SparseCraft

[ECCV'24] SparseCraft: Few-Shot Neural Reconstruction through Stereopsis Guided Geometric Linearization

Project | Paper | Code

DTU Dataset

We provide preprocessed DTU data and results for the tasks of novel view synthesis and surface reconstruction.

It contains the following directories:

sparsecraft_data
├── nvs # Novel View Synthesis task data and results
│   └── mvs_data
│       ├── scan103
│       ├── ...
│   └── results # Results for training using 3, 6, and 9 views
│       ├── 3v
│       │   ├── scan103
│       │   ├── ...
│       ├── 6v
│       │   ├── scan103
│       │   ├── ...
│       └── 9v
│           ├── scan103
│           ├── ...
└── reconstruction # Surface Reconstruction task data and results
    └── mvs_data # Surface reconstruction data uses a different set of scans and views than the novel view synthesis task
        ├── set0
        │   ├── scan105
        │   ├── ...
        └── set1
            ├── scan105
            ├── ...
    └── results
        ├── set0
        │   ├── scan105
        │   ├── ...
        └── set1
            ├── scan105

Note

The DTU dataset was preprocessed as follows:

The original data is from the NeuS Project. We use the same camera poses and intrinsics as the original data.
To obtain MVS data, we used the Colmap initialized with the original camera poses and intrinsics.
We provide a script that achieves this in scripts that you can run using the following command. Note that you will need to have Colmap installed on your machine: