Koke_Cacao
commited on
Commit
•
74db9af
1
Parent(s):
0974835
:sparkles: clean up code
Browse files- .style.yapf +6 -0
- scripts/README.md → README.md +6 -2
- scripts/attention.py +40 -101
- scripts/convert_mvdream_to_diffusers.py +37 -68
- scripts/models.py +123 -228
- scripts/pipeline_mvdream.py +97 -291
- vae/diffusion_pytorch_model.bin +1 -1
.style.yapf
ADDED
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[style]
|
2 |
+
based_on_style = google
|
3 |
+
spaces_before_comment = 1
|
4 |
+
indent_width: 4
|
5 |
+
split_before_logical_operator = true
|
6 |
+
column_limit = 1024
|
scripts/README.md → README.md
RENAMED
@@ -1,4 +1,8 @@
|
|
1 |
-
#
|
|
|
|
|
|
|
|
|
2 |
|
3 |
Download original MVDream checkpoint through one of the following sources:
|
4 |
|
@@ -14,5 +18,5 @@ wget https://raw.githubusercontent.com/bytedance/MVDream/main/mvdream/configs/sd
|
|
14 |
|
15 |
Hugging Face diffusers weights are converted by script:
|
16 |
```bash
|
17 |
-
python ./scripts/convert_mvdream_to_diffusers.py --checkpoint_path ./sd-v1.5-4view.pt --dump_path . --original_config_file ./sd-v1.yaml
|
18 |
```
|
|
|
1 |
+
# MVDream-HF
|
2 |
+
|
3 |
+
A huggingface implementation of MVDream, used for quick one-line download. See [huggingface repo](https://huggingface.co/KokeCacao/mvdream-hf/tree/main) that hosts sd-v1.5 version.
|
4 |
+
|
5 |
+
## Convert Original Weights to Diffusers
|
6 |
|
7 |
Download original MVDream checkpoint through one of the following sources:
|
8 |
|
|
|
18 |
|
19 |
Hugging Face diffusers weights are converted by script:
|
20 |
```bash
|
21 |
+
python ./scripts/convert_mvdream_to_diffusers.py --checkpoint_path ./sd-v1.5-4view.pt --dump_path . --original_config_file ./sd-v1.yaml --test
|
22 |
```
|
scripts/attention.py
CHANGED
@@ -11,7 +11,6 @@ from einops import rearrange, repeat
|
|
11 |
from typing import Optional, Any
|
12 |
from util import checkpoint
|
13 |
|
14 |
-
|
15 |
try:
|
16 |
import xformers
|
17 |
import xformers.ops
|
@@ -21,11 +20,12 @@ except:
|
|
21 |
|
22 |
# CrossAttn precision handling
|
23 |
import os
|
|
|
24 |
_ATTN_PRECISION = os.environ.get("ATTN_PRECISION", "fp32")
|
25 |
|
26 |
|
27 |
def uniq(arr):
|
28 |
-
return{el: True for el in arr}.keys()
|
29 |
|
30 |
|
31 |
def default(val, d):
|
@@ -47,6 +47,7 @@ def init_(tensor):
|
|
47 |
|
48 |
# feedforward
|
49 |
class GEGLU(nn.Module):
|
|
|
50 |
def __init__(self, dim_in, dim_out):
|
51 |
super().__init__()
|
52 |
self.proj = nn.Linear(dim_in, dim_out * 2)
|
@@ -57,20 +58,14 @@ class GEGLU(nn.Module):
|
|
57 |
|
58 |
|
59 |
class FeedForward(nn.Module):
|
|
|
60 |
def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.):
|
61 |
super().__init__()
|
62 |
inner_dim = int(dim * mult)
|
63 |
dim_out = default(dim_out, dim)
|
64 |
-
project_in = nn.Sequential(
|
65 |
-
|
66 |
-
|
67 |
-
) if not glu else GEGLU(dim, inner_dim)
|
68 |
-
|
69 |
-
self.net = nn.Sequential(
|
70 |
-
project_in,
|
71 |
-
nn.Dropout(dropout),
|
72 |
-
nn.Linear(inner_dim, dim_out)
|
73 |
-
)
|
74 |
|
75 |
def forward(self, x):
|
76 |
return self.net(x)
|
@@ -90,31 +85,16 @@ def Normalize(in_channels):
|
|
90 |
|
91 |
|
92 |
class SpatialSelfAttention(nn.Module):
|
|
|
93 |
def __init__(self, in_channels):
|
94 |
super().__init__()
|
95 |
self.in_channels = in_channels
|
96 |
|
97 |
self.norm = Normalize(in_channels)
|
98 |
-
self.q = torch.nn.Conv2d(in_channels,
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
padding=0)
|
103 |
-
self.k = torch.nn.Conv2d(in_channels,
|
104 |
-
in_channels,
|
105 |
-
kernel_size=1,
|
106 |
-
stride=1,
|
107 |
-
padding=0)
|
108 |
-
self.v = torch.nn.Conv2d(in_channels,
|
109 |
-
in_channels,
|
110 |
-
kernel_size=1,
|
111 |
-
stride=1,
|
112 |
-
padding=0)
|
113 |
-
self.proj_out = torch.nn.Conv2d(in_channels,
|
114 |
-
in_channels,
|
115 |
-
kernel_size=1,
|
116 |
-
stride=1,
|
117 |
-
padding=0)
|
118 |
|
119 |
def forward(self, x):
|
120 |
h_ = x
|
@@ -124,7 +104,7 @@ class SpatialSelfAttention(nn.Module):
|
|
124 |
v = self.v(h_)
|
125 |
|
126 |
# compute attention
|
127 |
-
b,c,h,w = q.shape
|
128 |
q = rearrange(q, 'b c h w -> b (h w) c')
|
129 |
k = rearrange(k, 'b c h w -> b c (h w)')
|
130 |
w_ = torch.einsum('bij,bjk->bik', q, k)
|
@@ -139,26 +119,24 @@ class SpatialSelfAttention(nn.Module):
|
|
139 |
h_ = rearrange(h_, 'b c (h w) -> b c h w', h=h)
|
140 |
h_ = self.proj_out(h_)
|
141 |
|
142 |
-
return x+h_
|
143 |
|
144 |
|
145 |
class CrossAttention(nn.Module):
|
|
|
146 |
def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0.):
|
147 |
super().__init__()
|
148 |
inner_dim = dim_head * heads
|
149 |
context_dim = default(context_dim, query_dim)
|
150 |
|
151 |
-
self.scale = dim_head
|
152 |
self.heads = heads
|
153 |
|
154 |
self.to_q = nn.Linear(query_dim, inner_dim, bias=False)
|
155 |
self.to_k = nn.Linear(context_dim, inner_dim, bias=False)
|
156 |
self.to_v = nn.Linear(context_dim, inner_dim, bias=False)
|
157 |
|
158 |
-
self.to_out = nn.Sequential(
|
159 |
-
nn.Linear(inner_dim, query_dim),
|
160 |
-
nn.Dropout(dropout)
|
161 |
-
)
|
162 |
|
163 |
def forward(self, x, context=None, mask=None):
|
164 |
h = self.heads
|
@@ -171,15 +149,15 @@ class CrossAttention(nn.Module):
|
|
171 |
q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
|
172 |
|
173 |
# force cast to fp32 to avoid overflowing
|
174 |
-
if _ATTN_PRECISION =="fp32":
|
175 |
-
with autocast(enabled=False, device_type
|
176 |
q, k = q.float(), k.float()
|
177 |
sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
|
178 |
else:
|
179 |
sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
|
180 |
-
|
181 |
del q, k
|
182 |
-
|
183 |
if mask is not None:
|
184 |
mask = rearrange(mask, 'b ... -> b (...)')
|
185 |
max_neg_value = -torch.finfo(sim.dtype).max
|
@@ -221,11 +199,7 @@ class MemoryEfficientCrossAttention(nn.Module):
|
|
221 |
|
222 |
b, _, _ = q.shape
|
223 |
q, k, v = map(
|
224 |
-
lambda t: t.unsqueeze(3)
|
225 |
-
.reshape(b, t.shape[1], self.heads, self.dim_head)
|
226 |
-
.permute(0, 2, 1, 3)
|
227 |
-
.reshape(b * self.heads, t.shape[1], self.dim_head)
|
228 |
-
.contiguous(),
|
229 |
(q, k, v),
|
230 |
)
|
231 |
|
@@ -234,32 +208,25 @@ class MemoryEfficientCrossAttention(nn.Module):
|
|
234 |
|
235 |
if mask is not None:
|
236 |
raise NotImplementedError
|
237 |
-
out = (
|
238 |
-
out.unsqueeze(0)
|
239 |
-
.reshape(b, self.heads, out.shape[1], self.dim_head)
|
240 |
-
.permute(0, 2, 1, 3)
|
241 |
-
.reshape(b, out.shape[1], self.heads * self.dim_head)
|
242 |
-
)
|
243 |
return self.to_out(out)
|
244 |
|
245 |
|
246 |
class BasicTransformerBlock(nn.Module):
|
247 |
ATTENTION_MODES = {
|
248 |
-
"softmax": CrossAttention,
|
249 |
"softmax-xformers": MemoryEfficientCrossAttention
|
250 |
}
|
251 |
-
|
252 |
-
|
253 |
super().__init__()
|
254 |
attn_mode = "softmax-xformers" if XFORMERS_IS_AVAILBLE else "softmax"
|
255 |
assert attn_mode in self.ATTENTION_MODES
|
256 |
attn_cls = self.ATTENTION_MODES[attn_mode]
|
257 |
self.disable_self_attn = disable_self_attn
|
258 |
-
self.attn1 = attn_cls(query_dim=dim, heads=n_heads, dim_head=d_head, dropout=dropout,
|
259 |
-
context_dim=context_dim if self.disable_self_attn else None) # is a self-attention if not self.disable_self_attn
|
260 |
self.ff = FeedForward(dim, dropout=dropout, glu=gated_ff)
|
261 |
-
self.attn2 = attn_cls(query_dim=dim, context_dim=context_dim,
|
262 |
-
heads=n_heads, dim_head=d_head, dropout=dropout) # is self-attn if context is none
|
263 |
self.norm1 = nn.LayerNorm(dim)
|
264 |
self.norm2 = nn.LayerNorm(dim)
|
265 |
self.norm3 = nn.LayerNorm(dim)
|
@@ -284,10 +251,8 @@ class SpatialTransformer(nn.Module):
|
|
284 |
Finally, reshape to image
|
285 |
NEW: use_linear for more efficiency instead of the 1x1 convs
|
286 |
"""
|
287 |
-
|
288 |
-
|
289 |
-
disable_self_attn=False, use_linear=False,
|
290 |
-
use_checkpoint=True):
|
291 |
super().__init__()
|
292 |
assert context_dim is not None
|
293 |
if not isinstance(context_dim, list):
|
@@ -296,25 +261,13 @@ class SpatialTransformer(nn.Module):
|
|
296 |
inner_dim = n_heads * d_head
|
297 |
self.norm = Normalize(in_channels)
|
298 |
if not use_linear:
|
299 |
-
self.proj_in = nn.Conv2d(in_channels,
|
300 |
-
inner_dim,
|
301 |
-
kernel_size=1,
|
302 |
-
stride=1,
|
303 |
-
padding=0)
|
304 |
else:
|
305 |
self.proj_in = nn.Linear(in_channels, inner_dim)
|
306 |
|
307 |
-
self.transformer_blocks = nn.ModuleList(
|
308 |
-
[BasicTransformerBlock(inner_dim, n_heads, d_head, dropout=dropout, context_dim=context_dim[d],
|
309 |
-
disable_self_attn=disable_self_attn, checkpoint=use_checkpoint)
|
310 |
-
for d in range(depth)]
|
311 |
-
)
|
312 |
if not use_linear:
|
313 |
-
self.proj_out = zero_module(nn.Conv2d(inner_dim,
|
314 |
-
in_channels,
|
315 |
-
kernel_size=1,
|
316 |
-
stride=1,
|
317 |
-
padding=0))
|
318 |
else:
|
319 |
self.proj_out = zero_module(nn.Linear(in_channels, inner_dim))
|
320 |
self.use_linear = use_linear
|
@@ -356,11 +309,9 @@ class BasicTransformerBlock3D(BasicTransformerBlock):
|
|
356 |
|
357 |
|
358 |
class SpatialTransformer3D(nn.Module):
|
359 |
-
''' 3D self-attention '''
|
360 |
-
|
361 |
-
|
362 |
-
disable_self_attn=False, use_linear=False,
|
363 |
-
use_checkpoint=True):
|
364 |
super().__init__()
|
365 |
assert context_dim is not None
|
366 |
if not isinstance(context_dim, list):
|
@@ -369,25 +320,13 @@ class SpatialTransformer3D(nn.Module):
|
|
369 |
inner_dim = n_heads * d_head
|
370 |
self.norm = Normalize(in_channels)
|
371 |
if not use_linear:
|
372 |
-
