref_id": "T214",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R126",
"head": {
"ref_id": "T215",
"role": "Arg1"
},
"tail": {
"ref_id": "T214",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R127",
"head": {
"ref_id": "T221",
"role": "Arg1"
},
"tail": {
"ref_id": "T219",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R128",
"head": {
"ref_id": "T220",
"role": "Arg1"
},
"tail": {
"ref_id": "T219",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R129",
"head": {
"ref_id": "T222",
"role": "Arg1"
},
"tail": {
"ref_id": "T223",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R130",
"head": {
"ref_id": "T223",
"role": "Arg1"
},
"tail": {
"ref_id": "T219",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R131",
"head": {
"ref_id": "T224",
"role": "Arg1"
},
"tail": {
"ref_id": "T225",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R132",
"head": {
"ref_id": "T226",
"role": "Arg1"
},
"tail": {
"ref_id": "T224",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R133",
"head": {
"ref_id": "T227",
"role": "Arg1"
},
"tail": {
"ref_id": "T228",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R134",
"head": {
"ref_id": "T228",
"role": "Arg1"
},
"tail": {
"ref_id": "T230",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R135",
"head": {
"ref_id": "T229",
"role": "Arg1"
},
"tail": {
"ref_id": "T230",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R136",
"head": {
"ref_id": "T233",
"role": "Arg1"
},
"tail": {
"ref_id": "T232",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R137",
"head": {
"ref_id": "T231",
"role": "Arg1"
},
"tail": {
"ref_id": "T233",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R138",
"head": {
"ref_id": "T234",
"role": "Arg1"
},
"tail": {
"ref_id": "T235",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R139",
"head": {
"ref_id": "T235",
"role": "Arg1"
},
"tail": {
"ref_id": "T236",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R140",
"head": {
"ref_id": "T237",
"role": "Arg1"
},
"tail": {
"ref_id": "T238",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R141",
"head": {
"ref_id": "T239",
"role": "Arg1"
},
"tail": {
"ref_id": "T240",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R142",
"head": {
"ref_id": "T242",
"role": "Arg1"
},
"tail": {
"ref_id": "T243",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R143",
"head": {
"ref_id": "T245",
"role": "Arg1"
},
"tail": {
"ref_id": "T246",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R144",
"head": {
"ref_id": "T247",
"role": "Arg1"
},
"tail": {
"ref_id": "T246",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R145",
"head": {
"ref_id": "T248",
"role": "Arg1"
},
"tail": {
"ref_id": "T246",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R146",
"head": {
"ref_id": "T249",
"role": "Arg1"
},
"tail": {
"ref_id": "T245",
"role": "Arg2"
},
"type": "semantically_same"
},
{
"id": "R147",
"head": {
"ref_id": "T251",
"role": "Arg1"
},
"tail": {
"ref_id": "T250",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R148",
"head": {
"ref_id": "T250",
"role": "Arg1"
},
"tail": {
"ref_id": "T249",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R149",
"head": {
"ref_id": "T253",
"role": "Arg1"
},
"tail": {
"ref_id": "T254",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R150",
"head": {
"ref_id": "T254",
"role": "Arg1"
},
"tail": {
"ref_id": "T255",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R151",
"head": {
"ref_id": "T259",
"role": "Arg1"
},
"tail": {
"ref_id": "T258",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R152",
"head": {
"ref_id": "T258",
"role": "Arg1"
},
"tail": {
"ref_id": "T257",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R153",
"head": {
"ref_id": "T260",
"role": "Arg1"
},
"tail": {
"ref_id": "T257",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R154",
"head": {
"ref_id": "T261",
"role": "Arg1"
},
"tail": {
"ref_id": "T259",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R155",
"head": {
"ref_id": "T262",
"role": "Arg1"
},
"tail": {
"ref_id": "T259",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R156",
"head": {
"ref_id": "T263",
"role": "Arg1"
},
"tail": {
"ref_id": "T259",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R157",
"head": {
"ref_id": "T265",
"role": "Arg1"
},
"tail": {
"ref_id": "T264",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R158",
"head": {
"ref_id": "T266",
"role": "Arg1"
},
"tail": {
"ref_id": "T265",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R159",
"head": {
"ref_id": "T267",
"role": "Arg1"
},
"tail": {
"ref_id": "T265",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R160",
"head": {
"ref_id": "T269",
"role": "Arg1"
},
"tail": {
"ref_id": "T268",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R161",
"head": {
"ref_id": "T267",
"role": "Arg1"
},
"tail": {
"ref_id": "T269",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R162",
"head": {
"ref_id": "T271",
"role": "Arg1"
},
"tail": {
"ref_id": "T270",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R163",
"head": {
"ref_id": "T273",
"role": "Arg1"
},
"tail": {
"ref_id": "T272",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R164",
"head": {
"ref_id": "T275",
"role": "Arg1"
},
"tail": {
"ref_id": "T274",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R165",
"head": {
"ref_id": "T276",
"role": "Arg1"
},
"tail": {
"ref_id": "T275",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R166",
"head": {
"ref_id": "T277",
"role": "Arg1"
},
"tail": {
"ref_id": "T275",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R167",
"head": {
"ref_id": "T278",
"role": "Arg1"
},
"tail": {
"ref_id": "T275",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R168",
"head": {
"ref_id": "T283",
"role": "Arg1"
},
"tail": {
"ref_id": "T282",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R169",
"head": {
"ref_id": "T284",
"role": "Arg1"
},
"tail": {
"ref_id": "T282",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R170",
"head": {
"ref_id": "T285",
"role": "Arg1"
},
"tail": {
"ref_id": "T282",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R171",
"head": {
"ref_id": "T286",
"role": "Arg1"
},
"tail": {
"ref_id": "T283",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R172",
"head": {
"ref_id": "T286",
"role": "Arg1"
},
"tail": {
"ref_id": "T284",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R173",
"head": {
"ref_id": "T286",
"role": "Arg1"
},
"tail": {
"ref_id": "T285",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R174",
"head": {
"ref_id": "T289",
"role": "Arg1"
},
"tail": {
"ref_id": "T288",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R175",
"head": {
"ref_id": "T287",
"role": "Arg1"
},
"tail": {
"ref_id": "T288",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R176",
"head": {
"ref_id": "T291",
"role": "Arg1"
},
"tail": {
"ref_id": "T290",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R177",
"head": {
"ref_id": "T290",
"role": "Arg1"
},
"tail": {
"ref_id": "T292",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R178",
"head": {
"ref_id": "T292",
"role": "Arg1"
},
"tail": {
"ref_id": "T288",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R179",
"head": {
"ref_id": "T294",
"role": "Arg1"
},
"tail": {
"ref_id": "T293",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R180",
"head": {
"ref_id": "T297",
"role": "Arg1"
},
"tail": {
"ref_id": "T296",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R181",
"head": {
"ref_id": "T302",
"role": "Arg1"
},
"tail": {
"ref_id": "T301",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R182",
"head": {
"ref_id": "T303",
"role": "Arg1"
},
"tail": {
"ref_id": "T304",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R183",
"head": {
"ref_id": "T305",
"role": "Arg1"
},
"tail": {
"ref_id": "T306",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R184",
"head": {
"ref_id": "T311",
"role": "Arg1"
},
"tail": {
"ref_id": "T310",
"role": "Arg2"
},
"type": "supports"
}
] |
A01 | <?xml version="1.0" encoding="UTF-8" standalone="no"?>
<Document xmlns:gate="http://www.gate.ac.uk" name="A01_S01_A_Powell_Optimization_Approach__for_Example-Based_Skinning_CITATION_PURPOSE_M_v1.xml">
878f0362cc6e7ed58f87ebe9655631d71db7b8522fb61a35b3a24df1c8fec64a
3ww8
http://dx.doi.org/10.1515/libr.2002.169
<Title>A Powell Optimization Approach for Example-Based Skinning in a Production Animation Environment</Title>
Xiao Xian ∗ John P. Lewis Nanyang Technological University Graphic Primitives Seah Hock Soon Nickson Fong Nanyang Technological University EggStoryCP Tian Feng Nanyang Technological University
<Abstract>We propose a layered framework for incorporating example-based skinning algorithms such as Pose Space Deformation or Shape-by-Example into an existing character animation system. The challenge in implementing example-based skinning in an existing system lies in the fact that it is generally believed that the interpolation of the examples is best performed before doing other skinning deformations (although there has been no analysis as to why this is the case), whereas the examples are specified by the user after the other deformations are performed. It is therefore necessary to invert the operation of these skinning and deformation operators. Existing systems typically allow layering of both basic skinning methods such as Skeleton Subspace Deformation (SSD) and other deformations such as lattices, etc., and commercial systems may further allow additional proprietary deformation algorithms as part of the character skinning. Unfortunately, understanding and accessing their various parameters can be laborious at best, and we do not have access to the algorithms in the case of commercial packages. The contributions of this paper are 1) a detailed analysis showing how inverting the skinning operations leads to better example interpolation, and 2) a demonstration</Abstract>
∗ xiao0003@ntu.edu.sg
that the black-box inverse can be accomplished in practice using Powell optimization, resulting in an improved example-based skinning capability for existing large animation systems.
<H2>Keywords: </H2>skinning, Powell optimization, computer animation
<H1>1 Introduction</H1>
With the help of modelling tools or capture devices, complicated 3D character models are widely used in fields of entertainment, virtual reality, medicine etc. The range of breathtaking realistic 3D models is only limited by the creativity of artists and resolution of devices. Driving 3D models in a natural and believable manner is not trivial, especially when the model is very detailed and playback of animation becomes quite heavy and time consuming. Each time when a frame goes wrong, a production cannot afford major revisions such as resculpting models or re-rigging skeletons. Therefore, providing a flexible and efficient solution to animation remains an open problem. Articulated character animation is a process of deforming the skin surface by manipulating influencing objects such as skeletons, IK, wire frames and Nurbs curves etc. Skeleton Subspace Deformation (SSD) is the predominant approach to character skinning at present. A nice review of SSD is given in [ 1 ]. SSD is widely used in games, virtual reality and other realtime applications due to its ease of implementation and low cost of computing. It provides the relation between characters and their underlying skeletons. Normally this relation is defined in the rest pose, and determines how characters move according to their skeletons thereafter. Sometimes, artists will edit the geometry of characters in the rest pose to fine-tune animations. This approach is not commonly applied, however, since editing in the rest pose will influence most other poses. On the other hand SSD is also notorious for artifacts at rotating elbows and extreme poses. For those applications that require visual fidelity, such as movies, SSD serves only as a basic framework, on which lots of more complicated deformation approaches are built as a compensation. Example based skinning methods such as Pose Space Deformation (PSD) are candidates for correcting SSD limitations. Example geometric models paired with underlying skeletons in different poses are provided by artists with carefully sculpting and posing. PSD smoothly interpolates these meshes in pose space and produces visually attractive animations. However, although PSD may be used as a compensation to the underlying SSD, and the animator specifies the PSD examples after the SSD has been performed, it is generally believed that the examples are best interpolated in the rest pose, before the SSD has been applied. Therefore the action of the SSD and any other deformations must be “inverted” in order to push the example compensation before these operations. Besides SSD, other skinning approaches such as rigid skinning, Free Form Deformation etc. can also be applied. Our goal is to incorporate examplebased skinning into a system having a variety of other skinning and deformation operations, and to be able to invert these operations regardless of their nature. Since SSD is the most representative in the family of basic skinning, we will discuss how it performs in the inverse operation of PSD scheme. For a simplified condition where only one joint rotation and two example poses are considered, we demonstrate this inverse strategy has a better performance than the same framework without it. The rest of this paper is organized as follows. After a review of related work, we will study details of SSD and PSD respectively in the third section. Then discussion of the inverse operation is presented in the fourth section including its implementation and reasons why this approach will improve the quality of deformation. In the fifth section, we propose a unified skinning framework by applying minimization theory with respect to other basic skinning schemes. Conclusion and some discussions of future work are presented in the last section.
1
<H1>2 Related Work</H1>
Besides the geometric solutions mentioned in the previous section, physical modelling and animation is another field providing realistic character simulations. Given physical principles, this category can generate more believable animation effects compared to its geometric counterpart. But they are seldom applied to interactive applications because of the high cost of computing and complicated algorithms. This paper is mainly dedicated to geometric solutions. Pose Space Deformation
[ 1 ] combines shape blending and Skeleton Subspace Deformation by formulating a scattered data interpolation problem over sculpted (or otherwise obtained) example poses. Character geometries in problematic poses will be re-sculpted by animators and then resulting displacement (referred as delta values in this paper) from the original geometries will be stored as “scattered data” for interpolation phase. The interpolation is performed in the pose space which consists of skeleton joints, or other potentially abstract controllers. Their values such as rotation degrees can be chosen as coordinates of the abstract pose space. After a model is posed and resculpted in different example poses, a multidimensional linear system is built by implementing an interpolation scheme using Radial Basis Functions (RBF), and the output of this system is the weights of each example pose. The final animation can be synthesized by linearly blending RBF functions with the solved weights. Related research efforts have improved the speed and power of example-based skinning. [ 2 ]
incorporate linear elements into RBF to produce constant changes between examples. [ 3 ] precompute principal components of the deformation influences for individual kinematic joints instead of storing displacements for key poses, thereby enabling realtime rendering large nonlinear finite element models of human hands. [ 4 ] introduce weighted pose space deformation for deforming realistic models of human hand. The latest work [ 5 ] identifies statistically relevant bones and approximates bone transforms from example mesh animations. Using established terminology from statistical modeling, these example-based approaches can be considered as non-parametric skin deformation methods. The data needed for these methods grows with the number of examples, but arbitrary deformations can be approximated as a result. Simpler parametric skinning approaches (of which SSD is the prototype) have a fixed number of parameters; these have also seen some development in recent years [ 6 ], [ 7 ]. Skinning using free form lattices [ 8 ], [ 9 ] or NURBS curves [ 10 ] instead of skeletons to drive character surface are also common practices in the entertainment production. Our framework implements existing PSD theory and the distinction is that we insert an optimization module into the PSD pipeline by applying a unified inverse approach assuming the knowledge of basic skinning is unavailable.
2
<H2>2.1 Our Contribution</H2>
We provide detailed reasons why and how the inverse operation can improve the results. For a simplified case, we show that the direction of deformed vertices from inverse skinning is a linear function of joint rotation, while in the forward approach, that direction is kept as a constant. This demonstration provides for the first time a clear theoretical reason why inverse operation is required. We formulate editing geometry in rest pose as an optimization problem and propose a unified framework which can be implemented on high-end commercial packages while allowing any proprietary skinning operators to be incorporated.
<H1>3 Skeleton Sub-Space Deformation</H1>
6 5 4 3 2 1 0
0 1 2 3 4 5 6 7 8 (a). (b).
Figure 1: (a). Skeleton Subspace Deformation; (b). Radial Basis Interpolation
Skeleton Subspace Deformation (SSD) is a basic algorithm that is used to define how the character surface deforms following movements of its underlying skeletons. The main idea is introduced by [ 11 ], and is also known as soft skinning, linear blending or Single Weight Enveloping (SWE). Due to its simplicity and efficiency, SSD is widely applied to interactive applications such as games and virtual reality, and it is implemented in most commercial animation packages. A skeleton should be rigged to a character surface beforehand, roughly based on the anatomy of the character and kinetic rules. The pose in which the skeleton is rigged normally is referred to as the rest pose. The basic relationship between surfaces and skeletons is defined at the rest pose, and all motions of the character will be influenced thereafter. If SSD is adopted to define this relation, each vertex or control point of the character surface is provided with a list of joints, that will influence it, along with the weight indicating the amount of influence. When the character is animated, the position of a vertex in the animated pose is the result of weighted linear blending of its transformation by each associated joint. We formulate SSD as:
1
M v p = SSD p (v r ) = ∑ ω k T pk v r k=1 r
For a vertex in rest pose v , its transformed position in pose p is v p . T pk means the kth joint’s transformation from rest pose to pose p. Readers can find details on how to compute T pk in
[ 1 ]. ω k is the corresponding weight. This weight is usually a function of distance between v r and its associated joints, and is defined when we apply SSD to the rigged character. Figure 1 (a). is a simple illustration of SSD with only two joint frames. Rectangles represent animated sections in each of two frames and the curve shows the blended result of both frames. Since vertex transformations can be easily implemented in the graphic card, SSD is very popular in circumstances that require animating a number of characters in real time. Some opportunities for control are provided to the animators. When a character goes wrong in some pose, animators can adjust joint influence weights. But the domain of adjusting one vertex in this way is strictly limited to the linear subspace formed by the vertex as transformed by joints influencing this vertex.
3
<H2>3.1 Pose Space Deformation</H2>
The famous SSD problem of “collapsed elbow” is recognized in
[ 1 ]
as being due to the fact that deforming is limited to a linear subspace. Because of this limitation SSD cannot synthesize many parts of a character skin involving complicated joint structures. Built on the SSD scheme, the Pose Space Deformation (PSD) is proposed by
[ 1 ] as a combination of SSD and shape blending providing nice solution to above mentioned problems. PSD can be performed in the following steps: • example pose setup: move the character to problematic poses, sculpt and store pose information (joint configuration x) including the amount of movement of each sculpted vertex (delta values d). • In the inverse PSD approach in this paper, we transform example models (d) to the rest pose (d r ): This step is trivial if the basic skinning, say SSD, is explicit. In our proposed framework, this step will be replaced by an optimization routine. • solving a linear system: we setup an interpolation scheme for delta Φ(x)ω = d and solve it to obtain weights ω for all example poses. • realtime synthesis: for an intermediate pose, we obtain the delta in the rest pose by interpolating example poses at runtime. We add this delta to the original character surface and then let SSD or any other skinning scheme finish the final transformation. For a vertex v, if sculpted in N example poses, then there are N delta d i , i = 0, . . . , N − 1 corresponding to each pose x i , i = 0, . . . , N − 1. These are converted to rest pose displacements using d i r = SSD −1 (d i ). We adopt Gaussian Radial Basis functions to interpolate d i r . First a N ∗ N matrix Φ is built with the (i, j)th element as φ ( x i − x j ), where x i − x j means the Euclidean distance between pose x i and pose x j , then we have a linear system:
2
W = Φ −1 D r
Here W and D r are column vectors with ith element ω i and d i r respectively. In the synthesis phase, for an intermediate pose x, we can obtain the delta d for this vertex by:
3
N−1 d = ∑ ω i φ ( x i − x ) i=0
For the Gaussian function φ (x) = e − σ x 2 2 , σ is used to control the “fall-off”. In Figure 1 (b)., we use Gaussian Radial Basis functions to interpolate 3 points. The blue and green curve represent σ = 1.0 and σ = 2.0 respectively. Other basis functions also can be candidates. Although PSD and improved example-based schemes have been discussed in many publications
[ 2 ], [ 3 ], [ 4 ]
, the reason why the inverse should be performed is still ambiguous. In the next section we will analyze this issue and demonstrate why inverting the SSD (and other deformations) in order to interpolate the examples in the rest pose is a the right choice.
<H1>4 Inverse Operation</H1>
This section will describe the implementation of our inverse algorithm and why it is an improvement. We still study SSD as the underlying skinning, since an explicit form of basic skinning can help to simplify our task of explanation. We call the PSD scheme without the inverse operation as “forward PSD”, and comparison to it will be used to demonstrate the superiority of the inverse method.
<H2>4.1 Implementation</H2>
In this sub-section, we will explain how inverse PSD works. For N examples, a vertex v is first transformed from rest pose by SSD to positions v i , i = 0, . . . , N − 1, then animators move it to example positions to obtain delta values d i , i = 0, . . . , N − 1. The final positions of v in example poses are v i + d i , i = 0, . . . , N − 1, and we call them target positions vt i . We summarize above operations as:
4
4
v i t = v i + d i = SSD i (v r ) + d i
where vr means the rest position of v and SSD i (∗) represents the equation 1. The “forward PSD” approach then concludes by interpolating d i as a function of pose. In the inverse approach we instead apply the inverse of SSD i (∗) to v i t to obtain a modified rest pose vertex v r i . The difference of vr i and vr produces new delta value d i r , which will be the input of linear system (equation 2) introduced in the previous section.
5
d i r = SSD −1 i (v t i ) − vr
In this step we need implement the inverse skinning operator SSD −1 . Since SSD is a 3D transformation, SSD −1 simply is the inverse transformation matrix generated by SSD. For the situation where other unknown skinning operations are adopted, we propose a unified framework which will be discussed in the following section. Next we build a new delta vector Dr with ith element as d i r , and replace D in equation 2 with d i r to get a new weight vector W r .
6
W r = Φ −1 Dr
In the synthesis phase, for an intermediate pose x we have:
7
N−1 d x r = ∑ ω i r φ ( x i − x )
i=0 then we add this d x r to v r and let SSD finish the rest of the job:
8
v x = SSD x (v r + d x r )
v x represents the final position of vertex v in pose x.