self.proj_in = nn.Conv2d(in_channels,
|
373 |
-
inner_dim,
|
374 |
-
kernel_size=1,
|
375 |
-
stride=1,
|
376 |
-
padding=0)
|
377 |
else:
|
378 |
self.proj_in = nn.Linear(in_channels, inner_dim)
|
379 |
|
380 |
-
self.transformer_blocks = nn.ModuleList(
|
381 |
-
[BasicTransformerBlock3D(inner_dim, n_heads, d_head, dropout=dropout, context_dim=context_dim[d],
|
382 |
-
disable_self_attn=disable_self_attn, checkpoint=use_checkpoint)
|
383 |
-
for d in range(depth)]
|
384 |
-
)
|
385 |
if not use_linear:
|
386 |
-
self.proj_out = zero_module(nn.Conv2d(inner_dim,
|
387 |
-
in_channels,
|
388 |
-
kernel_size=1,
|
389 |
-
stride=1,
|
390 |
-
padding=0))
|
391 |
else:
|
392 |
self.proj_out = zero_module(nn.Linear(in_channels, inner_dim))
|
393 |
self.use_linear = use_linear
|
@@ -411,4 +350,4 @@ class SpatialTransformer3D(nn.Module):
|
|
411 |
x = rearrange(x, 'b (h w) c -> b c h w', h=h, w=w).contiguous()
|
412 |
if not self.use_linear:
|
413 |
x = self.proj_out(x)
|
414 |
-
return x + x_in
|
|
|
11 |
from typing import Optional, Any
|
12 |
from util import checkpoint
|
13 |
|
|
|
14 |
try:
|
15 |
import xformers
|
16 |
import xformers.ops
|
|
|
20 |
|
21 |
# CrossAttn precision handling
|
22 |
import os
|
23 |
+
|
24 |
_ATTN_PRECISION = os.environ.get("ATTN_PRECISION", "fp32")
|
25 |
|
26 |
|
27 |
def uniq(arr):
|
28 |
+
return {el: True for el in arr}.keys()
|
29 |
|
30 |
|
31 |
def default(val, d):
|
|
|
47 |
|
48 |
# feedforward
|
49 |
class GEGLU(nn.Module):
|
50 |
+
|
51 |
def __init__(self, dim_in, dim_out):
|
52 |
super().__init__()
|
53 |
self.proj = nn.Linear(dim_in, dim_out * 2)
|
|
|
58 |
|
59 |
|
60 |
class FeedForward(nn.Module):
|
61 |
+
|
62 |
def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.):
|
63 |
super().__init__()
|
64 |
inner_dim = int(dim * mult)
|
65 |
dim_out = default(dim_out, dim)
|
66 |
+
project_in = nn.Sequential(nn.Linear(dim, inner_dim), nn.GELU()) if not glu else GEGLU(dim, inner_dim)
|
67 |
+
|
68 |
+
self.net = nn.Sequential(project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
69 |
|
70 |
def forward(self, x):
|
71 |
return self.net(x)
|
|
|
85 |
|
86 |
|
87 |
class SpatialSelfAttention(nn.Module):
|
88 |
+
|
89 |
def __init__(self, in_channels):
|
90 |
super().__init__()
|
91 |
self.in_channels = in_channels
|
92 |
|
93 |
self.norm = Normalize(in_channels)
|
94 |
+
self.q = torch.nn.Conv2d(in_channels, in_channels, kernel_size=1, stride=1, padding=0)
|
95 |
+
self.k = torch.nn.Conv2d(in_channels, in_channels, kernel_size=1, stride=1, padding=0)
|
96 |
+
self.v = torch.nn.Conv2d(in_channels, in_channels, kernel_size=1, stride=1, padding=0)
|
97 |
+
self.proj_out = torch.nn.Conv2d(in_channels, in_channels, kernel_size=1, stride=1, padding=0)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
|
99 |
def forward(self, x):
|
100 |
h_ = x
|
|
|
104 |
v = self.v(h_)
|
105 |
|
106 |
# compute attention
|
107 |
+
b, c, h, w = q.shape
|
108 |
q = rearrange(q, 'b c h w -> b (h w) c')
|
109 |
k = rearrange(k, 'b c h w -> b c (h w)')
|
110 |
w_ = torch.einsum('bij,bjk->bik', q, k)
|
|
|
119 |
h_ = rearrange(h_, 'b c (h w) -> b c h w', h=h)
|
120 |
h_ = self.proj_out(h_)
|
121 |
|
122 |
+
return x + h_
|
123 |
|
124 |
|
125 |
class CrossAttention(nn.Module):
|
126 |
+
|
127 |
def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0.):
|
128 |
super().__init__()
|
129 |
inner_dim = dim_head * heads
|
130 |
context_dim = default(context_dim, query_dim)
|
131 |
|
132 |
+
self.scale = dim_head**-0.5
|
133 |
self.heads = heads
|
134 |
|
135 |
self.to_q = nn.Linear(query_dim, inner_dim, bias=False)
|
136 |
self.to_k = nn.Linear(context_dim, inner_dim, bias=False)
|
137 |
self.to_v = nn.Linear(context_dim, inner_dim, bias=False)
|
138 |
|
139 |
+
self.to_out = nn.Sequential(nn.Linear(inner_dim, query_dim), nn.Dropout(dropout))
|
|
|
|
|
|
|
140 |
|
141 |
def forward(self, x, context=None, mask=None):
|
142 |
h = self.heads
|
|
|
149 |
q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
|
150 |
|
151 |
# force cast to fp32 to avoid overflowing
|
152 |
+
if _ATTN_PRECISION == "fp32":
|
153 |
+
with autocast(enabled=False, device_type='cuda'):
|
154 |
q, k = q.float(), k.float()
|
155 |
sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
|
156 |
else:
|
157 |
sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
|
158 |
+
|
159 |
del q, k
|
160 |
+
|
161 |
if mask is not None:
|
162 |
mask = rearrange(mask, 'b ... -> b (...)')
|
163 |
max_neg_value = -torch.finfo(sim.dtype).max
|
|
|
199 |
|
200 |
b, _, _ = q.shape
|
201 |
q, k, v = map(
|
202 |
+
lambda t: t.unsqueeze(3).reshape(b, t.shape[1], self.heads, self.dim_head).permute(0, 2, 1, 3).reshape(b * self.heads, t.shape[1], self.dim_head).contiguous(),
|
|
|
|
|
|
|
|
|
203 |
(q, k, v),
|
204 |
)
|
205 |
|
|
|
208 |
|
209 |
if mask is not None:
|
210 |
raise NotImplementedError
|
211 |
+
out = (out.unsqueeze(0).reshape(b, self.heads, out.shape[1], self.dim_head).permute(0, 2, 1, 3).reshape(b, out.shape[1], self.heads * self.dim_head))
|
|
|
|
|
|
|
|
|
|
|
212 |
return self.to_out(out)
|
213 |
|
214 |
|
215 |
class BasicTransformerBlock(nn.Module):
|
216 |
ATTENTION_MODES = {
|
217 |
+
"softmax": CrossAttention, # vanilla attention
|
218 |
"softmax-xformers": MemoryEfficientCrossAttention
|
219 |
}
|
220 |
+
|
221 |
+
def __init__(self, dim, n_heads, d_head, dropout=0., context_dim=None, gated_ff=True, checkpoint=True, disable_self_attn=False):
|
222 |
super().__init__()
|
223 |
attn_mode = "softmax-xformers" if XFORMERS_IS_AVAILBLE else "softmax"
|
224 |
assert attn_mode in self.ATTENTION_MODES
|
225 |
attn_cls = self.ATTENTION_MODES[attn_mode]
|
226 |
self.disable_self_attn = disable_self_attn
|
227 |
+
self.attn1 = attn_cls(query_dim=dim, heads=n_heads, dim_head=d_head, dropout=dropout, context_dim=context_dim if self.disable_self_attn else None) # is a self-attention if not self.disable_self_attn
|
|
|
228 |
self.ff = FeedForward(dim, dropout=dropout, glu=gated_ff)
|
229 |
+
self.attn2 = attn_cls(query_dim=dim, context_dim=context_dim, heads=n_heads, dim_head=d_head, dropout=dropout) # is self-attn if context is none
|
|
|
230 |
self.norm1 = nn.LayerNorm(dim)
|
231 |
self.norm2 = nn.LayerNorm(dim)
|
232 |
self.norm3 = nn.LayerNorm(dim)
|
|
|
251 |
Finally, reshape to image
|
252 |
NEW: use_linear for more efficiency instead of the 1x1 convs
|
253 |
"""
|
254 |
+
|
255 |
+
def __init__(self, in_channels, n_heads, d_head, depth=1, dropout=0., context_dim=None, disable_self_attn=False, use_linear=False, use_checkpoint=True):
|
|
|
|
|
256 |
super().__init__()
|
257 |
assert context_dim is not None
|
258 |
if not isinstance(context_dim, list):
|
|
|
261 |
inner_dim = n_heads * d_head
|
262 |
self.norm = Normalize(in_channels)
|
263 |
if not use_linear:
|
264 |
+
self.proj_in = nn.Conv2d(in_channels, inner_dim, kernel_size=1, stride=1, padding=0)
|
|
|
|
|
|
|
|
|
265 |
else:
|
266 |
self.proj_in = nn.Linear(in_channels, inner_dim)
|
267 |
|
268 |
+
self.transformer_blocks = nn.ModuleList([BasicTransformerBlock(inner_dim, n_heads, d_head, dropout=dropout, context_dim=context_dim[d], disable_self_attn=disable_self_attn, checkpoint=use_checkpoint) for d in range(depth)])
|
|
|
|
|
|
|
|
|
269 |
if not use_linear:
|
270 |
+
self.proj_out = zero_module(nn.Conv2d(inner_dim, in_channels, kernel_size=1, stride=1, padding=0))
|
|
|
|
|
|
|
|
|
271 |
else:
|
272 |
self.proj_out = zero_module(nn.Linear(in_channels, inner_dim))
|
273 |
self.use_linear = use_linear
|
|
|
309 |
|
310 |
|
311 |
class SpatialTransformer3D(nn.Module):
|
312 |
+
''' 3D self-attention '''
|
313 |
+
|
314 |
+
def __init__(self, in_channels, n_heads, d_head, depth=1, dropout=0., context_dim=None, disable_self_attn=False, use_linear=False, use_checkpoint=True):
|
|
|
|
|
315 |
super().__init__()
|
316 |
assert context_dim is not None
|
317 |
if not isinstance(context_dim, list):
|
|
|
320 |
inner_dim = n_heads * d_head
|
321 |
self.norm = Normalize(in_channels)
|
322 |
if not use_linear:
|
323 |
+
self.proj_in = nn.Conv2d(in_channels, inner_dim, kernel_size=1, stride=1, padding=0)
|
|
|
|
|
|
|
|
|
324 |
else:
|
325 |
self.proj_in = nn.Linear(in_channels, inner_dim)
|
326 |
|
327 |
+
self.transformer_blocks = nn.ModuleList([BasicTransformerBlock3D(inner_dim, n_heads, d_head, dropout=dropout, context_dim=context_dim[d], disable_self_attn=disable_self_attn, checkpoint=use_checkpoint) for d in range(depth)])
|
|
|
|
|
|
|
|
|
328 |
if not use_linear:
|
329 |
+
self.proj_out = zero_module(nn.Conv2d(inner_dim, in_channels, kernel_size=1, stride=1, padding=0))
|
|
|
|
|
|
|
|
|
330 |
else:
|
331 |
self.proj_out = zero_module(nn.Linear(in_channels, inner_dim))
|
332 |
self.use_linear = use_linear
|
|
|
350 |
x = rearrange(x, 'b (h w) c -> b c h w', h=h, w=w).contiguous()
|
351 |
if not self.use_linear:
|
352 |
x = self.proj_out(x)
|
353 |
+
return x + x_in
|
scripts/convert_mvdream_to_diffusers.py
CHANGED
@@ -3,6 +3,7 @@
|
|
3 |
import argparse
|
4 |
import torch
|
5 |
import sys
|
|
|
6 |
sys.path.insert(0, '../')
|
7 |
|
8 |
from transformers import (
|
@@ -126,9 +127,7 @@ logger = logging.get_logger(__name__)
|
|
126 |
# return config
|
127 |
|
128 |
|
129 |
-
def assign_to_checkpoint(
|
130 |
-
paths, checkpoint, old_checkpoint, attention_paths_to_split=None, additional_replacements=None, config=None
|
131 |
-
):
|
132 |
"""
|
133 |
This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits
|
134 |
attention layers, and takes into account additional replacements that may arise.