<H2>4.2 Deformation Direction Analysis</H2>
In this section we will analyze and compare the deforming effects of both forward PSD and inverse PSD, explaining why the inverse method is superior.. Given two examples as shown in Figure 2 (a) and (b) respectively, vertex v with the position v r in the rest pose ( 0 degrees ) is sculpted to a “target position” v ti in an example pose (90 degrees). The delta value in the first pose is zero. Then we apply forward and inverse PSD respectively to interpolate these two poses. For an intermediate pose x, we have two distinct deforming vertices resulting from two algorithms, as illustrated in Figure 3 , v ssd x , v x p , v x I p are the deformed positions from SSD, forward and inverse PSD in an intermediate pose x. We use two angles α p and α I p to analyze how directions of a deformed vertex change with the pose. In the forward case, α p is formed by the vector (v ssd x , v x p ) and the line y = Y v ssd x , where Y v x ssd is the y coordinate of v x ssd . For two examples shown in Figure 2 (a) and (b), we have delta values d 1 = [d 1x , d 1y ] and d 2 = [d 2x , d 2y ]. For equation 2: ω ω 2x 1x ω ω 2y 1y = φ φ 21 11 φ φ 12 22 −1 d d 2x 1x d d 1y 2y Because in the rest pose, we didn’t generate any movement for v , then d 1 = [d 1x , d 1y ] = [0, 0]. Taking the model in rest pose as an example is a common practice when applying shape interpolation, since interpolating effects from other examples should not change the original model in rest pose. Therefore, by solving above equation we have: ω 1x = φ 11 −1 d 1x + φ 12 −1 d 2x = φ 12 −1 d 2x ω 2x = φ 21 −1 d 1x + φ 22 −1 d 2x = φ 22 −1 d 2x = d 2x ω 1y = φ 11 −1 d 1y + φ 12 −1 d 2y = φ 12 −1 d 2y ω 2y = φ 21 −1 d 1y + φ 22 −1 d 2y = φ 22 −1 d 2y = d 2y where φ i −1 j is the (i, j)th element of Φ −1 , and if i = j, φ i −1 j = 1. Then in an intermediate pose x for α p , we have tan α p = d d x y . d y and d x are delta values in x, y coordinates computed from equation 3. With the above weight values, we have: tan α p = d d x y = ω ω 1x 1y φ φ (x (x − − x x 1 1 ) ) + + ω ω 2y 2x φ φ (x (x − − x x 2 2 ) ) = φ φ 12 12 −1 −1 d d 2x 2y φ φ (x (x − − x x 1 1 ) ) + + d d 2x 2y φ φ (x (x − − x x 2 2 ) ) = d 2y d 2x We can see this angle α p is a constant and depends only on the value of delta in the second pose d 2 = [d 2x , d 2y ]. Then we take a look at α I p in the inverse case. We transform two examples to rest pose to obtain delta values: d 1 r = [d 1x , d 1y ] = [0, 0] and d r 2 = [d 2x , d 2y ]. Since only the second joint is rotating, we simplify SSD as a rotation transformation ignoring other issues such as accumulating effects from the first joint: ⎛ ⎞ cos θ sin θ 0 SSD θ = ⎝ − sin θ cos θ 0 ⎠ . 0 0 1 where θ is the rotation angle of the second joint. Then for the vertex v r = [v 0x , v 0y ], SSD θ (v r ) transforms v from rest pose to [v SSD x , v SSD y ] = [v 0x cos θ − v 0y sin θ , v 0x sin θ + v 0y cos θ ]. In an intermediate pose x, we have its corresponding rest position as v r x = [v 0x + d x , v 0y + d y ], and here the [d x , d y ] are interpolated result computing from equation 3. We just apply the simplified SSD to v x r to obtain v I x p : v Inp x = (v 0x + d x ) cos θ − (v 0y + d y ) sin θ and v Inp y = (v 0x + d x ) sin θ + (v 0y + d y ) cos θ . Similarly, we compute the tangent of α I p : tan α I p = − v v Inp Inp x y − − v v SSD SSD x y = − d x sin θ + d y cos θ = − tan( β + θ ) d x cos θ − d y sin θ where tan β = d y = d 2y . Then we can see α I p = d x d 2x −( θ + β ), which is linearly proportional to the pose rotation θ . And now we take a look at a real cylinder model with one vertex sculpted in the second pose, shown in the Figure 4 . Forward PSD and the corresponding inverse PSD in the same poses (30, 45 and 60 degree of one rotated joint ) are illustrated respectively in Figure 5 . We can see that in forward case, the direction of deformed vertex always keeps the same with the example cylinder ( figure 4 ). For inverse PSD however, that direction is changed along with the rotation of the joint. The case described above is quite common in practice when animating shoulder, elbow, knee, hip-bone, neck, etc. All these parts would rotate from the rest pose with some angle to other poses. On the other hand, as a matter of experience, PSD is supposed to be a method as “local” correction, which means pose space should not be extended to a whole space that has to incorporate all influenced objects. Otherwise, large amount of unnecessary works of building examples will be required, and the distance between different poses is also meaningless. For example how to measure the distance between differing poses such as “lying down” and “pitching”?
Figure 2: (a). rest pose; (b). example pose with rotation of 90 degrees
Figure 3: deformation direction
5
Figure 4: A simple test case: two example poses with one vertex sculpted
<H1>5 A Unified Framework for Inverse Skinning Model</H1>
The above discussions assume that the basic skinning algorithm is SSD, but in many circumstances, other deformation schemes will be adopted
[ 9 ], [ 10 ], most of which have been implemented in most animation packages. Therefore we propose a unified framework in which no explicit inverse operation is necessitated. Given a basic skinning method supported by animation packages we can deform the original character model from rest pose to another specific pose. In the more general case, we need to replace SSD with SKINNING in the inverse skinning algorithm as such for equation 4: v i t = v i + d i = SKINNING i (v r ) + d i But this time we don’t implement the inverse of SKINNING as in equation 5. To find delta d i in the rest pose: v i = SKINNING i (v r ) + d i = SKINNING i (v r + d i ) we can setup a minimization problem to minimize the function:
6
Figure 5: Interpolated Poses: the rotated joint is at 30, 45 and 60 degree. left column: PSD; right column: inverse PSD.
9
f (d i ) = v i − SKINNING i (v r + d i ) 2
This function can be given to Powell’s method to find d i at the minimum of f (d ). For each example pose P i , we have a d i , then we can apply radial basis function to d i (i = 0 . . . n − 1) in pose space to obtain ω i (i = 0 . . . n − 1). In synthesis phase, a d x in an intermediate pose x can be computed by equation 3 based on its position x in pose space d x = ∑ n−1 i=0 ω i φ (||x − x i ||). Then we have the final synthesis result:
10
v x = SKINNING(v r + d x )
<H2>5.1 Direction Set (Powell’s) Methods for Minimization</H2>
For a minimization problem, there are many candidate algorithms according to the form of function, knowledge of the derivative, computing capacity, and requirements for the rate of convergence, etc. In our situation, the function form is not explicit, and the computing burden increases with the number of example poses increases. We will adopt Powell’s method as the solution to this minimization problem. One advantage of Powell’s classic method is that it does not need explicit computation of the function’s gradient
[ 12 ]. Because we are treating the skinning operations as a “black box”, their gradient is not available, so Powell’s method is suitable. Minimizing the function f (d ) in a particular direction is actually minimization problem of one variable, which is also called line minimization. Powell’s idea is trying to find each minimum of function f (d ) in different direction until f (d ) stops decreasing. How to choose the next direction is the main concern of Powell’s method, and it has been proved that after repeated cycles of M line minimizations on conjugate directions, the optimization will in due course converge to the minimum [ 12 ]. M is the dimensionality of the function f ().
<H2>5.2 System Overview</H2>
Figure 6: System Pipeline for Unified Inverse Skinning.
We implement this unified approach as a Maya plug-in. In Maya, “tweaking” is a procedure adding delta values to original surface vertices before any deformations. It is actu- ally Maya’s form of rest-pose editing for their built-in deformation operators. As presented in Figure 6 , the whole system is divided into two phases. The first phase is to find each delta in the rest pose corresponding to each example pose. Basic skinning provided by Maya is called in the loop of minimization scheme. The output of the first phase, the delta in the rest pose, is input to into the second phase that is a linear system performing RBF interpolation to obtain the PSD weights. In the synthesis process, for an intermediate pose x, a delta d x ( or d x r ) is synthesized by equation 7. The final deformed vertex is computed by Maya skinning as in equation 10.
7
<H2>5.3 Singular Poses</H2>
If the SSD transformation in equation 1 is singular, some types of inverse PSD deformation will not be possible, because any component of the desired deformation that lies in the null space of the SSD matrix will be ignored. Although singular cases are rare (one example is a joint with 180 o rotation and equal 2 1 , 2 1 weights on the two joint frames, which is an unrealistic case of selfintersection), it is possible to handle these cases with a small rearrangement of the inverse PSD approach. We reformulate the problem as f (y i ) = v i + w i − SKINNING i (v r + d i ) 2 + λ w i 2 where y i is a concatenated vector y i = [d i , w i ] and λ is an arbitrary small number. The final synthesis is then v x = SKINNING(v r + d x ) + w x where w x is interpolated after SKINNING by applying the same RBF scheme as used for d x (thus, only minimal code changed are required). The idea here is that, since w i is being minimized, it will be generally be zero, and will be non-zero only if it is not possible to obtain the desired deformation v i using SKINNING i (v r + d i ). In the case where the SSD transform is nearsingular, the solved d i can be much large than other d k , which can result in poorly posed interpolation. To address this case, we further modify the objective function as
f (y i ) = v i +w i −SKINNING i (v r +d i ) 2 + λ w i 2 + μ d i 2 where 0.0001 is a sufficient value for both λ and μ .
Figure 7: toad: a sequence of animated frames. left: PSD; right: inverse PSD.
Figure 8: toad: closeup of circled part from figure 9 . left: PSD; right: inverse PSD.
<H1>6 Conclusions and Discussions</H1>
Inverse skinning integrates SSD and shape interpolation more firmly than its forward rival. We demonstrate the direction of deformed vertex in inverse skinning is linearly proportional to joint rotations in a simplified example, while the forward PSD does not incorporate the direction information. Therefore the inverse approach presents better performance and more consistent interpolation ( Figure 7 to Figure 10 ). By formulating the inverse process as a minimization problem we propose a unified model not only for SSD but also for other skinning schemes, into which shape interpolation can be incorporated. But the minimizing process will introduce more cost. This cost depends on the size of deformed character, parameters of minimization methods (Powell) such as convergence precision, and the number of example poses. In addition the cost of the animation software must be considered (for example, the Maya API implements a run-time type interpretation system on all operations). The cost of the inverse operation is not critical, however, since it is a one time “setup” cost, and the compute time is insignificant compared to the human time required to sculpt the desired deformations. Once the linear system is solved, the synthesis is potentially realtime since no extra computing is involved in this process compared to the forward PSD. We implement this unified example-based approach as a Maya plugin. It interoperates with their closed-source “Smooth Skinning” deformation.
8
<H1>Acknowledgements</H1>
Authors would like to thank artists from EggStory Creative Production. We also thank several of the reviewers for comments which improved this work
<H1>References</H1>
[1] J. P. Lewis, Matt Cordner, and Nickson Fong. Pose space deformation: a unified approach to shape interpolation and skeleton-driven deformation. In SIGGRAPH 2000: Proceedings of the 27th annual conference on Computer graphics and interactive techniques, pages 165– 172, New York, NY, USA, 2000. ACM Press/Addison-Wesley Publishing Co.
[2] Peter-Pike J. Sloan, III Charles F. Rose, and Michael F. Cohen. Shape by example. In SI3D 2001: Proceedings of the 2001 symposium on Interactive 3D graphics, pages 135–143, New York, NY, USA, 2001. ACM Press.