|
@@ -144,6 +143,7 @@ def assign_to_checkpoint(
|
|
144 |
|
145 |
target_shape = (-1, channels) if len(old_tensor.shape) == 3 else (-1)
|
146 |
|
|
|
147 |
num_heads = old_tensor.shape[0] // config["num_head_channels"] // 3
|
148 |
|
149 |
old_tensor = old_tensor.reshape((num_heads, 3 * channels // num_heads) + old_tensor.shape[1:])
|
@@ -211,6 +211,7 @@ def renew_resnet_paths(old_list, n_shave_prefix_segments=0):
|
|
211 |
|
212 |
return mapping
|
213 |
|
|
|
214 |
def renew_attention_paths(old_list, n_shave_prefix_segments=0):
|
215 |
"""
|
216 |
Updates paths inside attentions to the new naming scheme (local renaming)
|
@@ -231,6 +232,7 @@ def renew_attention_paths(old_list, n_shave_prefix_segments=0):
|
|
231 |
|
232 |
return mapping
|
233 |
|
|
|
234 |
# def convert_ldm_unet_checkpoint(
|
235 |
# checkpoint, config, path=None, extract_ema=False, controlnet=False, skip_extract_state_dict=False
|
236 |
# ):
|
@@ -496,6 +498,7 @@ def create_vae_diffusers_config(original_config, image_size: int):
|
|
496 |
}
|
497 |
return config
|
498 |
|
|
|
499 |
def convert_ldm_vae_checkpoint(checkpoint, config):
|
500 |
# extract state dict for VAE
|
501 |
vae_state_dict = {}
|
@@ -528,26 +531,18 @@ def convert_ldm_vae_checkpoint(checkpoint, config):
|
|
528 |
|
529 |
# Retrieves the keys for the encoder down blocks only
|
530 |
num_down_blocks = len({".".join(layer.split(".")[:3]) for layer in vae_state_dict if "encoder.down" in layer})
|
531 |
-
down_blocks = {
|
532 |
-
layer_id: [key for key in vae_state_dict if f"down.{layer_id}" in key] for layer_id in range(num_down_blocks)
|
533 |
-
}
|
534 |
|
535 |
# Retrieves the keys for the decoder up blocks only
|
536 |
num_up_blocks = len({".".join(layer.split(".")[:3]) for layer in vae_state_dict if "decoder.up" in layer})
|
537 |
-
up_blocks = {
|
538 |
-
layer_id: [key for key in vae_state_dict if f"up.{layer_id}" in key] for layer_id in range(num_up_blocks)
|
539 |
-
}
|
540 |
|
541 |
for i in range(num_down_blocks):
|
542 |
resnets = [key for key in down_blocks[i] if f"down.{i}" in key and f"down.{i}.downsample" not in key]
|
543 |
|
544 |
if f"encoder.down.{i}.downsample.conv.weight" in vae_state_dict:
|
545 |
-
new_checkpoint[f"encoder.down_blocks.{i}.downsamplers.0.conv.weight"] = vae_state_dict.pop(
|
546 |
-
|
547 |
-
)
|
548 |
-
new_checkpoint[f"encoder.down_blocks.{i}.downsamplers.0.conv.bias"] = vae_state_dict.pop(
|
549 |
-
f"encoder.down.{i}.downsample.conv.bias"
|
550 |
-
)
|
551 |
|
552 |
paths = renew_vae_resnet_paths(resnets)
|
553 |
meta_path = {"old": f"down.{i}.block", "new": f"down_blocks.{i}.resnets"}
|
@@ -570,17 +565,11 @@ def convert_ldm_vae_checkpoint(checkpoint, config):
|
|
570 |
|
571 |
for i in range(num_up_blocks):
|
572 |
block_id = num_up_blocks - 1 - i
|
573 |
-
resnets = [
|
574 |
-
key for key in up_blocks[block_id] if f"up.{block_id}" in key and f"up.{block_id}.upsample" not in key
|
575 |
-
]
|
576 |
|
577 |
if f"decoder.up.{block_id}.upsample.conv.weight" in vae_state_dict:
|
578 |
-
new_checkpoint[f"decoder.up_blocks.{i}.upsamplers.0.conv.weight"] = vae_state_dict[
|
579 |
-
|
580 |
-
]
|
581 |
-
new_checkpoint[f"decoder.up_blocks.{i}.upsamplers.0.conv.bias"] = vae_state_dict[
|
582 |
-
f"decoder.up.{block_id}.upsample.conv.bias"
|
583 |
-
]
|
584 |
|
585 |
paths = renew_vae_resnet_paths(resnets)
|
586 |
meta_path = {"old": f"up.{block_id}.block", "new": f"up_blocks.{i}.resnets"}
|
@@ -618,6 +607,7 @@ def renew_vae_resnet_paths(old_list, n_shave_prefix_segments=0):
|
|
618 |
|
619 |
return mapping
|
620 |
|
|
|
621 |
def renew_vae_attention_paths(old_list, n_shave_prefix_segments=0):
|
622 |
"""
|
623 |
Updates paths inside attentions to the new naming scheme (local renaming)
|
@@ -659,12 +649,8 @@ def conv_attn_to_linear(checkpoint):
|
|
659 |
if checkpoint[key].ndim > 2:
|
660 |
checkpoint[key] = checkpoint[key][:, :, 0]
|
661 |
|
662 |
-
|
663 |
-
|
664 |
-
original_config_file,
|
665 |
-
extract_ema,
|
666 |
-
device
|
667 |
-
):
|
668 |
checkpoint = torch.load(checkpoint_path, map_location=device)
|
669 |
# print(f"Checkpoint: {checkpoint.keys()}")
|
670 |
torch.cuda.empty_cache()
|
@@ -702,9 +688,7 @@ def convert_from_original_mvdream_ckpt(
|
|
702 |
# print(f"Unet Config: {original_config.model.params.unet_config.params}")
|
703 |
unet: MultiViewUNetWrapperModel = MultiViewUNetWrapperModel(**original_config.model.params.unet_config.params)
|
704 |
# print(f"Unet State Dict: {unet.state_dict().keys()}")
|
705 |
-
unet.load_state_dict({
|
706 |
-
key.replace("model.diffusion_model.", "unet."): value for key, value in checkpoint.items() if key.replace("model.diffusion_model.", "unet.") in unet.state_dict()
|
707 |
-
})
|
708 |
for param_name, param in unet.state_dict().items():
|
709 |
set_module_tensor_to_device(unet, param_name, "cuda:0", value=param)
|
710 |
|
@@ -712,25 +696,21 @@ def convert_from_original_mvdream_ckpt(
|
|
712 |
vae_config = create_vae_diffusers_config(original_config, image_size=image_size)
|
713 |
converted_vae_checkpoint = convert_ldm_vae_checkpoint(checkpoint, vae_config)
|
714 |
|
715 |
-
if (
|
716 |
-
"model" in original_config
|
717 |
-
and "params" in original_config.model
|
718 |
-
and "scale_factor" in original_config.model.params
|
719 |
-
):
|
720 |
vae_scaling_factor = original_config.model.params.scale_factor
|
721 |
else:
|
722 |
-
vae_scaling_factor = 0.18215
|
723 |
|
724 |
vae_config["scaling_factor"] = vae_scaling_factor
|
725 |
|
726 |
with init_empty_weights():
|
727 |
vae = AutoencoderKL(**vae_config)
|
728 |
-
|
729 |
tokenizer: CLIPTokenizer = CLIPTokenizer.from_pretrained("openai/clip-vit-large-patch14")
|
730 |
text_encoder: CLIPTextModel = CLIPTextModel.from_pretrained("openai/clip-vit-large-patch14").to(device=torch.device("cuda:0")) # type: ignore
|
731 |
|
732 |
for param_name, param in converted_vae_checkpoint.items():
|
733 |
-
set_module_tensor_to_device(vae, param_name, "cuda:0", value=param)
|
734 |
|
735 |
pipe = MVDreamStableDiffusionPipeline(
|
736 |
vae=vae,
|
@@ -746,30 +726,20 @@ def convert_from_original_mvdream_ckpt(
|
|
746 |
if __name__ == "__main__":
|
747 |
parser = argparse.ArgumentParser()
|
748 |
|
749 |
-
parser.add_argument(
|
750 |
-
"--checkpoint_path", default=None, type=str, required=True, help="Path to the checkpoint to convert."
|
751 |
-
)
|
752 |
parser.add_argument(
|
753 |
"--original_config_file",
|
754 |
default=None,
|
755 |
type=str,
|
756 |
help="The YAML config file corresponding to the original architecture.",
|
757 |
)
|
758 |
-
parser.add_argument(
|
759 |
-
"--extract_ema",
|
760 |
-
action="store_true",
|
761 |
-
help=(
|
762 |
-
"Only relevant for checkpoints that have both EMA and non-EMA weights. Whether to extract the EMA weights"
|
763 |
-
" or not. Defaults to `False`. Add `--extract_ema` to extract the EMA weights. EMA weights usually yield"
|
764 |
-
" higher quality images for inference. Non-EMA weights are usually better to continue fine-tuning."
|
765 |
-
),
|
766 |
-
)
|
767 |
parser.add_argument(
|
768 |
"--to_safetensors",
|
769 |
action="store_true",
|
770 |
help="Whether to store pipeline in safetensors format or not.",
|
771 |
)
|
772 |
-
parser.add_argument("--half", action="store_true", help="Save weights in half precision.")
|
|
|
773 |
parser.add_argument("--dump_path", default=None, type=str, required=True, help="Path to the output model.")
|
774 |
parser.add_argument("--device", type=str, help="Device to use (e.g. cpu, cuda:0, cuda:1, etc.)")
|
775 |
args = parser.parse_args()
|
@@ -777,22 +747,21 @@ if __name__ == "__main__":
|
|
777 |
pipe = convert_from_original_mvdream_ckpt(
|
778 |
checkpoint_path=args.checkpoint_path,
|
779 |
original_config_file=args.original_config_file,
|
780 |
-
extract_ema=args.extract_ema,
|
781 |
device=args.device,
|
782 |
)
|
783 |
|
784 |
if args.half:
|
785 |
pipe.to(torch_dtype=torch.float16)
|
786 |
-
|
787 |
-
|
788 |
-
|
789 |
-
|
790 |
-
|
791 |
-
|
792 |
-
|
793 |
-
|
794 |
-
|
795 |
-
|
796 |
-
|
797 |
-
|
798 |
-
pipe.save_pretrained(args.dump_path, safe_serialization=args.to_safetensors)
|
|
|
3 |
import argparse
|
4 |
import torch
|
5 |
import sys
|
6 |
+
|
7 |
sys.path.insert(0, '../')
|
8 |
|
9 |
from transformers import (
|
|
|
127 |
# return config
|
128 |
|
129 |
|
130 |
+
def assign_to_checkpoint(paths, checkpoint, old_checkpoint, attention_paths_to_split=None, additional_replacements=None, config=None):
|
|
|
|
|
131 |
"""
|
132 |
This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits
|
133 |
attention layers, and takes into account additional replacements that may arise.