[3] Paul G. Kry, Doug L. James, and Dinesh K. Pai. Eigenskin: real time large deformation character skinning in hardware. In SCA 2002: Proceedings of the 2002 ACM SIGGRAPH/Eurographics symposium on Computer animation, pages 153–159, New York, NY, USA, 2002. ACM Press.
[4] Tsuneya Kurihara and Natsuki Miyata. Modeling deformable human hands from medical images. In SCA 2004: Proceedings of the 2004 ACM SIGGRAPH/Eurographics symposium on Computer animation, pages 355–363, New York, NY, USA, 2004. ACM Press.
[5] Doug L. James and Christopher D. Twigg. Skinning mesh animations. ACM Trans. Graph., 24(3):399–407, 2005.
[6] Xiaohuan Corina Wang and Cary Phillips. Multi-weight enveloping: least-squares approximation techniques for skin animation. In SCA 2002: Proceedings of the 2002 ACM SIGGRAPH/Eurographics symposium on Computer animation, pages 129–138, New York, NY, USA, 2002. ACM Press.
[7] Alex Mohr and Michael Gleicher. Building efficient, accurate character skins from examples. ACM Trans. Graph., 22(3):562– 568, 2003.
[8] Karan Singh and Evangelos Kokkevis. Skinning characters using surface oriented free-form deformations. In Graphics Interface, pages 35–42, 2000.
[9] Ron MacCracken and Kenneth I. Joy. Free-form deformations with lattices of arbitrary topology. In SIGGRAPH 96: Proceedings of the 23rd annual conference on Computer graphics and interactive techniques, pages 181–188, New York, NY, USA, 1996. ACM Press.
[10] Karan Singh and Eugene Fiume. Wires: a geometric deformation technique. In SIGGRAPH 98: Proceedings of the 25th annual conference on Computer graphics and interactive techniques, pages 405– 414, New York, NY, USA, 1998. ACM Press.
[11] N. Magnenat-Thalmann, R. Laperrire, and D. Thalmann. Joint-dependent local deformations for hand animation and object grasping. In Proceedings on Graphics interface ’88, pages 26–33, Toronto, Ont., Canada, Canada, 1988. Canadian Information Processing Society.
[12] William H. Press, William T. Vetterling, Saul A. Teukolsky, and Brian P. Flannery. Numerical Recipes in C++: the art of scientific computing. 2002.
9
Figure 9: PSD vs. Inverse PSD: we have two examples, one in rest pose and the other is sculpted when the thumb is rotated down 66.8 degrees. SSD shows an obvious collapse. The intermediate poses from SSD, PSD and inverse PSD are shown in the second (45 degrees) and third (60 degrees) row respectively.
examples
interpolated models computed by Powell optimization approach
Figure 10: human arm
10
</Document>
| [
{
"offsets": [
[
2417,
2522
]
],
"text": "['complicated 3D character models are widely used in fields of entertainment, virtual reality, medicine etc']",
"type": "background_claim",
"id": "T1"
},
{
"offsets": [
[
2524,
2640
]
],
"text": "['The range of breathtaking realistic 3D models is only limited by the creativity of artists and resolution of devices']",
"type": "background_claim",
"id": "T2"
},
{
"offsets": [
[
2855,
2897
]
],
"text": "['a production cannot afford major revisions']",
"type": "background_claim",
"id": "T4"
},
{
"offsets": [
[
2961,
3041
]
],
"text": "['providing a flexible and efficient solution to animation remains an open problem']",
"type": "own_claim",
"id": "T5"
},
{
"offsets": [
[
3211,
3307
]
],
"text": "['Skeleton Subspace Deformation (SSD) is the predominant approach to character skinning at present']",
"type": "background_claim",
"id": "T6"
},
{
"offsets": [
[
2642,
2709
]
],
"text": "['Driving 3D models in a natural and believable manner is not trivial']",
"type": "background_claim",
"id": "T3"
},
{
"offsets": [
[
3433,
3459
]
],
"text": "['its ease of implementation']",
"type": "data",
"id": "T7"
},
{
"offsets": [
[
3464,
3485
]
],
"text": "['low cost of computing']",
"type": "data",
"id": "T8"
},
{
"offsets": [
[
3349,
3425
]
],
"text": "['SSD is widely used in games, virtual reality and other realtime applications']",
"type": "background_claim",
"id": "T9"
},
{
"offsets": [
[
3702,
3787
]
],
"text": "['artists will edit the geometry of characters in the rest pose to fine-tune animations']",
"type": "background_claim",
"id": "T10"
},
{
"offsets": [
[
3789,
3826
]
],
"text": "['This approach is not commonly applied']",
"type": "background_claim",
"id": "T11"
},
{
"offsets": [
[
3843,
3899
]
],
"text": "['editing in the rest pose will influence most other poses']",
"type": "background_claim",
"id": "T12"
},
{
"offsets": [
[
3919,
3991
]
],
"text": "['SSD is also notorious for artifacts at rotating elbows and extreme poses']",
"type": "background_claim",
"id": "T13"
},
{
"offsets": [
[
4186,
4299
]
],
"text": "['Example based skinning methods such as Pose Space Deformation (PSD) are candidates for correcting SSD limitations']",
"type": "background_claim",
"id": "T14"
},
{
"offsets": [
[
4555,
4610
]
],
"text": "['PSD may be used as a compensation to the underlying SSD']",
"type": "background_claim",
"id": "T15"
},
{
"offsets": [
[
4616,
4688
]
],
"text": "['the animator specifies the PSD examples after the SSD has been performed']",
"type": "background_claim",
"id": "T16"
},
{
"offsets": [
[
4720,
4804
]
],
"text": "['the examples are best interpolated in the rest pose, before the SSD has been applied']",
"type": "background_claim",
"id": "T17"
},
{
"offsets": [
[
4816,
4949
]
],
"text": "['the action of the SSD and any other deformations must be “inverted” in order to push the example compensation before these operations']",
"type": "background_claim",
"id": "T18"
},
{
"offsets": [
[
5264,
5326
]
],
"text": "['SSD is the most representative in the family of basic skinning']",
"type": "background_claim",
"id": "T19"
},
{
"offsets": [
[
5510,
5591
]
],
"text": "['this inverse strategy has a better performance than the same framework without it']",
"type": "own_claim",
"id": "T20"
},
{
"offsets": [
[
5866,
5919
]
],
"text": "['this approach will improve the quality of deformation']",
"type": "own_claim",
"id": "T21"
},
{
"offsets": [
[
6489,
6540
]
],
"text": "['they are seldom applied to interactive applications']",
"type": "background_claim",
"id": "T22"
},
{
"offsets": [
[
6552,
6578
]
],
"text": "['the high cost of computing']",
"type": "data",
"id": "T23"
},
{
"offsets": [
[
6583,
6605
]
],
"text": "['complicated algorithms']",
"type": "data",
"id": "T24"
},
{
"offsets": [
[
6385,
6483
]
],
"text": "['this category can generate more believable animation effects compared to its geometric counterpart']",
"type": "background_claim",
"id": "T25"
},
{
"offsets": [
[
6662,
6684
]
],
"text": "['Pose Space Deformation']",
"type": "background_claim",
"id": "T26"
},
{
"offsets": [
[
6692,
6855
]
],
"text": "['combines shape blending and Skeleton Subspace Deformation by formulating a scattered data interpolation problem over sculpted (or otherwise obtained) example poses']",
"type": "background_claim",
"id": "T27"
},
{
"offsets": [
[
6686,
6691
]
],
"text": "['[ 1 ]']",
"type": "data",
"id": "T28"
},
{
"offsets": [
[
7666,
7750
]
],
"text": "['Related research efforts have improved the speed and power of example-based skinning']",
"type": "background_claim",
"id": "T29"
},
{
"offsets": [
[
7759,
7840
]
],
"text": "['incorporate linear elements into RBF to produce constant changes between examples']",
"type": "background_claim",
"id": "T30"
},
{
"offsets": [
[
7752,
7757
]
],
"text": "['[ 2 ]']",
"type": "data",
"id": "T31"
},
{
"offsets": [
[
7848,
7988
]
],
"text": "['precompute principal components of the deformation influences for individual kinematic joints instead of storing displacements for key poses']",
"type": "background_claim",
"id": "T32"
},
{
"offsets": [
[
7998,
8078
]
],
"text": "['enabling realtime rendering large nonlinear finite element models of human hands']",
"type": "background_claim",
"id": "T33"
},
{
"offsets": [
[
7842,
7847
]
],
"text": "['[ 3 ]']",
"type": "data",
"id": "T34"
},
{
"offsets": [
[
8086,
8172
]
],
"text": "['introduce weighted pose space deformation for deforming realistic models of human hand']",
"type": "background_claim",
"id": "T35"
},
{
"offsets": [
[
8080,
8085
]
],
"text": "['[ 4 ]']",
"type": "data",
"id": "T36"
},
{
"offsets": [
[
8356,
8447
]
],
"text": "['these example-based approaches can be considered as non-parametric skin deformation methods']",
"type": "background_claim",
"id": "T37"
},
{
"offsets": [
[
8196,
8297
]
],
"text": "['identifies statistically relevant bones and approximates bone transforms from example mesh animations']",
"type": "background_claim",
"id": "T38"
},
{
"offsets": [
[
8190,
8195
]
],
"text": "['[ 5 ]']",
"type": "data",
"id": "T39"
},
{
"offsets": [
[
8684,
8737
]
],
"text": "['these have also seen some development in recent years']",
"type": "background_claim",
"id": "T40"
},
{
"offsets": [
[
8738,
8743
]
],
"text": "['[ 6 ]']",
"type": "data",
"id": "T41"
},
{
"offsets": [
[
8745,
8750
]
],
"text": "['[ 7 ]']",
"type": "data",
"id": "T42"
},
{
"offsets": [
[
8870,
8927
]
],
"text": "['are also common practices in the entertainment production']",
"type": "background_claim",
"id": "T43"
},
{
"offsets": [
[
8752,
8785
]
],
"text": "['Skinning using free form lattices']",
"type": "background_claim",
"id": "T44"
},
{
"offsets": [
[
8802,
8814
]
],
"text": "['NURBS curves']",
"type": "background_claim",
"id": "T45"
},
{
"offsets": [
[
8815,
8821
]
],
"text": "['[ 10 ]']",
"type": "data",
"id": "T46"
},
{
"offsets": [
[
8786,
8791
]
],
"text": "['[ 8 ]']",
"type": "data",
"id": "T47"
},
{
"offsets": [
[
8793,
8798
]
],
"text": "['[ 9 ]']",
"type": "data",
"id": "T48"
},
{
"offsets": [
[
9263,
9308
]
],
"text": "['the inverse operation can improve the results']",
"type": "own_claim",
"id": "T49"
},
{
"offsets": [
[
9590,
9619
]
],
"text": "['inverse operation is required']",
"type": "own_claim",
"id": "T50"
},
{
"offsets": [
[
10442,
10525
]
],
"text": "['SSD is widely applied to interactive applications such as games and virtual reality']",
"type": "background_claim",
"id": "T51"
},
{
"offsets": [
[
10531,
10586
]
],
"text": "['it is implemented in most commercial animation packages']",
"type": "background_claim",
"id": "T52"
},
{
"offsets": [