|
|
|
143 |
|
144 |
target_shape = (-1, channels) if len(old_tensor.shape) == 3 else (-1)
|
145 |
|
146 |
+
assert config is not None
|
147 |
num_heads = old_tensor.shape[0] // config["num_head_channels"] // 3
|
148 |
|
149 |
old_tensor = old_tensor.reshape((num_heads, 3 * channels // num_heads) + old_tensor.shape[1:])
|
|
|
211 |
|
212 |
return mapping
|
213 |
|
214 |
+
|
215 |
def renew_attention_paths(old_list, n_shave_prefix_segments=0):
|
216 |
"""
|
217 |
Updates paths inside attentions to the new naming scheme (local renaming)
|
|
|
232 |
|
233 |
return mapping
|
234 |
|
235 |
+
|
236 |
# def convert_ldm_unet_checkpoint(
|
237 |
# checkpoint, config, path=None, extract_ema=False, controlnet=False, skip_extract_state_dict=False
|
238 |
# ):
|
|
|
498 |
}
|
499 |
return config
|
500 |
|
501 |
+
|
502 |
def convert_ldm_vae_checkpoint(checkpoint, config):
|
503 |
# extract state dict for VAE
|
504 |
vae_state_dict = {}
|
|
|
531 |
|
532 |
# Retrieves the keys for the encoder down blocks only
|
533 |
num_down_blocks = len({".".join(layer.split(".")[:3]) for layer in vae_state_dict if "encoder.down" in layer})
|
534 |
+
down_blocks = {layer_id: [key for key in vae_state_dict if f"down.{layer_id}" in key] for layer_id in range(num_down_blocks)}
|
|
|
|
|
535 |
|
536 |
# Retrieves the keys for the decoder up blocks only
|
537 |
num_up_blocks = len({".".join(layer.split(".")[:3]) for layer in vae_state_dict if "decoder.up" in layer})
|
538 |
+
up_blocks = {layer_id: [key for key in vae_state_dict if f"up.{layer_id}" in key] for layer_id in range(num_up_blocks)}
|
|
|
|
|
539 |
|
540 |
for i in range(num_down_blocks):
|
541 |
resnets = [key for key in down_blocks[i] if f"down.{i}" in key and f"down.{i}.downsample" not in key]
|
542 |
|
543 |
if f"encoder.down.{i}.downsample.conv.weight" in vae_state_dict:
|
544 |
+
new_checkpoint[f"encoder.down_blocks.{i}.downsamplers.0.conv.weight"] = vae_state_dict.pop(f"encoder.down.{i}.downsample.conv.weight")
|
545 |
+
new_checkpoint[f"encoder.down_blocks.{i}.downsamplers.0.conv.bias"] = vae_state_dict.pop(f"encoder.down.{i}.downsample.conv.bias")
|
|
|
|
|
|
|
|
|
546 |
|
547 |
paths = renew_vae_resnet_paths(resnets)
|
548 |
meta_path = {"old": f"down.{i}.block", "new": f"down_blocks.{i}.resnets"}
|
|
|
565 |
|
566 |
for i in range(num_up_blocks):
|
567 |
block_id = num_up_blocks - 1 - i
|
568 |
+
resnets = [key for key in up_blocks[block_id] if f"up.{block_id}" in key and f"up.{block_id}.upsample" not in key]
|
|
|
|
|
569 |
|
570 |
if f"decoder.up.{block_id}.upsample.conv.weight" in vae_state_dict:
|
571 |
+
new_checkpoint[f"decoder.up_blocks.{i}.upsamplers.0.conv.weight"] = vae_state_dict[f"decoder.up.{block_id}.upsample.conv.weight"]
|
572 |
+
new_checkpoint[f"decoder.up_blocks.{i}.upsamplers.0.conv.bias"] = vae_state_dict[f"decoder.up.{block_id}.upsample.conv.bias"]
|
|
|
|
|
|
|
|
|
573 |
|
574 |
paths = renew_vae_resnet_paths(resnets)
|
575 |
meta_path = {"old": f"up.{block_id}.block", "new": f"up_blocks.{i}.resnets"}
|
|
|
607 |
|
608 |
return mapping
|
609 |
|
610 |
+
|
611 |
def renew_vae_attention_paths(old_list, n_shave_prefix_segments=0):
|
612 |
"""
|
613 |
Updates paths inside attentions to the new naming scheme (local renaming)
|
|
|
649 |
if checkpoint[key].ndim > 2:
|
650 |
checkpoint[key] = checkpoint[key][:, :, 0]
|
651 |
|
652 |
+
|
653 |
+
def convert_from_original_mvdream_ckpt(checkpoint_path, original_config_file, device):
|
|
|
|
|
|
|
|
|
654 |
checkpoint = torch.load(checkpoint_path, map_location=device)
|
655 |
# print(f"Checkpoint: {checkpoint.keys()}")
|
656 |
torch.cuda.empty_cache()
|
|
|
688 |
# print(f"Unet Config: {original_config.model.params.unet_config.params}")
|
689 |
unet: MultiViewUNetWrapperModel = MultiViewUNetWrapperModel(**original_config.model.params.unet_config.params)
|
690 |
# print(f"Unet State Dict: {unet.state_dict().keys()}")
|
691 |
+
unet.load_state_dict({key.replace("model.diffusion_model.", "unet."): value for key, value in checkpoint.items() if key.replace("model.diffusion_model.", "unet.") in unet.state_dict()})
|
|
|
|
|
692 |
for param_name, param in unet.state_dict().items():
|
693 |
set_module_tensor_to_device(unet, param_name, "cuda:0", value=param)
|
694 |
|
|
|
696 |
vae_config = create_vae_diffusers_config(original_config, image_size=image_size)
|
697 |
converted_vae_checkpoint = convert_ldm_vae_checkpoint(checkpoint, vae_config)
|
698 |
|
699 |
+
if ("model" in original_config and "params" in original_config.model and "scale_factor" in original_config.model.params):
|
|
|
|
|
|
|
|
|
700 |
vae_scaling_factor = original_config.model.params.scale_factor
|
701 |
else:
|
702 |
+
vae_scaling_factor = 0.18215 # default SD scaling factor
|
703 |
|
704 |
vae_config["scaling_factor"] = vae_scaling_factor
|
705 |
|
706 |
with init_empty_weights():
|
707 |
vae = AutoencoderKL(**vae_config)
|
708 |
+
|
709 |
tokenizer: CLIPTokenizer = CLIPTokenizer.from_pretrained("openai/clip-vit-large-patch14")
|
710 |
text_encoder: CLIPTextModel = CLIPTextModel.from_pretrained("openai/clip-vit-large-patch14").to(device=torch.device("cuda:0")) # type: ignore
|
711 |
|
712 |
for param_name, param in converted_vae_checkpoint.items():
|
713 |
+
set_module_tensor_to_device(vae, param_name, "cuda:0", value=param)
|
714 |
|
715 |
pipe = MVDreamStableDiffusionPipeline(
|
716 |
vae=vae,
|
|
|
726 |
if __name__ == "__main__":
|
727 |
parser = argparse.ArgumentParser()
|
728 |
|
729 |
+
parser.add_argument("--checkpoint_path", default=None, type=str, required=True, help="Path to the checkpoint to convert.")
|
|
|
|
|
730 |
parser.add_argument(
|
731 |
"--original_config_file",
|
732 |
default=None,
|
733 |
type=str,
|
734 |
help="The YAML config file corresponding to the original architecture.",
|
735 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
736 |
parser.add_argument(
|
737 |
"--to_safetensors",
|
738 |
action="store_true",
|
739 |
help="Whether to store pipeline in safetensors format or not.",
|
740 |
)
|
741 |
+
parser.add_argument("--half", action="store_true", help="Save weights in half precision.")
|
742 |
+
parser.add_argument("--test", help="Whether to test inference after convertion.")
|
743 |
parser.add_argument("--dump_path", default=None, type=str, required=True, help="Path to the output model.")
|
744 |
parser.add_argument("--device", type=str, help="Device to use (e.g. cpu, cuda:0, cuda:1, etc.)")
|
745 |
args = parser.parse_args()
|
|
|
747 |
pipe = convert_from_original_mvdream_ckpt(
|
748 |
checkpoint_path=args.checkpoint_path,
|
749 |
original_config_file=args.original_config_file,
|
|
|
750 |
device=args.device,
|
751 |
)
|
752 |
|
753 |
if args.half:
|
754 |
pipe.to(torch_dtype=torch.float16)
|
755 |
+
|
756 |
+
if args.test:
|
757 |
+
images = pipe(
|
758 |
+
prompt="Head of Hatsune Miku",
|
759 |
+
negative_prompt="painting, bad quality, flat",
|
760 |
+
output_type="pil",
|
761 |
+
guidance_scale=7.5,
|
762 |
+
num_inference_steps=50,
|
763 |
+
)
|
764 |
+
for i, image in enumerate(images):
|
765 |
+
image.save(f"image_{i}.png") # type: ignore
|
766 |
+
|
767 |
+
pipe.save_pretrained(args.dump_path, safe_serialization=args.to_safetensors)
|
scripts/models.py
CHANGED
@@ -82,29 +82,19 @@ class Upsample(nn.Module):
|
|
82 |
upsampling occurs in the inner-two dimensions.
|
83 |
"""
|
84 |
|
85 |
-
def __init__(self,
|
86 |
-
channels,
|
87 |
-
use_conv,
|
88 |
-
dims=2,
|
89 |
-
out_channels=None,
|
90 |
-
padding=1):
|
91 |
super().__init__()
|
92 |
self.channels = channels
|
93 |
self.out_channels = out_channels or channels
|
94 |
self.use_conv = use_conv
|
95 |
self.dims = dims
|
96 |
if use_conv:
|
97 |
-
self.conv = conv_nd(dims,
|
98 |
-
self.channels,
|
99 |
-
self.out_channels,
|
100 |
-
3,
|
101 |
-
padding=padding)
|
102 |
|
103 |
def forward(self, x):
|
104 |
assert x.shape[1] == self.channels
|
105 |
if self.dims == 3:
|
106 |
-
x = F.interpolate(x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2),
|
107 |
-
mode="nearest")
|
108 |
else:
|
109 |
x = F.interpolate(x, scale_factor=2, mode="nearest")
|
110 |
if self.use_conv:
|
@@ -121,12 +111,7 @@ class Downsample(nn.Module):
|
|
121 |
downsampling occurs in the inner-two dimensions.
|
122 |
"""
|
123 |
|
124 |
-
def __init__(self,
|
125 |
-
channels,
|
126 |
-
use_conv,
|
127 |
-
dims=2,
|
128 |
-
out_channels=None,
|
129 |
-
padding=1):
|
130 |
super().__init__()
|
131 |
self.channels = channels
|
132 |
self.out_channels = out_channels or channels
|
@@ -134,12 +119,7 @@ class Downsample(nn.Module):
|
|
134 |
self.dims = dims
|
135 |
stride = 2 if dims != 3 else (1, 2, 2)
|
136 |
if use_conv:
|
137 |
-
self.op = conv_nd(dims,
|
138 |
-
self.channels,
|
139 |
-
self.out_channels,
|
140 |
-
3,
|
141 |
-
stride=stride,
|
142 |
-
padding=padding)
|
143 |
else:
|
144 |
assert self.channels == self.out_channels
|
145 |
self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride)
|
@@ -208,33 +188,22 @@ class ResBlock(TimestepBlock):
|
|
208 |
nn.SiLU(),
|
209 |
linear(
|
210 |
emb_channels,
|
211 |
-
2 * self.out_channels
|
212 |
-
if use_scale_shift_norm else self.out_channels,
|
213 |
),
|
214 |
)
|
215 |
self.out_layers = nn.Sequential(
|
216 |
normalization(self.out_channels),
|
217 |
nn.SiLU(),
|
218 |
nn.Dropout(p=dropout),
|
219 |
-
zero_module(
|
220 |
-
conv_nd(dims,
|
221 |
-
self.out_channels,
|
222 |
-
self.out_channels,
|
223 |
-
3,
|
224 |
-
padding=1)),
|
225 |
)
|
226 |
|
227 |
if self.out_channels == channels:
|
228 |
self.skip_connection = nn.Identity()
|
229 |
elif use_conv:
|
230 |
-
self.skip_connection = conv_nd(dims,
|
231 |
-
channels,
|
232 |
-
self.out_channels,
|
233 |
-
3,
|
234 |
-
padding=1)
|
235 |
else:
|
236 |
-
self.skip_connection = conv_nd(dims, channels, self.out_channels,
|
237 |
-
1)
|
238 |
|
239 |
def forward(self, x, emb):
|
240 |
"""
|
@@ -243,8 +212,7 @@ class ResBlock(TimestepBlock):
|
|
243 |
:param emb: an [N x emb_channels] Tensor of timestep embeddings.
|
244 |
:return: an [N x C x ...] Tensor of outputs.
|
245 |
"""
|
246 |
-
return checkpoint(self._forward, (x, emb), self.parameters(),
|
247 |
-
self.use_checkpoint)
|
248 |
|
249 |
def _forward(self, x, emb):
|
250 |
if self.updown:
|
@@ -289,9 +257,7 @@ class AttentionBlock(nn.Module):
|
|
289 |
if num_head_channels == -1:
|
290 |
self.num_heads = num_heads
|
291 |
else:
|
292 |
-
assert (
|
293 |
-
channels % num_head_channels == 0
|
294 |
-
), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}"
|
295 |
self.num_heads = channels // num_head_channels
|
296 |
self.use_checkpoint = use_checkpoint
|
297 |
self.norm = normalization(channels)
|
@@ -306,9 +272,7 @@ class AttentionBlock(nn.Module):
|
|
306 |
self.proj_out = zero_module(conv_nd(1, channels, channels, 1))
|
307 |
|
308 |
def forward(self, x):
|
309 |
-
return checkpoint(
|
310 |
-
self._forward, (x, ), self.parameters(), True
|
311 |
-
) # TODO: check checkpoint usage, is True # TODO: fix the .half call!!!
|
312 |
#return pt_checkpoint(self._forward, x) # pytorch
|
313 |
|
314 |
def _forward(self, x):
|
@@ -358,12 +322,9 @@ class QKVAttentionLegacy(nn.Module):
|
|
358 |
bs, width, length = qkv.shape
|
359 |
assert width % (3 * self.n_heads) == 0
|
360 |
ch = width // (3 * self.n_heads)
|
361 |
-
q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch,
|
362 |
-
dim=1)
|
363 |
scale = 1 / math.sqrt(math.sqrt(ch))
|
364 |
-
weight = th.einsum(
|
365 |
-
"bct,bcs->bts", q * scale,
|
366 |
-
k * scale) # More stable with f16 than dividing afterwards
|
367 |
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
|
368 |
a = th.einsum("bts,bcs->bct", weight, v)
|
369 |
return a.reshape(bs, -1, length)
|
@@ -397,10 +358,9 @@ class QKVAttention(nn.Module):
|
|
397 |
"bct,bcs->bts",
|
398 |
(q * scale).view(bs * self.n_heads, ch, length),
|
399 |
(k * scale).view(bs * self.n_heads, ch, length),
|
400 |
-
)
|
401 |
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
|
402 |
-
a = th.einsum("bts,bcs->bct", weight,
|
403 |
-
v.reshape(bs * self.n_heads, ch, length))
|
404 |
return a.reshape(bs, -1, length)
|
405 |
|
406 |
@staticmethod
|
@@ -450,41 +410,40 @@ class MultiViewUNetModel(nn.Module):
|
|
450 |
"""
|
451 |
|
452 |
def __init__(
|
453 |
-
|
454 |
-
|
455 |
-
|
456 |
-
|
457 |
-
|
458 |
-
|
459 |
-
|
460 |
-
|
461 |
-
|
462 |
-
|
463 |
-
|
464 |
-
|
465 |
-
|
466 |
-
|
467 |
-
|
468 |
-
|
469 |
-
|
470 |
-
|
471 |
-
|
472 |
-
|
473 |
-
|
474 |
-
|
475 |
-
|
476 |
-
|
477 |
-
|
478 |
-
|
479 |
-
|
480 |
-
|
481 |
-
|
482 |
-
|
483 |
-
|
484 |
-
|
485 |
):
|
486 |
super().__init__()
|
487 |
-
assert num_classes is not None
|
488 |
if use_spatial_transformer:
|
489 |
assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'
|
490 |
|
@@ -511,26 +470,19 @@ class MultiViewUNetModel(nn.Module):
|
|
511 |
self.num_res_blocks = len(channel_mult) * [num_res_blocks]
|
512 |
else:
|
513 |
if len(num_res_blocks) != len(channel_mult):
|
514 |
-
raise ValueError(
|
515 |
-
|
516 |
-
"as a list/tuple (per-level) with the same length as channel_mult"
|
517 |
-
)
|
518 |
self.num_res_blocks = num_res_blocks
|
519 |
if disable_self_attentions is not None:
|
520 |
# should be a list of booleans, indicating whether to disable self-attention in TransformerBlocks or not
|
521 |
assert len(disable_self_attentions) == len(channel_mult)
|
522 |
if num_attention_blocks is not None:
|
523 |
assert len(num_attention_blocks) == len(self.num_res_blocks)
|
524 |
-
assert all(
|
525 |
-
|
526 |
-
|
527 |
-
|
528 |
-
|
529 |
-
print(
|
530 |
-
f"Constructor of UNetModel received num_attention_blocks={num_attention_blocks}. "
|
531 |
-
f"This option has LESS priority than attention_resolutions {attention_resolutions}, "
|
532 |
-
f"i.e., in cases where num_attention_blocks[i] > 0 but 2**i not in attention_resolutions, "
|
533 |
-
f"attention will still not be set.")