[
10411,
10425
]
],
"text": "['its simplicity']",
"type": "data",
"id": "T53"
},
{
"offsets": [
[
10430,
10440
]
],
"text": "['efficiency']",
"type": "data",
"id": "T54"
},
{
"offsets": [
[
12084,
12179
]
],
"text": "['SSD is very popular in circumstances that require animating a number of characters in real time']",
"type": "background_claim",
"id": "T55"
},
{
"offsets": [
[
12014,
12082
]
],
"text": "['vertex transformations can be easily implemented in the graphic card']",
"type": "data",
"id": "T56"
},
{
"offsets": [
[
12335,
12494
]
],
"text": "['the domain of adjusting one vertex in this way is strictly limited to the linear subspace formed by the vertex as transformed by joints influencing this vertex']",
"type": "background_claim",
"id": "T57"
},
{
"offsets": [
[
12243,
12329
]
],
"text": "['When a character goes wrong in some pose, animators can adjust joint influence weights']",
"type": "background_claim",
"id": "T58"
},
{
"offsets": [
[
12741,
12832
]
],
"text": "['SSD cannot synthesize many parts of a character skin involving complicated joint structures']",
"type": "background_claim",
"id": "T59"
},
{
"offsets": [
[
12671,
12712
]
],
"text": "['deforming is limited to a linear subspace']",
"type": "background_claim",
"id": "T60"
},
{
"offsets": [
[
12634,
12639
]
],
"text": "['[ 1 ]']",
"type": "data",
"id": "T61"
},
{
"offsets": [
[
12917,
13008
]
],
"text": "['a combination of SSD and shape blending providing nice solution to above mentioned problems']",
"type": "background_claim",
"id": "T62"
},
{
"offsets": [
[
12908,
12913
]
],
"text": "['[ 1 ]']",
"type": "data",
"id": "T63"
},
{
"offsets": [
[
14983,
15062
]
],
"text": "['PSD and improved example-based schemes have been discussed in many publications']",
"type": "background_claim",
"id": "T64"
},
{
"offsets": [
[
15064,
15069
]
],
"text": "['[ 2 ]']",
"type": "data",
"id": "T65"
},
{
"offsets": [
[
15071,
15076
]
],
"text": "['[ 3 ]']",
"type": "data",
"id": "T66"
},
{
"offsets": [
[
15078,
15083
]
],
"text": "['[ 4 ]']",
"type": "data",
"id": "T67"
},
{
"offsets": [
[
15086,
15151
]
],
"text": "['the reason why the inverse should be performed is still ambiguous']",
"type": "own_claim",
"id": "T68"
},
{
"offsets": [
[
15220,
15338
]
],
"text": "['inverting the SSD (and other deformations) in order to interpolate the examples in the rest pose is a the right choice']",
"type": "own_claim",
"id": "T69"
},
{
"offsets": [
[
15478,
15498
]
],
"text": "['it is an improvement']",
"type": "own_claim",
"id": "T70"
},
{
"offsets": [
[
15754,
15791
]
],
"text": "['the superiority of the inverse method']",
"type": "own_claim",
"id": "T71"
},
{
"offsets": [
[
16991,
17058
]
],
"text": "['SSD −1 simply is the inverse transformation matrix generated by SSD']",
"type": "own_claim",
"id": "T72"
},
{
"offsets": [
[
16963,
16989
]
],
"text": "['SSD is a 3D transformation']",
"type": "data",
"id": "T73"
},
{
"offsets": [
[
17950,
17980
]
],
"text": "['the inverse method is superior']",
"type": "own_claim",
"id": "T74"
},
{
"offsets": [
[
19098,
19196
]
],
"text": "['Taking the model in rest pose as an example is a common practice when applying shape interpolation']",
"type": "background_claim",
"id": "T75"
},
{
"offsets": [
[
19204,
19295
]
],
"text": "['interpolating effects from other examples should not change the original model in rest pose']",
"type": "background_claim",
"id": "T76"
},
{
"offsets": [
[
19334,
19603
]
],
"text": "['we have: ω 1x = φ 11 −1 d 1x + φ 12 −1 d 2x = φ 12 −1 d 2x ω 2x = φ 21 −1 d 1x + φ 22 −1 d 2x = φ 22 −1 d 2x = d 2x ω 1y = φ 11 −1 d 1y + φ 12 −1 d 2y = φ 12 −1 d 2y ω 2y = φ 21 −1 d 1y + φ 22 −1 d 2y = φ 22 −1 d 2y = d 2y where φ i −1 j is the (i, j)th element of Φ −1']",
"type": "own_claim",
"id": "T77"
},
{
"offsets": [
[
20353,
20469
]
],
"text": "['we simplify SSD as a rotation transformation ignoring other issues such as accumulating effects from the first joint']",
"type": "own_claim",
"id": "T78"
},
{
"offsets": [
[
20318,
20351
]
],
"text": "['only the second joint is rotating']",
"type": "own_claim",
"id": "T79"
},
{
"offsets": [
[
21667,
21764
]
],
"text": "['in forward case, the direction of deformed vertex always keeps the same with the example cylinder']",
"type": "own_claim",
"id": "T80"
},
{
"offsets": [
[
21765,
21777
]
],
"text": "['( figure 4 )']",
"type": "data",
"id": "T81"
},
{
"offsets": [
[
21779,
21866
]
],
"text": "['For inverse PSD however, that direction is changed along with the rotation of the joint']",
"type": "own_claim",
"id": "T82"
},
{
"offsets": [
[
22106,
22158
]
],
"text": "['PSD is supposed to be a method as “local” correction']",
"type": "own_claim",
"id": "T83"
},
{
"offsets": [
[
22172,
22269
]
],
"text": "['pose space should not be extended to a whole space that has to incorporate all influenced objects']",
"type": "own_claim",
"id": "T84"
},
{
"offsets": [
[
22082,
22104
]
],
"text": "['a matter of experience']",
"type": "data",
"id": "T85"
},
{
"offsets": [
[
22282,
22353
]
],
"text": "['large amount of unnecessary works of building examples will be required']",
"type": "own_claim",
"id": "T86"
},
{
"offsets": [
[
22359,
22415
]
],
"text": "['the distance between different poses is also meaningless']",
"type": "own_claim",
"id": "T87"
},
{
"offsets": [
[
23066,
23128
]
],
"text": "['most of which have been implemented in most animation packages']",
"type": "background_claim",
"id": "T88"
},
{
"offsets": [
[
22985,
23049
]
],
"text": "['in many circumstances, other deformation schemes will be adopted']",
"type": "background_claim",
"id": "T89"
},
{
"offsets": [
[
23058,
23064
]
],
"text": "['[ 10 ]']",
"type": "data",
"id": "T90"
},
{
"offsets": [
[
23051,
23056
]
],
"text": "['[ 9 ]']",
"type": "data",
"id": "T91"
},
{
"offsets": [
[
23289,
23371
]
],
"text": "['we can deform the original character model from rest pose to another specific pose']",
"type": "own_claim",
"id": "T92"
},
{
"offsets": [
[
25039,
25152
]
],
"text": "['One advantage of Powell’s classic method is that it does not need explicit computation of the function’s gradient']",
"type": "background_claim",
"id": "T93"
},
{
"offsets": [
[
25154,
25160
]
],
"text": "['[ 12 ]']",
"type": "data",
"id": "T94"
},
{
"offsets": [
[
25264,
25291
]
],
"text": "['Powell’s method is suitable']",
"type": "own_claim",
"id": "T95"
},
{
"offsets": [
[
25170,
25259
]
],
"text": "['we are treating the skinning operations as a “black box”, their gradient is not available']",
"type": "data",
"id": "T96"
},
{
"offsets": [
[
25659,
25789
]
],
"text": "['after repeated cycles of M line minimizations on conjugate directions, the optimization will in due course converge to the minimum']",
"type": "background_claim",
"id": "T97"
},
{
"offsets": [
[
25790,
25796
]
],
"text": "['[ 12 ]']",
"type": "data",
"id": "T98"
},
{
"offsets": [
[
26918,
27029
]
],
"text": "['If the SSD transformation in equation 1 is singular, some types of inverse PSD deformation will not be possible']",
"type": "own_claim",
"id": "T99"
},
{
"offsets": [
[
27039,
27141
]
],
"text": "['any component of the desired deformation that lies in the null space of the SSD matrix will be ignored']",
"type": "data",
"id": "T100"
},
{
"offsets": [
[
27152,
27175
]
],
"text": "['singular cases are rare']",
"type": "own_claim",
"id": "T101"
},
{
"offsets": [
[
27324,
27415
]
],
"text": "['it is possible to handle these cases with a small rearrangement of the inverse PSD approach']",
"type": "own_claim",
"id": "T102"
},
{
"offsets": [
[
27177,
27321
]
],
"text": "['one example is a joint with 180 o rotation and equal 2 1 , 2 1 weights on the two joint frames, which is an unrealistic case of selfintersection']",
"type": "data",
"id": "T103"
},
{
"offsets": [
[
27847,
27875
]
],
"text": "['it will be generally be zero']",
"type": "own_claim",
"id": "T104"
},
{
"offsets": [
[
27823,
27845
]
],
"text": "['w i is being minimized']",
"type": "data",
"id": "T105"
},
{
"offsets": [
[
27881,
27992
]
],
"text": "['will be non-zero only if it is not possible to obtain the desired deformation v i using SKINNING i (v r + d i )']",
"type": "own_claim",
"id": "T106"
},
{
"offsets": [
[
28651,
28742
]
],
"text": "['Inverse skinning integrates SSD and shape interpolation more firmly than its forward rival.']",
"type": "own_claim",
"id": "T107"
},
{
"offsets": [
[
28960,
29042
]
],
"text": "['the inverse approach presents better performance and more consistent interpolation']",
"type": "own_claim",
"id": "T108"
},
{
"offsets": [
[
29045,
29066
]
],
"text": "['Figure 7 to Figure 10']",
"type": "data",
"id": "T109"
},
{
"offsets": [
[
28758,
28948
]
],
"text": "['the direction of deformed vertex in inverse skinning is linearly proportional to joint rotations in a simplified example, while the forward PSD does not incorporate the direction information']",
"type": "own_claim",
"id": "T110"
},
{
"offsets": [
[
29268,
29315
]
],
"text": "['the minimizing process will introduce more cost']",
"type": "own_claim",
"id": "T111"
},
{
"offsets": [
[
29640,
29689
]
],
"text": "['The cost of the inverse operation is not critical']",
"type": "own_claim",
"id": "T112"
},
{
"offsets": [
[
29706,
29735
]
],
"text": "['it is a one time “setup” cost']",
"type": "own_claim",
"id": "T113"
},
{
"offsets": [
[
29741,
29845
]
],
"text": "['the compute time is insignificant compared to the human time required to sculpt the desired deformations']",
"type": "own_claim",
"id": "T114"
},
{
"offsets": [
[
2835,
2853
]
],
"text": "['a frame goes wrong']",
"type": "data",
"id": "T115"
},
{
"offsets": [
[
6364,
6383
]
],
"text": "['physical principles']",
"type": "data",
"id": "T116"
},
{
"offsets": [
[
2727,
2753
]
],
"text": "['the model is very detailed']",
"type": "data",
"id": "T117"
},
{
"offsets": [
[
2758,
2818
]
],
"text": "['playback of animation becomes quite heavy and time consuming']",
"type": "data",
"id": "T118"
},
{
"offsets": [
[
3993,
4044
]
],
"text": "['For those applications that require visual fidelity']",
"type": "background_claim",
"id": "T119"
},
{
"offsets": [
[
4062,
4098
]
],