|
534 |
|
535 |
self.attention_resolutions = attention_resolutions
|
536 |
self.dropout = dropout
|
@@ -562,42 +514,36 @@ class MultiViewUNetModel(nn.Module):
|
|
562 |
|
563 |
if self.num_classes is not None:
|
564 |
if isinstance(self.num_classes, int):
|
565 |
-
self.label_emb = nn.Embedding(num_classes, time_embed_dim)
|
566 |
elif self.num_classes == "continuous":
|
567 |
print("setting up linear c_adm embedding layer")
|
568 |
self.label_emb = nn.Linear(1, time_embed_dim)
|
569 |
elif self.num_classes == "sequential":
|
570 |
assert adm_in_channels is not None
|
571 |
-
self.label_emb = nn.Sequential(
|
572 |
-
|
573 |
-
|
574 |
-
|
575 |
-
|
576 |
-
))
|
577 |
else:
|
578 |
raise ValueError()
|
579 |
|
580 |
-
self.input_blocks = nn.ModuleList([
|
581 |
-
TimestepEmbedSequential(
|
582 |
-
conv_nd(dims, in_channels, model_channels, 3, padding=1))
|
583 |
-
])
|
584 |
self._feature_size = model_channels
|
585 |
input_block_chans = [model_channels]
|
586 |
ch = model_channels
|
587 |
ds = 1
|
588 |
for level, mult in enumerate(channel_mult):
|
589 |
for nr in range(self.num_res_blocks[level]):
|
590 |
-
layers: List[Any] = [
|
591 |
-
|
592 |
-
|
593 |
-
|
594 |
-
|
595 |
-
|
596 |
-
|
597 |
-
|
598 |
-
|
599 |
-
)
|
600 |
-
]
|
601 |
ch = mult * model_channels
|
602 |
if ds in attention_resolutions:
|
603 |
if num_head_channels == -1:
|
@@ -613,44 +559,29 @@ class MultiViewUNetModel(nn.Module):
|
|
613 |
else:
|
614 |
disabled_sa = False
|
615 |
|
616 |
-
if num_attention_blocks is None or nr < num_attention_blocks[
|
617 |
-
|
618 |
-
|
619 |
-
|
620 |
-
|
621 |
-
|
622 |
-
|
623 |
-
|
624 |
-
use_new_attention_order=use_new_attention_order,
|
625 |
-
) if not use_spatial_transformer else
|
626 |
-
SpatialTransformer3D(
|
627 |
-
ch,
|
628 |
-
num_heads,
|
629 |
-
dim_head,
|
630 |
-
depth=transformer_depth,
|
631 |
-
context_dim=context_dim,
|
632 |
-
disable_self_attn=disabled_sa,
|
633 |
-
use_linear=use_linear_in_transformer,
|
634 |
-
use_checkpoint=use_checkpoint))
|
635 |
self.input_blocks.append(TimestepEmbedSequential(*layers))
|
636 |
self._feature_size += ch
|
637 |
input_block_chans.append(ch)
|
638 |
if level != len(channel_mult) - 1:
|
639 |
out_ch = ch
|
640 |
-
self.input_blocks.append(
|
641 |
-
|
642 |
-
|
643 |
-
|
644 |
-
|
645 |
-
|
646 |
-
|
647 |
-
|
648 |
-
|
649 |
-
|
650 |
-
down=True,
|
651 |
-
) if resblock_updown else Downsample(
|
652 |
-
ch, conv_resample, dims=dims, out_channels=out_ch))
|
653 |
-
)
|
654 |
ch = out_ch
|
655 |
input_block_chans.append(ch)
|
656 |
ds *= 2
|
@@ -679,16 +610,8 @@ class MultiViewUNetModel(nn.Module):
|
|
679 |
num_heads=num_heads,
|
680 |
num_head_channels=dim_head,
|
681 |
use_new_attention_order=use_new_attention_order,
|
682 |
-
) if not use_spatial_transformer else
|
683 |
-
|
684 |
-
ch,
|
685 |
-
num_heads,
|
686 |
-
dim_head,
|
687 |
-
depth=transformer_depth,
|
688 |
-
context_dim=context_dim,
|
689 |
-
disable_self_attn=disable_middle_self_attn,
|
690 |
-
use_linear=use_linear_in_transformer,
|
691 |
-
use_checkpoint=use_checkpoint),
|
692 |
ResBlock(
|
693 |
ch,
|
694 |
time_embed_dim,
|
@@ -704,17 +627,15 @@ class MultiViewUNetModel(nn.Module):
|
|
704 |
for level, mult in list(enumerate(channel_mult))[::-1]:
|
705 |
for i in range(self.num_res_blocks[level] + 1):
|
706 |
ich = input_block_chans.pop()
|
707 |
-
layers = [
|
708 |
-
|
709 |
-
|
710 |
-
|
711 |
-
|
712 |
-
|
713 |
-
|
714 |
-
|
715 |
-
|
716 |
-
)
|
717 |
-
]
|
718 |
ch = model_channels * mult
|
719 |
if ds in attention_resolutions:
|
720 |
if num_head_channels == -1:
|
@@ -730,39 +651,26 @@ class MultiViewUNetModel(nn.Module):
|
|
730 |
else:
|
731 |
disabled_sa = False
|
732 |
|
733 |
-
if num_attention_blocks is None or i < num_attention_blocks[
|
734 |
-
|
735 |
-
layers.append(
|
736 |
-
AttentionBlock(
|
737 |
-
ch,
|
738 |
-
use_checkpoint=use_checkpoint,
|
739 |
-
num_heads=num_heads_upsample,
|
740 |
-
num_head_channels=dim_head,
|
741 |
-
use_new_attention_order=use_new_attention_order,
|
742 |
-
) if not use_spatial_transformer else
|
743 |
-
SpatialTransformer3D(
|
744 |
-
ch,
|
745 |
-
num_heads,
|
746 |
-
dim_head,
|
747 |
-
depth=transformer_depth,
|
748 |
-
context_dim=context_dim,
|
749 |
-
disable_self_attn=disabled_sa,
|
750 |
-
use_linear=use_linear_in_transformer,
|
751 |
-
use_checkpoint=use_checkpoint))
|
752 |
-
if level and i == self.num_res_blocks[level]:
|
753 |
-
out_ch = ch
|
754 |
-
layers.append(
|
755 |
-
ResBlock(
|
756 |
ch,
|
757 |
-
time_embed_dim,
|
758 |
-
dropout,
|
759 |
-
out_channels=out_ch,
|
760 |
-
dims=dims,
|
761 |
use_checkpoint=use_checkpoint,
|
762 |
-
|
763 |
-
|
764 |
-
|
765 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
766 |
ds //= 2
|
767 |
self.output_blocks.append(TimestepEmbedSequential(*layers))
|
768 |
self._feature_size += ch
|
@@ -770,8 +678,7 @@ class MultiViewUNetModel(nn.Module):
|
|
770 |
self.out = nn.Sequential(
|
771 |
normalization(ch),
|
772 |
nn.SiLU(),
|
773 |
-
zero_module(
|
774 |
-
conv_nd(dims, model_channels, out_channels, 3, padding=1)),
|
775 |
)
|
776 |
if self.predict_codebook_ids:
|
777 |
self.id_predictor = nn.Sequential(
|
@@ -796,14 +703,7 @@ class MultiViewUNetModel(nn.Module):
|
|
796 |
self.middle_block.apply(convert_module_to_f32)
|
797 |
self.output_blocks.apply(convert_module_to_f32)
|
798 |
|
799 |
-
def forward(self,
|
800 |
-
x,
|
801 |
-
timesteps=None,
|
802 |
-
context=None,
|
803 |
-
y: Optional[Tensor] = None,
|
804 |
-
camera=None,
|
805 |
-
num_frames=1,
|
806 |
-
**kwargs):
|
807 |
"""
|
808 |
Apply the model to an input batch.
|
809 |
:param x: an [(N x F) x C x ...] Tensor of inputs. F is the number of frames (views).
|
@@ -813,15 +713,10 @@ class MultiViewUNetModel(nn.Module):
|
|
813 |
:param num_frames: a integer indicating number of frames for tensor reshaping.
|
814 |
:return: an [(N x F) x C x ...] Tensor of outputs. F is the number of frames (views).
|
815 |
"""
|
816 |
-
assert x.shape[
|
817 |
-
|
818 |
-
assert (y is not None) == (
|
819 |
-
self.num_classes is not None
|
820 |
-
), "must specify y if and only if the model is class-conditional"
|
821 |
hs = []
|
822 |
-
t_emb = timestep_embedding(timesteps,
|
823 |
-
self.model_channels,
|
824 |
-
repeat_only=False)
|
825 |
emb = self.time_embed(t_emb)
|
826 |
|
827 |
if self.num_classes is not None:
|
|
|
82 |
upsampling occurs in the inner-two dimensions.
|
83 |
"""
|
84 |
|
85 |
+
def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1):
|
|
|
|
|
|
|
|
|
|
|
86 |
super().__init__()
|
87 |
self.channels = channels
|
88 |
self.out_channels = out_channels or channels
|
89 |
self.use_conv = use_conv
|
90 |
self.dims = dims
|
91 |
if use_conv:
|
92 |
+
self.conv = conv_nd(dims, self.channels, self.out_channels, 3, padding=padding)
|
|
|
|
|
|
|
|
|
93 |
|
94 |
def forward(self, x):
|
95 |
assert x.shape[1] == self.channels
|
96 |
if self.dims == 3:
|
97 |
+
x = F.interpolate(x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode="nearest")
|
|
|
98 |
else:
|
99 |
x = F.interpolate(x, scale_factor=2, mode="nearest")
|
100 |
if self.use_conv:
|
|
|
111 |
downsampling occurs in the inner-two dimensions.
|
112 |
"""
|
113 |
|
114 |
+
def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1):
|
|
|
|
|
|
|
|
|
|
|
115 |
super().__init__()
|
116 |
self.channels = channels
|
117 |
self.out_channels = out_channels or channels
|
|
|
119 |
self.dims = dims
|
120 |
stride = 2 if dims != 3 else (1, 2, 2)
|
121 |
if use_conv:
|
122 |
+
self.op = conv_nd(dims, self.channels, self.out_channels, 3, stride=stride, padding=padding)
|
|
|
|
|
|
|
|
|
|
|
123 |
else:
|
124 |
assert self.channels == self.out_channels
|
125 |
self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride)
|
|
|
188 |
nn.SiLU(),
|
189 |
linear(
|
190 |
emb_channels,
|
191 |
+
2 * self.out_channels if use_scale_shift_norm else self.out_channels,
|
|
|
192 |
),
|
193 |
)
|
194 |
self.out_layers = nn.Sequential(
|
195 |
normalization(self.out_channels),
|
196 |
nn.SiLU(),
|
197 |
nn.Dropout(p=dropout),
|
198 |
+
zero_module(conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1)),
|
|
|
|
|
|
|
|
|
|
|
199 |
)
|
200 |
|
201 |
if self.out_channels == channels:
|
202 |
self.skip_connection = nn.Identity()
|
203 |
elif use_conv:
|
204 |
+
self.skip_connection = conv_nd(dims, channels, self.out_channels, 3, padding=1)
|
|
|
|
|
|
|
|
|
205 |
else:
|
206 |
+
self.skip_connection = conv_nd(dims, channels, self.out_channels, 1)
|
|
|
207 |
|
208 |
def forward(self, x, emb):
|
209 |
"""
|
|
|
212 |
:param emb: an [N x emb_channels] Tensor of timestep embeddings.
|
213 |
:return: an [N x C x ...] Tensor of outputs.
|
214 |
"""
|
215 |
+
return checkpoint(self._forward, (x, emb), self.parameters(), self.use_checkpoint)
|
|
|
216 |
|
217 |
def _forward(self, x, emb):
|
218 |
if self.updown:
|
|
|
257 |
if num_head_channels == -1:
|
258 |
self.num_heads = num_heads
|
259 |
else:
|
260 |
+
assert (channels % num_head_channels == 0), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}"
|
|
|
|
|
261 |
self.num_heads = channels // num_head_channels
|
262 |
self.use_checkpoint = use_checkpoint
|
263 |
self.norm = normalization(channels)
|
|
|
272 |
self.proj_out = zero_module(conv_nd(1, channels, channels, 1))
|
273 |
|
274 |
def forward(self, x):
|
275 |
+
return checkpoint(self._forward, (x,), self.parameters(), True) # TODO: check checkpoint usage, is True # TODO: fix the .half call!!!
|
|
|
|
|
276 |
#return pt_checkpoint(self._forward, x) # pytorch
|
277 |
|
278 |
def _forward(self, x):
|
|
|
322 |
bs, width, length = qkv.shape
|
323 |
assert width % (3 * self.n_heads) == 0
|
324 |
ch = width // (3 * self.n_heads)
|
325 |
+
q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1)
|
|
|
326 |
scale = 1 / math.sqrt(math.sqrt(ch))
|
327 |
+
weight = th.einsum("bct,bcs->bts", q * scale, k * scale) # More stable with f16 than dividing afterwards
|
|
|
|
|
328 |
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
|
329 |
a = th.einsum("bts,bcs->bct", weight, v)
|
330 |
return a.reshape(bs, -1, length)
|
|
|
358 |
"bct,bcs->bts",
|
359 |
(q * scale).view(bs * self.n_heads, ch, length),
|
360 |
(k * scale).view(bs * self.n_heads, ch, length),
|
361 |
+
) # More stable with f16 than dividing afterwards
|
362 |
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
|
363 |
+
a = th.einsum("bts,bcs->bct", weight, v.reshape(bs * self.n_heads, ch, length))
|
|
|
364 |
return a.reshape(bs, -1, length)
|
365 |
|
366 |
@staticmethod
|
|
|
410 |
"""
|
411 |
|
412 |
def __init__(
|
413 |
+
self,
|
414 |
+
image_size,
|
415 |
+
in_channels,
|
416 |
+
model_channels,
|
417 |
+
out_channels,
|
418 |
+
num_res_blocks,
|
419 |
+
attention_resolutions,
|
420 |
+
dropout=0,
|
421 |
+
channel_mult=(1, 2, 4, 8),
|
422 |
+
conv_resample=True,
|
423 |
+
dims=2,
|
424 |
+
num_classes=None,
|
425 |
+
use_checkpoint=False,
|
426 |
+
use_fp16=False,
|
427 |
+
use_bf16=False,
|
428 |
+
num_heads=-1,
|
429 |
+
num_head_channels=-1,
|
430 |
+
num_heads_upsample=-1,
|
431 |
+
use_scale_shift_norm=False,
|
432 |
+
resblock_updown=False,
|
433 |
+
use_new_attention_order=False,
|
434 |
+
use_spatial_transformer=False, # custom transformer support
|
435 |
+
transformer_depth=1, # custom transformer support
|
436 |
+
context_dim=None, # custom transformer support
|
437 |
+
n_embed=None, # custom support for prediction of discrete ids into codebook of first stage vq model
|
438 |
+
legacy=True,
|
439 |
+
disable_self_attentions=None,
|
440 |
+
num_attention_blocks=None,
|
441 |
+
disable_middle_self_attn=False,
|
442 |
+
use_linear_in_transformer=False,
|
443 |
+
adm_in_channels=None,
|
444 |
+
camera_dim=None,
|
445 |
):
|
446 |
super().__init__()
|
|
|
447 |
if use_spatial_transformer:
|
448 |
assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'
|
449 |
|
|
|
470 |
self.num_res_blocks = len(channel_mult) * [num_res_blocks]
|
471 |
else:
|
472 |
if len(num_res_blocks) != len(channel_mult):
|
473 |
+
raise ValueError("provide num_res_blocks either as an int (globally constant) or "
|
474 |
+
"as a list/tuple (per-level) with the same length as channel_mult")
|
|
|
|
|
475 |
self.num_res_blocks = num_res_blocks
|
476 |
if disable_self_attentions is not None:
|
477 |
# should be a list of booleans, indicating whether to disable self-attention in TransformerBlocks or not
|
478 |
assert len(disable_self_attentions) == len(channel_mult)
|
479 |
if num_attention_blocks is not None:
|
480 |
assert len(num_attention_blocks) == len(self.num_res_blocks)
|
481 |
+
assert all(map(lambda i: self.num_res_blocks[i] >= num_attention_blocks[i], range(len(num_attention_blocks))))
|
482 |
+
print(f"Constructor of UNetModel received num_attention_blocks={num_attention_blocks}. "
|
483 |
+
f"This option has LESS priority than attention_resolutions {attention_resolutions}, "
|
484 |
+
f"i.e., in cases where num_attention_blocks[i] > 0 but 2**i not in attention_resolutions, "
|
485 |
+
f"attention will still not be set.")
|
|
|
|
|
|
|
|
|
|
|
486 |
|
487 |
self.attention_resolutions = attention_resolutions
|
488 |
self.dropout = dropout
|
|
|
514 |
|
515 |
if self.num_classes is not None:
|
516 |
if isinstance(self.num_classes, int):
|
517 |
+
self.label_emb = nn.Embedding(self.num_classes, time_embed_dim)
|
518 |
elif self.num_classes == "continuous":
|
519 |
print("setting up linear c_adm embedding layer")
|
520 |
self.label_emb = nn.Linear(1, time_embed_dim)
|
521 |
elif self.num_classes == "sequential":
|
522 |
assert adm_in_channels is not None
|
523 |
+
self.label_emb = nn.Sequential(nn.Sequential(
|
524 |
+
linear(adm_in_channels, time_embed_dim),
|
525 |
+
nn.SiLU(),
|
526 |
+
linear(time_embed_dim, time_embed_dim),
|
527 |
+
))
|
|
|
528 |
else:
|
529 |
raise ValueError()
|
530 |
|
531 |
+
self.input_blocks = nn.ModuleList([TimestepEmbedSequential(conv_nd(dims, in_channels, model_channels, 3, padding=1))])
|
|
|
|
|
|
|
532 |
self._feature_size = model_channels
|
533 |
input_block_chans = [model_channels]
|
534 |
ch = model_channels
|
535 |
ds = 1
|
536 |
for level, mult in enumerate(channel_mult):
|
537 |
for nr in range(self.num_res_blocks[level]):
|
538 |
+
layers: List[Any] = [ResBlock(
|
539 |
+
ch,
|
540 |
+
time_embed_dim,
|
541 |
+
dropout,
|
542 |
+
out_channels=mult * model_channels,
|
543 |
+
dims=dims,
|
544 |
+
use_checkpoint=use_checkpoint,
|
545 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
546 |
+
)]
|
|
|
|
|
547 |
ch = mult * model_channels
|
548 |
if ds in attention_resolutions:
|
549 |
if num_head_channels == -1:
|
|
|
559 |
else:
|
560 |
disabled_sa = False
|
561 |
|
562 |
+
if num_attention_blocks is None or nr < num_attention_blocks[level]:
|
563 |
+
layers.append(AttentionBlock(
|
564 |
+
ch,
|
565 |
+
use_checkpoint=use_checkpoint,
|
566 |
+
num_heads=num_heads,
|
567 |
+
num_head_channels=dim_head,
|
568 |
+
use_new_attention_order=use_new_attention_order,
|
569 |
+
) if not use_spatial_transformer else SpatialTransformer3D(ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim, disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer, use_checkpoint=use_checkpoint))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
570 |
self.input_blocks.append(TimestepEmbedSequential(*layers))
|
571 |
self._feature_size += ch
|
572 |
input_block_chans.append(ch)
|
573 |
if level != len(channel_mult) - 1:
|
574 |
out_ch = ch
|
575 |
+
self.input_blocks.append(TimestepEmbedSequential(ResBlock(
|
576 |
+
ch,
|
577 |
+
time_embed_dim,
|
578 |
+
dropout,
|
579 |
+
out_channels=out_ch,
|
580 |
+
dims=dims,
|
581 |
+
use_checkpoint=use_checkpoint,
|
582 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
583 |
+
down=True,
|
584 |
+
) if resblock_updown else Downsample(ch, conv_resample, dims=dims, out_channels=out_ch)))
|
|
|
|
|
|
|
|
|
585 |
ch = out_ch
|
586 |
input_block_chans.append(ch)
|
587 |
ds *= 2
|
|
|
610 |
num_heads=num_heads,
|
611 |
num_head_channels=dim_head,
|
612 |
use_new_attention_order=use_new_attention_order,
|
613 |
+
) if not use_spatial_transformer else SpatialTransformer3D( # always uses a self-attn
|
614 |
+
ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim, disable_self_attn=disable_middle_self_attn, use_linear=use_linear_in_transformer, use_checkpoint=use_checkpoint),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
615 |
ResBlock(
|
616 |
ch,
|
617 |
time_embed_dim,
|
|
|
627 |
for level, mult in list(enumerate(channel_mult))[::-1]:
|
628 |
for i in range(self.num_res_blocks[level] + 1):
|
629 |
ich = input_block_chans.pop()
|
630 |
+
layers = [ResBlock(
|
631 |
+
ch + ich,
|
632 |
+
time_embed_dim,
|
633 |
+
dropout,
|
634 |
+
out_channels=model_channels * mult,
|
635 |
+
dims=dims,
|
636 |
+
use_checkpoint=use_checkpoint,
|
637 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
638 |
+
)]
|
|
|
|
|
639 |
ch = model_channels * mult
|
640 |
if ds in attention_resolutions:
|
641 |
if num_head_channels == -1:
|
|
|
651 |
else:
|
652 |
disabled_sa = False
|
653 |
|
654 |
+
if num_attention_blocks is None or i < num_attention_blocks[level]:
|
655 |
+
layers.append(AttentionBlock(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
656 |
ch,
|
|
|
|
|
|
|
|
|
657 |
use_checkpoint=use_checkpoint,
|
658 |
+
num_heads=num_heads_upsample,
|
659 |
+
num_head_channels=dim_head,
|
660 |
+
use_new_attention_order=use_new_attention_order,
|
661 |
+
) if not use_spatial_transformer else SpatialTransformer3D(ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim, disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer, use_checkpoint=use_checkpoint))
|
662 |
+
if level and i == self.num_res_blocks[level]:
|
663 |
+
out_ch = ch
|
664 |
+
layers.append(ResBlock(
|
665 |
+
ch,
|
666 |
+
time_embed_dim,
|
667 |
+
dropout,
|
668 |
+
out_channels=out_ch,
|
669 |
+
dims=dims,
|
670 |
+
use_checkpoint=use_checkpoint,
|
671 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
672 |
+
up=True,
|
673 |
+
) if resblock_updown else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch))
|
674 |
ds //= 2
|
675 |
self.output_blocks.append(TimestepEmbedSequential(*layers))
|
676 |
self._feature_size += ch
|
|
|
678 |
self.out = nn.Sequential(
|
679 |
normalization(ch),
|
680 |
nn.SiLU(),
|
681 |
+
zero_module(conv_nd(dims, model_channels, out_channels, 3, padding=1)),
|
|
|
682 |
)
|
683 |
if self.predict_codebook_ids:
|
684 |
self.id_predictor = nn.Sequential(
|
|
|
703 |
self.middle_block.apply(convert_module_to_f32)
|
704 |
self.output_blocks.apply(convert_module_to_f32)
|
705 |
|
706 |
+
def forward(self, x, timesteps=None, context=None, y: Optional[Tensor] = None, camera=None, num_frames=1, **kwargs):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
707 |
"""
|
708 |
Apply the model to an input batch.
|
709 |
:param x: an [(N x F) x C x ...] Tensor of inputs. F is the number of frames (views).
|
|
|
713 |
:param num_frames: a integer indicating number of frames for tensor reshaping.
|
714 |
:return: an [(N x F) x C x ...] Tensor of outputs. F is the number of frames (views).
|
715 |
"""
|
716 |
+
assert x.shape[0] % num_frames == 0, "[UNet] input batch size must be dividable by num_frames!"