"text": "['SSD serves only as a basic framework']",
"type": "background_claim",
"id": "T120"
},
{
"offsets": [
[
4439,
4535
]
],
"text": "['PSD smoothly interpolates these meshes in pose space and produces visually attractive animations']",
"type": "background_claim",
"id": "T121"
},
{
"offsets": [
[
4951,
4989
]
],
"text": "['Besides SSD, other skinning approaches']",
"type": "background_claim",
"id": "T122"
},
{
"offsets": [
[
5041,
5060
]
],
"text": "['can also be applied']",
"type": "background_claim",
"id": "T123"
},
{
"offsets": [
[
4998,
5012
]
],
"text": "['rigid skinning']",
"type": "data",
"id": "T124"
},
{
"offsets": [
[
5014,
5035
]
],
"text": "['Free Form Deformation']",
"type": "data",
"id": "T125"
},
{
"offsets": [
[
6265,
6356
]
],
"text": "['physical modelling and animation is another field providing realistic character simulations']",
"type": "background_claim",
"id": "T126"
},
{
"offsets": [
[
8578,
8682
]
],
"text": "['Simpler parametric skinning approaches (of which SSD is the prototype) have a fixed number of parameters']",
"type": "background_claim",
"id": "T127"
},
{
"offsets": [
[
9346,
9441
]
],
"text": "['the direction of deformed vertices from inverse skinning is a linear function of joint rotation']",
"type": "own_claim",
"id": "T128"
},
{
"offsets": [
[
9449,
9510
]
],
"text": "['in the forward approach, that direction is kept as a constant']",
"type": "own_claim",
"id": "T129"
},
{
"offsets": [
[
9695,
9848
]
],
"text": "['propose a unified framework which can be implemented on high-end commercial packages while allowing any proprietary skinning operators to be incorporated']",
"type": "own_claim",
"id": "T130"
},
{
"offsets": [
[
10115,
10276
]
],
"text": "['Skeleton Subspace Deformation (SSD) is a basic algorithm that is used to define how the character surface deforms following movements of its underlying skeletons']",
"type": "background_claim",
"id": "T131"
},
{
"offsets": [
[
10309,
10315
]
],
"text": "['[ 11 ]']",
"type": "data",
"id": "T132"
},
{
"offsets": [
[
10949,
10987
]
],
"text": "['SSD is adopted to define this relation']",
"type": "data",
"id": "T133"
},
{
"offsets": [
[
10989,
11158
]
],
"text": "['each vertex or control point of the character surface is provided with a list of joints, that will influence it, along with the weight indicating the amount of influence']",
"type": "background_claim",
"id": "T134"
},
{
"offsets": [
[
11192,
11326
]
],
"text": "['the position of a vertex in the animated pose is the result of weighted linear blending of its transformation by each associated joint']",
"type": "background_claim",
"id": "T135"
},
{
"offsets": [
[
11165,
11190
]
],
"text": "['the character is animated']",
"type": "data",
"id": "T136"
},
{
"offsets": [
[
12572,
12615
]
],
"text": "['The famous SSD problem of “collapsed elbow”']",
"type": "background_claim",
"id": "T137"
},
{
"offsets": [
[
13915,
13942
]
],
"text": "['sculpted in N example poses']",
"type": "data",
"id": "T138"
},
{
"offsets": [
[
13900,
13910
]
],
"text": "['a vertex v']",
"type": "data",
"id": "T139"
},
{
"offsets": [
[
13949,
14047
]
],
"text": "['there are N delta d i , i = 0, . . . , N − 1 corresponding to each pose x i , i = 0, . . . , N − 1']",
"type": "own_claim",
"id": "T140"
},
{
"offsets": [
[
15515,
15545
]
],
"text": "['SSD as the underlying skinning']",
"type": "own_claim",
"id": "T141"
},
{
"offsets": [
[
15553,
15632
]
],
"text": "['an explicit form of basic skinning can help to simplify our task of explanation']",
"type": "own_claim",
"id": "T142"
},
{
"offsets": [
[
19004,
19059
]
],
"text": "['in the rest pose, we didn’t generate any movement for v']",
"type": "data",
"id": "T143"
},
{
"offsets": [
[
19067,
19096
]
],
"text": "['d 1 = [d 1x , d 1y ] = [0, 0]']",
"type": "own_claim",
"id": "T144"
},
{
"offsets": [
[
19308,
19333
]
],
"text": "['by solving above equation']",
"type": "data",
"id": "T145"
},
{
"offsets": [
[
19613,
19618
]
],
"text": "['i = j']",
"type": "data",
"id": "T146"
},
{
"offsets": [
[
19620,
19632
]
],
"text": "['φ i −1 j = 1']",
"type": "own_claim",
"id": "T147"
},
{
"offsets": [
[
20030,
20137
]
],
"text": "['this angle α p is a constant and depends only on the value of delta in the second pose d 2 = [d 2x , d 2y ]']",
"type": "own_claim",
"id": "T148"
},
{
"offsets": [
[
19817,
20018
]
],
"text": "['tan α p = d d x y = ω ω 1x 1y φ φ (x (x − − x x 1 1 ) ) + + ω ω 2y 2x φ φ (x (x − − x x 2 2 ) ) = φ φ 12 12 −1 −1 d d 2x 2y φ φ (x (x − − x x 1 1 ) ) + + d d 2x 2y φ φ (x (x − − x x 2 2 ) ) = d 2y d 2x']",
"type": "own_claim",
"id": "T149"
},
{
"offsets": [
[
19683,
19700
]
],
"text": "['tan α p = d d x y']",
"type": "own_claim",
"id": "T150"
},
{
"offsets": [
[
21868,
21920
]
],
"text": "['The case described above is quite common in practice']",
"type": "own_claim",
"id": "T151"
},
{
"offsets": [
[
21926,
21973
]
],
"text": "['animating shoulder, elbow, knee, hip-bone, neck']",
"type": "data",
"id": "T152"
},
{
"offsets": [
[
21980,
22058
]
],
"text": "['All these parts would rotate from the rest pose with some angle to other poses']",
"type": "own_claim",
"id": "T153"
},
{
"offsets": [
[
23151,
23225
]
],
"text": "['a unified framework in which no explicit inverse operation is necessitated']",
"type": "own_claim",
"id": "T154"
},
{
"offsets": [
[
23233,
23288
]
],
"text": "['a basic skinning method supported by animation packages']",
"type": "data",
"id": "T155"
},
{
"offsets": [
[
23494,
23537
]
],
"text": "['v i t = v i + d i = SKINNING i (v r ) + d i']",
"type": "own_claim",
"id": "T156"
},
{
"offsets": [
[
23399,
23492
]
],
"text": "['we need to replace SSD with SKINNING in the inverse skinning algorithm as such for equation 4']",
"type": "own_claim",
"id": "T157"
},
{
"offsets": [
[
23376,
23397
]
],
"text": "['the more general case']",
"type": "data",
"id": "T158"
},
{
"offsets": [
[
23613,
23765
]
],
"text": "['To find delta d i in the rest pose: v i = SKINNING i (v r ) + d i = SKINNING i (v r + d i ) we can setup a minimization problem to minimize the function']",
"type": "own_claim",
"id": "T159"
},
{
"offsets": [
[
24028,
24110
]
],
"text": "['This function can be given to Powell’s method to find d i at the minimum of f (d )']",
"type": "own_claim",
"id": "T160"
},
{
"offsets": [
[
24112,
24153
]
],
"text": "['For each example pose P i , we have a d i']",
"type": "data",
"id": "T161"
},
{
"offsets": [
[
24161,
24270
]
],
"text": "['we can apply radial basis function to d i (i = 0 . . . n − 1) in pose space to obtain ω i (i = 0 . . . n − 1)']",
"type": "own_claim",
"id": "T162"
},
{
"offsets": [
[
25293,
25398
]
],
"text": "['Minimizing the function f (d ) in a particular direction is actually minimization problem of one variable']",
"type": "own_claim",
"id": "T163"
},
{
"offsets": [
[
25440,
25556
]
],
"text": "['Powell’s idea is trying to find each minimum of function f (d ) in different direction until f (d ) stops decreasing']",
"type": "background_claim",
"id": "T164"
},
{
"offsets": [
[
25558,
25629
]
],
"text": "['How to choose the next direction is the main concern of Powell’s method']",
"type": "background_claim",
"id": "T165"
},
{
"offsets": [
[
26052,
26159
]
],
"text": "['In Maya, “tweaking” is a procedure adding delta values to original surface vertices before any deformations']",
"type": "background_claim",
"id": "T166"
},
{
"offsets": [
[
26161,
26251
]
],
"text": "['It is actu- ally Maya’s form of rest-pose editing for their built-in deformation operators']",
"type": "background_claim",
"id": "T167"
},
{
"offsets": [
[
26280,
26323
]
],
"text": "['the whole system is divided into two phases']",
"type": "own_claim",
"id": "T168"
},
{
"offsets": [
[
26269,
26277
]
],
"text": "['Figure 6']",
"type": "data",
"id": "T169"
},
{
"offsets": [
[
26325,
26414
]
],
"text": "['The first phase is to find each delta in the rest pose corresponding to each example pose']",
"type": "own_claim",
"id": "T170"
},
{
"offsets": [
[
26570,
26665
]
],
"text": "['the second phase that is a linear system performing RBF interpolation to obtain the PSD weights']",
"type": "own_claim",
"id": "T171"
},
{
"offsets": [
[
27753,
27791
]
],
"text": "['only minimal code changed are required']",
"type": "own_claim",
"id": "T172"
},
{
"offsets": [
[
27596,
27745
]
],
"text": "['The final synthesis is then v x = SKINNING(v r + d x ) + w x where w x is interpolated after SKINNING by applying the same RBF scheme as used for d x']",
"type": "own_claim",
"id": "T173"
},
{
"offsets": [
[
28012,
28045
]
],
"text": "['the SSD transform is nearsingular']",
"type": "data",
"id": "T174"
},
{
"offsets": [
[
28047,
28143
]
],
"text": "['the solved d i can be much large than other d k , which can result in poorly posed interpolation']",
"type": "own_claim",
"id": "T175"
},
{
"offsets": [
[
29881,
29918
]
],
"text": "['the synthesis is potentially realtime']",
"type": "own_claim",
"id": "T176"
},
{
"offsets": [
[
29925,
29971
]
],
"text": "['no extra computing is involved in this process']",
"type": "own_claim",
"id": "T177"
},
{
"offsets": [
[
29852,
29879
]
],
"text": "['the linear system is solved']",
"type": "data",
"id": "T178"
},
{
"offsets": [
[
5433,
5493
]
],
"text": "['only one joint rotation and two example poses are considered']",
"type": "data",
"id": "T179"
},
{
"offsets": [
[
3309,
3338
]
],
"text": "['A nice review of SSD is given']",
"type": "background_claim",
"id": "T180"
},
{
"offsets": [
[
3344,
3345
]
],
"text": "['1']",
"type": "data",
"id": "T181"
},
{
"offsets": [
[
2906,
2924
]
],
"text": "['resculpting models']",
"type": "data",
"id": "T182"
},
{
"offsets": [
[
2928,
2948
]
],
"text": "['re-rigging skeletons']",
"type": "data",
"id": "T183"
}
] | [
{
"id": "R1",
"head": {
"ref_id": "T4",
"role": "Arg1"
},
"tail": {
"ref_id": "T5",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R2",
"head": {
"ref_id": "T7",
"role": "Arg1"
},
"tail": {
"ref_id": "T9",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R3",
"head": {
"ref_id": "T8",
"role": "Arg1"
},
"tail": {
"ref_id": "T9",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R4",
"head": {
"ref_id": "T12",