|
717 |
+
assert (y is not None) == (self.num_classes is not None), "must specify y if and only if the model is class-conditional"
|
|
|
|
|
|
|
718 |
hs = []
|
719 |
+
t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
|
|
|
|
|
720 |
emb = self.time_embed(t_emb)
|
721 |
|
722 |
if self.num_classes is not None:
|
scripts/pipeline_mvdream.py
CHANGED
@@ -1,9 +1,8 @@
|
|
|
|
|
|
1 |
import inspect
|
2 |
from typing import Any, Callable, Dict, List, Optional, Union
|
3 |
-
|
4 |
-
import torch
|
5 |
from transformers import CLIPTextModel, CLIPTokenizer
|
6 |
-
|
7 |
from diffusers import AutoencoderKL, DiffusionPipeline
|
8 |
from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
|
9 |
from diffusers.utils import (
|
@@ -13,22 +12,16 @@ from diffusers.utils import (
|
|
13 |
logging,
|
14 |
replace_example_docstring,
|
15 |
)
|
16 |
-
|
17 |
-
try:
|
18 |
-
from diffusers import randn_tensor # old import
|
19 |
-
except ImportError:
|
20 |
-
from diffusers.utils.torch_utils import randn_tensor # new import
|
21 |
-
|
22 |
from diffusers.configuration_utils import FrozenDict
|
23 |
-
import numpy as np
|
24 |
from diffusers.schedulers import DDIMScheduler
|
25 |
-
|
26 |
-
|
27 |
-
|
|
|
28 |
|
29 |
-
|
30 |
|
31 |
-
|
32 |
|
33 |
def create_camera_to_world_matrix(elevation, azimuth):
|
34 |
elevation = np.radians(elevation)
|
@@ -59,29 +52,21 @@ def create_camera_to_world_matrix(elevation, azimuth):
|
|
59 |
def convert_opengl_to_blender(camera_matrix):
|
60 |
if isinstance(camera_matrix, np.ndarray):
|
61 |
# Construct transformation matrix to convert from OpenGL space to Blender space
|
62 |
-
flip_yz = np.array([[1, 0, 0, 0], [0, 0, -1, 0], [0, 1, 0, 0],
|
63 |
-
[0, 0, 0, 1]])
|
64 |
camera_matrix_blender = np.dot(flip_yz, camera_matrix)
|
65 |
else:
|
66 |
# Construct transformation matrix to convert from OpenGL space to Blender space
|
67 |
-
flip_yz = torch.tensor([[1, 0, 0, 0], [0, 0, -1, 0], [0, 1, 0, 0],
|
68 |
-
[0, 0, 0, 1]])
|
69 |
if camera_matrix.ndim == 3:
|
70 |
flip_yz = flip_yz.unsqueeze(0)
|
71 |
-
camera_matrix_blender = torch.matmul(flip_yz.to(camera_matrix),
|
72 |
-
camera_matrix)
|
73 |
return camera_matrix_blender
|
74 |
|
75 |
|
76 |
-
def get_camera(num_frames,
|
77 |
-
elevation=15,
|
78 |
-
azimuth_start=0,
|
79 |
-
azimuth_span=360,
|
80 |
-
blender_coord=True):
|
81 |
angle_gap = azimuth_span / num_frames
|
82 |
cameras = []
|
83 |
-
for azimuth in np.arange(azimuth_start, azimuth_span + azimuth_start,
|
84 |
-
angle_gap):
|
85 |
camera_matrix = create_camera_to_world_matrix(elevation, azimuth)
|
86 |
if blender_coord:
|
87 |
camera_matrix = convert_opengl_to_blender(camera_matrix)
|
@@ -101,36 +86,25 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
101 |
):
|
102 |
super().__init__()
|
103 |
|
104 |
-
if hasattr(scheduler.config,
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
" file")
|
113 |
-
deprecate("steps_offset!=1",
|
114 |
-
"1.0.0",
|
115 |
-
deprecation_message,
|
116 |
-
standard_warn=False)
|
117 |
new_config = dict(scheduler.config)
|
118 |
new_config["steps_offset"] = 1
|
119 |
scheduler._internal_dict = FrozenDict(new_config)
|
120 |
|
121 |
-
if hasattr(scheduler.config,
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
" nice if you could open a Pull request for the `scheduler/scheduler_config.json` file"
|
129 |
-
)
|
130 |
-
deprecate("clip_sample not set",
|
131 |
-
"1.0.0",
|
132 |
-
deprecation_message,
|
133 |
-
standard_warn=False)
|
134 |
new_config = dict(scheduler.config)
|
135 |
new_config["clip_sample"] = False
|
136 |
scheduler._internal_dict = FrozenDict(new_config)
|
@@ -142,8 +116,7 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
142 |
tokenizer=tokenizer,
|
143 |
text_encoder=text_encoder,
|
144 |
)
|
145 |
-
self.vae_scale_factor = 2**(len(self.vae.config.block_out_channels) -
|
146 |
-
1)
|
147 |
self.register_to_config(requires_safety_checker=False)
|
148 |
|
149 |
def enable_vae_slicing(self):
|
@@ -189,16 +162,13 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
189 |
if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
|
190 |
from accelerate import cpu_offload
|
191 |
else:
|
192 |
-
raise ImportError(
|
193 |
-
"`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher"
|
194 |
-
)
|
195 |
|
196 |
device = torch.device(f"cuda:{gpu_id}")
|
197 |
|
198 |
if self.device.type != "cpu":
|
199 |
self.to("cpu", silence_dtype_warnings=True)
|
200 |
-
torch.cuda.empty_cache(
|
201 |
-
) # otherwise we don't see the memory savings (but they probably exist)
|
202 |
|
203 |
for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
|
204 |
cpu_offload(cpu_offloaded_model, device)
|
@@ -210,26 +180,20 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
210 |
method is called, and the model remains in GPU until the next model runs. Memory savings are lower than with
|
211 |
`enable_sequential_cpu_offload`, but performance is much better due to the iterative execution of the `unet`.
|
212 |
"""
|
213 |
-
if is_accelerate_available() and is_accelerate_version(
|
214 |
-
">=", "0.17.0.dev0"):
|
215 |
from accelerate import cpu_offload_with_hook
|
216 |
else:
|
217 |
-
raise ImportError(
|
218 |
-
"`enable_model_offload` requires `accelerate v0.17.0` or higher."
|
219 |
-
)
|
220 |
|
221 |
device = torch.device(f"cuda:{gpu_id}")
|
222 |
|
223 |
if self.device.type != "cpu":
|
224 |
self.to("cpu", silence_dtype_warnings=True)
|
225 |
-
torch.cuda.empty_cache(
|
226 |
-
) # otherwise we don't see the memory savings (but they probably exist)
|
227 |
|
228 |
hook = None
|
229 |
for cpu_offloaded_model in [self.text_encoder, self.unet, self.vae]:
|
230 |
-
_, hook = cpu_offload_with_hook(cpu_offloaded_model,
|
231 |
-
device,
|
232 |
-
prev_module_hook=hook)
|
233 |
|
234 |
# We'll offload the last model manually.
|
235 |
self.final_offload_hook = hook
|
@@ -244,9 +208,7 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
244 |
if not hasattr(self.unet, "_hf_hook"):
|
245 |
return self.device
|
246 |
for module in self.unet.modules():
|
247 |
-
if (hasattr(module, "_hf_hook")
|
248 |
-
and hasattr(module._hf_hook, "execution_device")
|
249 |
-
and module._hf_hook.execution_device is not None):
|
250 |
return torch.device(module._hf_hook.execution_device)
|
251 |
return self.device
|
252 |
|
@@ -257,8 +219,6 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
257 |
num_images_per_prompt,
|
258 |
do_classifier_free_guidance: bool,
|
259 |
negative_prompt=None,
|
260 |
-
prompt_embeds: Optional[torch.FloatTensor] = None,
|
261 |
-
negative_prompt_embeds: Optional[torch.FloatTensor] = None,
|
262 |
):
|
263 |
r"""
|
264 |
Encodes the prompt into text encoder hidden states.
|
@@ -289,67 +249,55 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
289 |
elif prompt is not None and isinstance(prompt, list):
|
290 |
batch_size = len(prompt)
|
291 |
else:
|
292 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
293 |
|
294 |
-
if
|
295 |
-
|
296 |
-
|
297 |
-
|
298 |
-
max_length=self.tokenizer.model_max_length,
|
299 |
-
truncation=True,
|
300 |
-
return_tensors="pt",
|
301 |
-
)
|
302 |
-
text_input_ids = text_inputs.input_ids
|
303 |
-
untruncated_ids = self.tokenizer(prompt,
|
304 |
-
padding="longest",
|
305 |
-
return_tensors="pt").input_ids
|
306 |
-
|
307 |
-
if untruncated_ids.shape[-1] >= text_input_ids.shape[
|
308 |
-
-1] and not torch.equal(text_input_ids, untruncated_ids):
|
309 |
-
removed_text = self.tokenizer.batch_decode(
|
310 |
-
untruncated_ids[:, self.tokenizer.model_max_length - 1:-1])
|
311 |
-
logger.warning(
|
312 |
-
"The following part of your input was truncated because CLIP can only handle sequences up to"
|
313 |
-
f" {self.tokenizer.model_max_length} tokens: {removed_text}"
|
314 |
-
)
|
315 |
|
316 |
-
|
317 |
-
|
318 |
-
|
319 |
-
|
320 |
-
attention_mask = None
|
321 |
|
322 |
-
|
323 |
-
|
324 |
-
|
325 |
-
|
326 |
-
|
327 |
|
328 |
-
prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype,
|
329 |
-
device=device)
|
330 |
|
331 |
bs_embed, seq_len, _ = prompt_embeds.shape
|
332 |
# duplicate text embeddings for each generation per prompt, using mps friendly method
|
333 |
prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
|
334 |
-
prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt,
|
335 |
-
seq_len, -1)
|
336 |
|
337 |
# get unconditional embeddings for classifier free guidance
|
338 |
-
if do_classifier_free_guidance
|
339 |
uncond_tokens: List[str]
|
340 |
if negative_prompt is None:
|
341 |
uncond_tokens = [""] * batch_size
|
342 |
elif type(prompt) is not type(negative_prompt):
|
343 |
-
raise TypeError(
|
344 |
-
|
345 |
-
f" {type(prompt)}.")
|
346 |
elif isinstance(negative_prompt, str):
|
347 |
uncond_tokens = [negative_prompt]
|
348 |
elif batch_size != len(negative_prompt):
|
349 |
-
raise ValueError(
|
350 |
-
|
351 |
-
|
352 |
-
" the batch size of `prompt`.")
|
353 |
else:
|
354 |
uncond_tokens = negative_prompt
|
355 |
|
@@ -362,8 +310,7 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
362 |
return_tensors="pt",
|
363 |
)
|
364 |
|
365 |
-
if hasattr(self.text_encoder.config, "use_attention_mask"
|
366 |
-
) and self.text_encoder.config.use_attention_mask:
|
367 |
attention_mask = uncond_input.attention_mask.to(device)
|
368 |
else:
|
369 |
attention_mask = None
|
@@ -374,17 +321,13 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
374 |
)
|
375 |
negative_prompt_embeds = negative_prompt_embeds[0]
|
376 |
|
377 |
-
if do_classifier_free_guidance:
|
378 |
# duplicate unconditional embeddings for each generation per prompt, using mps friendly method
|
379 |
seq_len = negative_prompt_embeds.shape[1]
|
380 |
|
381 |
-
negative_prompt_embeds = negative_prompt_embeds.to(
|
382 |
-
dtype=self.text_encoder.dtype, device=device)
|
383 |
|
384 |
-
negative_prompt_embeds = negative_prompt_embeds.repeat(
|
385 |
-
|
386 |
-
negative_prompt_embeds = negative_prompt_embeds.view(
|
387 |
-
batch_size * num_images_per_prompt, seq_len, -1)
|
388 |
|
389 |
# For classifier free guidance, we need to do two forward passes.
|
390 |
# Here we concatenate the unconditional and text embeddings into a single batch
|
@@ -407,42 +350,25 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
407 |
# eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
|
408 |
# and should be between [0, 1]
|
409 |
|
410 |
-
accepts_eta = "eta" in set(
|
411 |
-
inspect.signature(self.scheduler.step).parameters.keys())
|
412 |
extra_step_kwargs = {}
|
413 |
if accepts_eta:
|
414 |
extra_step_kwargs["eta"] = eta
|
415 |
|
416 |
# check if the scheduler accepts generator
|
417 |
-
accepts_generator = "generator" in set(
|
418 |
-
inspect.signature(self.scheduler.step).parameters.keys())
|
419 |
if accepts_generator:
|
420 |
extra_step_kwargs["generator"] = generator
|
421 |
return extra_step_kwargs
|
422 |
|
423 |
-
def prepare_latents(self,
|
424 |
-
|
425 |
-
num_channels_latents,
|
426 |
-
height,
|
427 |
-
width,
|
428 |
-
dtype,
|
429 |
-
device,
|
430 |
-
generator,
|
431 |
-
latents=None):
|
432 |
-
shape = (batch_size, num_channels_latents,
|
433 |
-
height // self.vae_scale_factor,
|
434 |
-
width // self.vae_scale_factor)
|
435 |
if isinstance(generator, list) and len(generator) != batch_size:
|
436 |
-
raise ValueError(
|
437 |
-
|
438 |
-
f" size of {batch_size}. Make sure the batch size matches the length of the generators."