"role": "Arg1"
},
"tail": {
"ref_id": "T11",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R5",
"head": {
"ref_id": "T11",
"role": "Arg1"
},
"tail": {
"ref_id": "T10",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R6",
"head": {
"ref_id": "T16",
"role": "Arg1"
},
"tail": {
"ref_id": "T17",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R7",
"head": {
"ref_id": "T15",
"role": "Arg1"
},
"tail": {
"ref_id": "T17",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R8",
"head": {
"ref_id": "T17",
"role": "Arg1"
},
"tail": {
"ref_id": "T18",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R9",
"head": {
"ref_id": "T20",
"role": "Arg1"
},
"tail": {
"ref_id": "T21",
"role": "Arg2"
},
"type": "semantically_same"
},
{
"id": "R10",
"head": {
"ref_id": "T23",
"role": "Arg1"
},
"tail": {
"ref_id": "T22",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R11",
"head": {
"ref_id": "T24",
"role": "Arg1"
},
"tail": {
"ref_id": "T22",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R12",
"head": {
"ref_id": "T22",
"role": "Arg1"
},
"tail": {
"ref_id": "T25",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R13",
"head": {
"ref_id": "T26",
"role": "Arg1"
},
"tail": {
"ref_id": "T27",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R14",
"head": {
"ref_id": "T28",
"role": "Arg1"
},
"tail": {
"ref_id": "T27",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R15",
"head": {
"ref_id": "T31",
"role": "Arg1"
},
"tail": {
"ref_id": "T30",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R16",
"head": {
"ref_id": "T30",
"role": "Arg1"
},
"tail": {
"ref_id": "T29",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R17",
"head": {
"ref_id": "T34",
"role": "Arg1"
},
"tail": {
"ref_id": "T32",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R18",
"head": {
"ref_id": "T32",
"role": "Arg1"
},
"tail": {
"ref_id": "T33",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R19",
"head": {
"ref_id": "T33",
"role": "Arg1"
},
"tail": {
"ref_id": "T29",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R20",
"head": {
"ref_id": "T36",
"role": "Arg1"
},
"tail": {
"ref_id": "T35",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R21",
"head": {
"ref_id": "T35",
"role": "Arg1"
},
"tail": {
"ref_id": "T29",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R22",
"head": {
"ref_id": "T39",
"role": "Arg1"
},
"tail": {
"ref_id": "T38",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R23",
"head": {
"ref_id": "T38",
"role": "Arg1"
},
"tail": {
"ref_id": "T29",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R24",
"head": {
"ref_id": "T41",
"role": "Arg1"
},
"tail": {
"ref_id": "T40",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R25",
"head": {
"ref_id": "T42",
"role": "Arg1"
},
"tail": {
"ref_id": "T40",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R26",
"head": {
"ref_id": "T44",
"role": "Arg1"
},
"tail": {
"ref_id": "T45",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R27",
"head": {
"ref_id": "T43",
"role": "Arg1"
},
"tail": {
"ref_id": "T45",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R28",
"head": {
"ref_id": "T47",
"role": "Arg1"
},
"tail": {
"ref_id": "T44",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R29",
"head": {
"ref_id": "T48",
"role": "Arg1"
},
"tail": {
"ref_id": "T44",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R30",
"head": {
"ref_id": "T46",
"role": "Arg1"
},
"tail": {
"ref_id": "T45",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R31",
"head": {
"ref_id": "T49",
"role": "Arg1"
},
"tail": {
"ref_id": "T21",
"role": "Arg2"
},
"type": "semantically_same"
},
{
"id": "R32",
"head": {
"ref_id": "T54",
"role": "Arg1"
},
"tail": {
"ref_id": "T51",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R33",
"head": {
"ref_id": "T53",
"role": "Arg1"
},
"tail": {
"ref_id": "T51",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R34",
"head": {
"ref_id": "T54",
"role": "Arg1"
},
"tail": {
"ref_id": "T52",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R35",
"head": {
"ref_id": "T53",
"role": "Arg1"
},
"tail": {
"ref_id": "T52",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R36",
"head": {
"ref_id": "T56",
"role": "Arg1"
},
"tail": {
"ref_id": "T55",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R37",
"head": {
"ref_id": "T57",
"role": "Arg1"
},
"tail": {
"ref_id": "T58",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R38",
"head": {
"ref_id": "T61",
"role": "Arg1"
},
"tail": {
"ref_id": "T60",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R39",
"head": {
"ref_id": "T60",
"role": "Arg1"
},
"tail": {
"ref_id": "T59",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R40",
"head": {
"ref_id": "T63",
"role": "Arg1"
},
"tail": {
"ref_id": "T62",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R41",
"head": {
"ref_id": "T65",
"role": "Arg1"
},
"tail": {
"ref_id": "T64",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R42",
"head": {
"ref_id": "T66",
"role": "Arg1"
},
"tail": {
"ref_id": "T64",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R43",
"head": {
"ref_id": "T67",
"role": "Arg1"
},
"tail": {
"ref_id": "T64",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R44",
"head": {
"ref_id": "T64",
"role": "Arg1"
},
"tail": {
"ref_id": "T68",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R45",
"head": {
"ref_id": "T69",
"role": "Arg1"
},
"tail": {
"ref_id": "T50",
"role": "Arg2"
},
"type": "semantically_same"
},
{
"id": "R46",
"head": {
"ref_id": "T70",
"role": "Arg1"
},
"tail": {
"ref_id": "T69",
"role": "Arg2"
},
"type": "semantically_same"
},
{
"id": "R47",
"head": {
"ref_id": "T71",
"role": "Arg1"
},
"tail": {
"ref_id": "T70",
"role": "Arg2"
},
"type": "semantically_same"
},
{
"id": "R48",
"head": {
"ref_id": "T50",
"role": "Arg1"
},
"tail": {
"ref_id": "T49",
"role": "Arg2"
},
"type": "semantically_same"
},
{
"id": "R49",
"head": {
"ref_id": "T73",
"role": "Arg1"
},
"tail": {
"ref_id": "T72",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R50",
"head": {
"ref_id": "T74",
"role": "Arg1"
},
"tail": {
"ref_id": "T71",
"role": "Arg2"
},
"type": "semantically_same"
},
{
"id": "R51",
"head": {
"ref_id": "T76",
"role": "Arg1"
},
"tail": {
"ref_id": "T75",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R52",
"head": {
"ref_id": "T75",
"role": "Arg1"
},
"tail": {
"ref_id": "T77",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R53",
"head": {
"ref_id": "T79",
"role": "Arg1"
},
"tail": {
"ref_id": "T78",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R54",
"head": {
"ref_id": "T81",
"role": "Arg1"
},
"tail": {
"ref_id": "T80",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R55",
"head": {
"ref_id": "T81",
"role": "Arg1"
},
"tail": {
"ref_id": "T82",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R56",
"head": {
"ref_id": "T83",
"role": "Arg1"
},
"tail": {
"ref_id": "T84",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R57",
"head": {
"ref_id": "T85",
"role": "Arg1"
},
"tail": {
"ref_id": "T83",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R58",
"head": {
"ref_id": "T86",
"role": "Arg1"
},
"tail": {
"ref_id": "T84",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R59",
"head": {
"ref_id": "T87",
"role": "Arg1"
},
"tail": {
"ref_id": "T84",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R60",
"head": {
"ref_id": "T90",
"role": "Arg1"
},
"tail": {
"ref_id": "T89",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R61",
"head": {
"ref_id": "T91",
"role": "Arg1"
},
"tail": {
"ref_id": "T89",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R62",
"head": {
"ref_id": "T94",
"role": "Arg1"
},
"tail": {
"ref_id": "T93",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R63",
"head": {
"ref_id": "T96",
"role": "Arg1"
},
"tail": {
"ref_id": "T95",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R64",
"head": {
"ref_id": "T98",
"role": "Arg1"
},
"tail": {
"ref_id": "T97",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R65",
"head": {
"ref_id": "T100",
"role": "Arg1"
},
"tail": {
"ref_id": "T99",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R66",
"head": {
"ref_id": "T103",
"role": "Arg1"
},
"tail": {
"ref_id": "T101",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R67",
"head": {
"ref_id": "T99",
"role": "Arg1"
},
"tail": {
"ref_id": "T102",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R68",
"head": {
"ref_id": "T105",
"role": "Arg1"
},
"tail": {
"ref_id": "T104",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R69",
"head": {
"ref_id": "T105",
"role": "Arg1"
},
"tail": {
"ref_id": "T106",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R70",
"head": {
"ref_id": "T109",
"role": "Arg1"
},
"tail": {
"ref_id": "T108",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R71",
"head": {
"ref_id": "T110",
"role": "Arg1"
},
"tail": {
"ref_id": "T108",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R72",
"head": {
"ref_id": "T108",
"role": "Arg1"
},
"tail": {
"ref_id": "T74",
"role": "Arg2"
},
"type": "semantically_same"
},
{
"id": "R73",
"head": {
"ref_id": "T113",
"role": "Arg1"
},
"tail": {
"ref_id": "T112",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R74",
"head": {
"ref_id": "T114",
"role": "Arg1"
},
"tail": {
"ref_id": "T112",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R75",
"head": {
"ref_id": "T115",
"role": "Arg1"
},
"tail": {
"ref_id": "T4",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R76",
"head": {
"ref_id": "T14",
"role": "Arg1"
},
"tail": {
"ref_id": "T15",
"role": "Arg2"
},
"type": "semantically_same"
},
{
"id": "R77",
"head": {
"ref_id": "T116",
"role": "Arg1"
},
"tail": {
"ref_id": "T25",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R78",
"head": {
"ref_id": "T117",
"role": "Arg1"
},
"tail": {
"ref_id": "T3",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R79",
"head": {
"ref_id": "T118",
"role": "Arg1"
},
"tail": {
"ref_id": "T3",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R80",
"head": {
"ref_id": "T9",
"role": "Arg1"
},
"tail": {
"ref_id": "T6",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R81",
"head": {
"ref_id": "T119",
"role": "Arg1"
},
"tail": {
"ref_id": "T120",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R82",
"head": {
"ref_id": "T122",
"role": "Arg1"
},
"tail": {
"ref_id": "T123",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R83",
"head": {
"ref_id": "T124",
"role": "Arg1"
},
"tail": {
"ref_id": "T122",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R84",
"head": {
"ref_id": "T125",
"role": "Arg1"
},
"tail": {
"ref_id": "T122",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R85",
"head": {
"ref_id": "T41",
"role": "Arg1"
},
"tail": {
"ref_id": "T127",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R86",
"head": {
"ref_id": "T42",
"role": "Arg1"
},
"tail": {
"ref_id": "T127",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R87",
"head": {
"ref_id": "T129",
"role": "Arg1"
},
"tail": {
"ref_id": "T128",
"role": "Arg2"
},
"type": "contradicts"
},
{
"id": "R88",
"head": {
"ref_id": "T132",
"role": "Arg1"
},
"tail": {
"ref_id": "T131",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R89",
"head": {
"ref_id": "T133",
"role": "Arg1"
},
"tail": {
"ref_id": "T134",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R90",
"head": {
"ref_id": "T136",
"role": "Arg1"
},
"tail": {
"ref_id": "T135",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R91",
"head": {
"ref_id": "T60",
"role": "Arg1"
},
"tail": {
"ref_id": "T137",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R92",
"head": {
"ref_id": "T139",
"role": "Arg1"
},
"tail": {
"ref_id": "T138",
"role": "Arg2"
},
"type": "parts_of_same"
},
{
"id": "R93",
"head": {
"ref_id": "T138",
"role": "Arg1"
},
"tail": {
"ref_id": "T140",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R94",
"head": {
"ref_id": "T142",
"role": "Arg1"
},
"tail": {
"ref_id": "T141",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R95",
"head": {
"ref_id": "T143",
"role": "Arg1"
},
"tail": {
"ref_id": "T144",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R96",
"head": {
"ref_id": "T145",
"role": "Arg1"
},
"tail": {
"ref_id": "T77",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R97",
"head": {
"ref_id": "T146",
"role": "Arg1"
},
"tail": {
"ref_id": "T147",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R98",
"head": {
"ref_id": "T149",
"role": "Arg1"
},
"tail": {
"ref_id": "T148",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R99",
"head": {
"ref_id": "T150",
"role": "Arg1"
},
"tail": {
"ref_id": "T149",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R100",
"head": {
"ref_id": "T152",
"role": "Arg1"
},
"tail": {
"ref_id": "T151",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R101",
"head": {
"ref_id": "T88",
"role": "Arg1"
},
"tail": {
"ref_id": "T154",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R102",
"head": {
"ref_id": "T155",
"role": "Arg1"
},
"tail": {
"ref_id": "T92",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R103",
"head": {
"ref_id": "T158",
"role": "Arg1"
},
"tail": {
"ref_id": "T157",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R104",
"head": {
"ref_id": "T157",
"role": "Arg1"
},
"tail": {
"ref_id": "T156",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R105",
"head": {
"ref_id": "T161",
"role": "Arg1"
},
"tail": {
"ref_id": "T162",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R106",
"head": {
"ref_id": "T169",
"role": "Arg1"
},
"tail": {
"ref_id": "T168",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R107",
"head": {
"ref_id": "T170",
"role": "Arg1"
},
"tail": {
"ref_id": "T168",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R108",
"head": {
"ref_id": "T171",
"role": "Arg1"
},
"tail": {
"ref_id": "T168",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R109",
"head": {
"ref_id": "T173",
"role": "Arg1"
},
"tail": {
"ref_id": "T172",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R110",
"head": {
"ref_id": "T174",
"role": "Arg1"
},
"tail": {
"ref_id": "T175",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R111",
"head": {
"ref_id": "T177",
"role": "Arg1"
},
"tail": {
"ref_id": "T176",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R112",
"head": {
"ref_id": "T178",
"role": "Arg1"
},
"tail": {
"ref_id": "T176",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R113",
"head": {
"ref_id": "T179",
"role": "Arg1"
},
"tail": {
"ref_id": "T20",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R114",
"head": {
"ref_id": "T181",
"role": "Arg1"
},
"tail": {
"ref_id": "T180",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R115",
"head": {
"ref_id": "T182",
"role": "Arg1"
},
"tail": {
"ref_id": "T4",
"role": "Arg2"
},
"type": "supports"
},
{
"id": "R116",
"head": {
"ref_id": "T183",
"role": "Arg1"
},
"tail": {
"ref_id": "T4",
"role": "Arg2"
},
"type": "supports"
}
] |
A18 | "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n<Document xmlns:gate=\"http://www.gat(...TRUNCATED) | [{"offsets":[[1966,2035]],"text":"['The dream of animating complex living creatures with pure comput(...TRUNCATED) | [{"id":"R1","head":{"ref_id":"T1","role":"Arg1"},"tail":{"ref_id":"T2","role":"Arg2"},"type":"parts_(...TRUNCATED) |
A36 | "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n<Document xmlns:gate=\"http://www.gat(...TRUNCATED) | [{"offsets":[[2177,2309]],"text":"[' Today’s cloth simulators for animation, visual effects, games(...TRUNCATED) | [{"id":"R1","head":{"ref_id":"T6","role":"Arg1"},"tail":{"ref_id":"T5","role":"Arg2"},"type":"suppor(...TRUNCATED) |
A27 | "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n<Document xmlns:gate=\"http://www.gat(...TRUNCATED) | [{"offsets":[[3125,3186]],"text":"['Previous approaches have made great strides towards this goal']"(...TRUNCATED) | [{"id":"R1","head":{"ref_id":"T2","role":"Arg1"},"tail":{"ref_id":"T1","role":"Arg2"},"type":"contra(...TRUNCATED) |
A09 | "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n<Document xmlns:gate=\"http://www.gat(...TRUNCATED) | [{"offsets":[[3101,3203]],"text":"['Significant work has occurred in graphics for deforming articula(...TRUNCATED) | [{"id":"R1","head":{"ref_id":"T2","role":"Arg1"},"tail":{"ref_id":"T1","role":"Arg2"},"type":"suppor(...TRUNCATED) |
A14 | "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n<Document xmlns:gate=\"http://www.gat(...TRUNCATED) | [{"offsets":[[1907,1965]],"text":"['Authoring human motion is difficult for computer animators']","t(...TRUNCATED) | [{"id":"R1","head":{"ref_id":"T2","role":"Arg1"},"tail":{"ref_id":"T1","role":"Arg2"},"type":"suppor(...TRUNCATED) |
A32 | "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n<Document xmlns:gate=\"http://www.gat(...TRUNCATED) | [{"offsets":[[1824,1864]],"text":"['Wrinkles add life to garments in fashion']","type":"background_c(...TRUNCATED) | [{"id":"R1","head":{"ref_id":"T3","role":"Arg1"},"tail":{"ref_id":"T2","role":"Arg2"},"type":"suppor(...TRUNCATED) |
A23 | "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n<Document xmlns:gate=\"http://www.gat(...TRUNCATED) | [{"offsets":[[6973,7125]],"text":"['Computer graphics simulations of fluid behavior are in demand in(...TRUNCATED) | [{"id":"R1","head":{"ref_id":"T7","role":"Arg1"},"tail":{"ref_id":"T6","role":"Arg2"},"type":"contra(...TRUNCATED) |
A05 | "<?xml version=\"1.0\" encoding=\"UTF-8\" standalone=\"no\"?>\n<Document xmlns:gate=\"http://www.gat(...TRUNCATED) | [{"offsets":[[2142,2276]],"text":"['It is possible to accomplish this by having an artist sculpt an (...TRUNCATED) | [{"id":"R1","head":{"ref_id":"T2","role":"Arg1"},"tail":{"ref_id":"T1","role":"Arg2"},"type":"contra(...TRUNCATED) |
End of preview. Expand
in Dataset Viewer.
Dataset Card for "sciarg"
Dataset Summary
The SciArg dataset is an extension of the Dr. Inventor corpus (Fisas et al., 2015, 2016) with an annotation layer containing fine-grained argumentative components and relations. It is the first argument-annotated corpus of scientific publications (in English), which allows for joint analyses of argumentation and other rhetorical dimensions of scientific writing.
Supported Tasks and Leaderboards
Languages
The language in the dataset is English.
Dataset Structure
Data Instances
[More Information Needed]
Data Fields
document_id: the base file name, e.g. "A28"text: the parsed text of the scientific publication in the XML formattext_bound_annotations: span annotations that mark argumentative discourse units (ADUs). Each entry has the following fields:offsets,text,type, andid.relations: binary relation annotations that mark the argumentative relations that hold between a head and a tail ADU. Each entry has the following fields:id,head,tail, andtypewhereheadandtaileach have the fields:ref_idandrole.
Data Splits
The dataset consists of a single train split that has 40 documents.
Dataset Creation
Curation Rationale
[More Information Needed]
Source Data
Initial Data Collection and Normalization
[More Information Needed]
Who are the source language producers?
[More Information Needed]
Annotations
Annotation process
[More Information Needed]
Who are the annotators?
[More Information Needed]
Personal and Sensitive Information
[More Information Needed]
Considerations for Using the Data
Social Impact of Dataset
[More Information Needed]
Discussion of Biases
[More Information Needed]
Other Known Limitations
[More Information Needed]
Additional Information
Dataset Curators
[More Information Needed]
Licensing Information
[More Information Needed]
Citation Information
@inproceedings{lauscher2018b,
title = {An argument-annotated corpus of scientific publications},
booktitle = {Proceedings of the 5th Workshop on Mining Argumentation},
publisher = {Association for Computational Linguistics},
author = {Lauscher, Anne and Glava\v{s}, Goran and Ponzetto, Simone Paolo},
address = {Brussels, Belgium},
year = {2018},
pages = {40–46}
}
Contributions
Thanks to @github-username for adding this dataset.
- Downloads last month
- 54
Homepage:
github.com
Repository:
github.com
Paper:
An argument-annotated corpus of scientific publications
Size of downloaded dataset files:
1.13 MB
Size of the auto-converted Parquet files:
1.65 MB
Number of rows:
40
Models trained or fine-tuned on DFKI-SLT/sciarg