|
439 |
-
)
|
440 |
|
441 |
if latents is None:
|
442 |
-
latents = randn_tensor(shape,
|
443 |
-
generator=generator,
|
444 |
-
device=device,
|
445 |
-
dtype=dtype)
|
446 |
else:
|
447 |
latents = latents.to(device)
|
448 |
|
@@ -451,7 +377,6 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
451 |
return latents
|
452 |
|
453 |
@torch.no_grad()
|
454 |
-
@replace_example_docstring(EXAMPLE_DOC_STRING)
|
455 |
def __call__(
|
456 |
self,
|
457 |
height: int = 256,
|
@@ -462,87 +387,11 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
462 |
negative_prompt: str = "bad quality",
|
463 |
num_images_per_prompt: int = 1,
|
464 |
eta: float = 0.0,
|
465 |
-
generator: Optional[Union[torch.Generator,
|
466 |
-
List[torch.Generator]]] = None,
|
467 |
output_type: Optional[str] = "pil",
|
468 |
-
|
469 |
-
callback: Optional[Callable[[int, int, torch.FloatTensor],
|
470 |
-
None]] = None,
|
471 |
callback_steps: int = 1,
|
472 |
-
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
|
473 |
-
controlnet_conditioning_scale: float = 1.0,
|
474 |
):
|
475 |
-
r"""
|
476 |
-
Function invoked when calling the pipeline for generation.
|
477 |
-
|
478 |
-
Args:
|
479 |
-
input_imgs (`PIL` or `List[PIL]`, *optional*):
|
480 |
-
The single input image for each 3D object
|
481 |
-
prompt_imgs (`PIL` or `List[PIL]`, *optional*):
|
482 |
-
Same as input_imgs, but will be used later as an image prompt condition, encoded by CLIP feature
|
483 |
-
height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
|
484 |
-
The height in pixels of the generated image.
|
485 |
-
width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
|
486 |
-
The width in pixels of the generated image.
|
487 |
-
num_inference_steps (`int`, *optional*, defaults to 50):
|
488 |
-
The number of denoising steps. More denoising steps usually lead to a higher quality image at the
|
489 |
-
expense of slower inference.
|
490 |
-
guidance_scale (`float`, *optional*, defaults to 7.5):
|
491 |
-
Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
|
492 |
-
`guidance_scale` is defined as `w` of equation 2. of [Imagen
|
493 |
-
Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
|
494 |
-
1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
|
495 |
-
usually at the expense of lower image quality.
|
496 |
-
negative_prompt (`str` or `List[str]`, *optional*):
|
497 |
-
The prompt or prompts not to guide the image generation. If not defined, one has to pass
|
498 |
-
`negative_prompt_embeds`. instead. If not defined, one has to pass `negative_prompt_embeds`. instead.
|
499 |
-
Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
|
500 |
-
num_images_per_prompt (`int`, *optional*, defaults to 1):
|
501 |
-
The number of images to generate per prompt.
|
502 |
-
eta (`float`, *optional*, defaults to 0.0):
|
503 |
-
Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
|
504 |
-
[`schedulers.DDIMScheduler`], will be ignored for others.
|
505 |
-
generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
|
506 |
-
One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
|
507 |
-
to make generation deterministic.
|
508 |
-
latents (`torch.FloatTensor`, *optional*):
|
509 |
-
Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
|
510 |
-
generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
|
511 |
-
tensor will ge generated by sampling using the supplied random `generator`.
|
512 |
-
prompt_embeds (`torch.FloatTensor`, *optional*):
|
513 |
-
Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
|
514 |
-
provided, text embeddings will be generated from `prompt` input argument.
|
515 |
-
negative_prompt_embeds (`torch.FloatTensor`, *optional*):
|
516 |
-
Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
|
517 |
-
weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
|
518 |
-
argument.
|
519 |
-
output_type (`str`, *optional*, defaults to `"pil"`):
|
520 |
-
The output format of the generate image. Choose between
|
521 |
-
[PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
|
522 |
-
return_dict (`bool`, *optional*, defaults to `True`):
|
523 |
-
Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
|
524 |
-
plain tuple.
|
525 |
-
callback (`Callable`, *optional*):
|
526 |
-
A function that will be called every `callback_steps` steps during inference. The function will be
|
527 |
-
called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
|
528 |
-
callback_steps (`int`, *optional*, defaults to 1):
|
529 |
-
The frequency at which the `callback` function will be called. If not specified, the callback will be
|
530 |
-
called at every step.
|
531 |
-
cross_attention_kwargs (`dict`, *optional*):
|
532 |
-
A kwargs dictionary that if specified is passed along to the `AttnProcessor` as defined under
|
533 |
-
`self.processor` in
|
534 |
-
[diffusers.cross_attention](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/cross_attention.py).
|
535 |
-
|
536 |
-
Examples:
|
537 |
-
|
538 |
-
Returns:
|
539 |
-
[`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
|
540 |
-
[`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
|
541 |
-
When returning a tuple, the first element is a list with the generated images, and the second element is a
|
542 |
-
list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
|
543 |
-
(nsfw) content, according to the `safety_checker`.
|
544 |
-
"""
|
545 |
-
# 0. Default height and width to unet
|
546 |
batch_size = 4
|
547 |
device = torch.device("cuda:0")
|
548 |
|
@@ -553,7 +402,7 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
553 |
# corresponds to doing no classifier free guidance.
|
554 |
do_classifier_free_guidance = guidance_scale > 1.0
|
555 |
|
556 |
-
#
|
557 |
self.scheduler.set_timesteps(num_inference_steps, device=device)
|
558 |
timesteps = self.scheduler.timesteps
|
559 |
|
@@ -563,26 +412,10 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
563 |
num_images_per_prompt=num_images_per_prompt,
|
564 |
do_classifier_free_guidance=do_classifier_free_guidance,
|
565 |
negative_prompt=negative_prompt,
|
566 |
-
)
|
567 |
prompt_embeds_neg, prompt_embeds_pos = _prompt_embeds.chunk(2)
|
568 |
-
|
569 |
-
_, prompt_embeds_pos_2 = self._encode_prompt(
|
570 |
-
prompt="watermellon",
|
571 |
-
device=device,
|
572 |
-
num_images_per_prompt=num_images_per_prompt,
|
573 |
-
do_classifier_free_guidance=do_classifier_free_guidance,
|
574 |
-
negative_prompt=negative_prompt,
|
575 |
-
).chunk(2) # type: ignore
|
576 |
-
|
577 |
-
_, prompt_embeds_pos_4 = self._encode_prompt(
|
578 |
-
prompt="long hair",
|
579 |
-
device=device,
|
580 |
-
num_images_per_prompt=num_images_per_prompt,
|
581 |
-
do_classifier_free_guidance=do_classifier_free_guidance,
|
582 |
-
negative_prompt=negative_prompt,
|
583 |
-
).chunk(2) # type: ignore
|
584 |
|
585 |
-
#
|
586 |
latents: torch.Tensor = self.prepare_latents(
|
587 |
batch_size * num_images_per_prompt,
|
588 |
4,
|
@@ -594,33 +427,23 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
594 |
None,
|
595 |
)
|
596 |
|
597 |
-
#
|
598 |
extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
|
599 |
|
600 |
-
#
|
601 |
-
num_warmup_steps = len(
|
602 |
-
timesteps) - num_inference_steps * self.scheduler.order
|
603 |
with self.progress_bar(total=num_inference_steps) as progress_bar:
|
604 |
for i, t in enumerate(timesteps):
|
605 |
# expand the latents if we are doing classifier free guidance
|
606 |
multiplier = 2 if do_classifier_free_guidance else 1
|
607 |
latent_model_input = torch.cat([latents] * multiplier)
|
608 |
-
latent_model_input = self.scheduler.scale_model_input(
|
609 |
-
latent_model_input, t)
|
610 |
|
611 |
# predict the noise residual
|
612 |
-
# print(
|
613 |
-
# f"shape of latent_model_input: {latent_model_input.shape}"
|
614 |
-
# ) # [2*4, 4, 32, 32]
|
615 |
-
# print(f"shape of prompt_embeds: {prompt_embeds.shape}"
|
616 |
-
# ) # [2*4, 77, 768]
|
617 |
-
# print(f"shape of camera: {camera.shape}") # [4, 16]
|
618 |
noise_pred = self.unet.forward(
|
619 |
x=latent_model_input,
|
620 |
-
timesteps=torch.tensor([t] * 4 * multiplier,
|
621 |
-
|
622 |
-
context=torch.cat([prompt_embeds_neg] * 4 +
|
623 |
-
[prompt_embeds_pos, prompt_embeds_pos_2, prompt_embeds_pos, prompt_embeds_pos_4]),
|
624 |
num_frames=4,
|
625 |
camera=torch.cat([camera] * multiplier),
|
626 |
)
|
@@ -628,46 +451,29 @@ class MVDreamStableDiffusionPipeline(DiffusionPipeline):
|
|
628 |
# perform guidance
|
629 |
if do_classifier_free_guidance:
|
630 |
noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
|
631 |
-
noise_pred = noise_pred_uncond + guidance_scale * (
|
632 |
-
noise_pred_text - noise_pred_uncond)
|
633 |
|
634 |
# compute the previous noisy sample x_t -> x_t-1
|
635 |
# latents = self.scheduler.step(noise_pred.to(dtype=torch.float32), t, latents.to(dtype=torch.float32)).prev_sample.to(prompt_embeds.dtype)
|
636 |
-
latents: torch.Tensor = self.scheduler.step(
|
637 |
-
noise_pred,
|
638 |
-
t,
|
639 |
-
latents,
|
640 |
-
**extra_step_kwargs,
|
641 |
-
return_dict=False)[0]
|
642 |
|
643 |
# call the callback, if provided
|
644 |
-
if i == len(timesteps) - 1 or (
|
645 |
-
(i + 1) > num_warmup_steps and
|
646 |
-
(i + 1) % self.scheduler.order == 0):
|
647 |
progress_bar.update()
|
648 |
if callback is not None and i % callback_steps == 0:
|
649 |
-
callback(i, t, latents)
|
650 |
|
651 |
-
#
|
652 |
if output_type == "latent":
|
653 |
image = latents
|
654 |
elif output_type == "pil":
|
655 |
-
# 8. Post-processing
|
656 |
image = self.decode_latents(latents)
|
657 |
-
# 10. Convert to PIL
|
658 |
image = self.numpy_to_pil(image)
|
659 |
else:
|
660 |
-
# 8. Post-processing
|
661 |
image = self.decode_latents(latents)
|
662 |
|
663 |
# Offload last model to CPU
|
664 |
-
if hasattr(
|
665 |
-
self,
|
666 |
-
"final_offload_hook") and self.final_offload_hook is not None:
|
667 |
self.final_offload_hook.offload()
|
668 |
|
669 |
-
|
670 |
-
return image
|
671 |
-
|
672 |
-
return StableDiffusionPipelineOutput(images=image,
|
673 |
-
nsfw_content_detected=None)
|
|
|
1 |
+
import torch
|
2 |
+
import numpy as np
|
3 |
import inspect
|
4 |
from typing import Any, Callable, Dict, List, Optional, Union
|
|
|
|
|
5 |
from transformers import CLIPTextModel, CLIPTokenizer
|
|
|
6 |
from diffusers import AutoencoderKL, DiffusionPipeline
|
7 |
from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
|
8 |
from diffusers.utils import (
|
|
|
12 |
logging,
|
13 |
replace_example_docstring,
|
14 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
from diffusers.configuration_utils import FrozenDict
|
|
|
16 |
from diffusers.schedulers import DDIMScheduler
|
17 |
+
try:
|
18 |
+
from diffusers import randn_tensor # old import # type: ignore
|
19 |
+
except ImportError:
|
20 |
+
from diffusers.utils.torch_utils import randn_tensor # new import # type: ignore
|
21 |
|
22 |
+
from models import MultiViewUNetWrapperModel
|
23 |
|
24 |
+
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
|
25 |
|
26 |
def create_camera_to_world_matrix(elevation, azimuth):
|
27 |
elevation = np.radians(elevation)
|
|
|
52 |
def convert_opengl_to_blender(camera_matrix):
|
53 |
if isinstance(camera_matrix, np.ndarray):
|
54 |
# Construct transformation matrix to convert from OpenGL space to Blender space
|
55 |
+
flip_yz = np.array([[1, 0, 0, 0], [0, 0, -1, 0], [0, 1, 0, 0], [0, 0, 0, 1]])
|
|
|
56 |
camera_matrix_blender = np.dot(flip_yz, camera_matrix)
|
57 |
else:
|
58 |
# Construct transformation matrix to convert from OpenGL space to Blender space
|
59 |
+
flip_yz = torch.tensor([[1, 0, 0, 0], [0, 0, -1, 0], [0, 1, 0, 0], [0, 0, 0, 1]])
|
|
|
60 |
if camera_matrix.ndim == 3:
|
61 |
flip_yz = flip_yz.unsqueeze(0)
|
62 |
+
camera_matrix_blender = torch.matmul(flip_yz.to(camera_matrix), camera_matrix)
|
|
|
63 |
return camera_matrix_blender
|
64 |
|
65 |
|
66 |
+
def get_camera(num_frames, elevation=15, azimuth_start=0, azimuth_span=360, blender_coord=True):
|
|
|
|
|
|
|
|
|
67 |
angle_gap = azimuth_span / num_frames
|
68 |
cameras = []
|
69 |
+
for azimuth in np.arange(azimuth_start, azimuth_span + azimuth_start, angle_gap):
|
|
|
70 |
camera_matrix = create_camera_to_world_matrix(elevation, azimuth)
|
71 |
if blender_coord:
|
72 |
camera_matrix = convert_opengl_to_blender(camera_matrix)
|
|
|
86 |
):
|
87 |
super().__init__()
|
88 |
|
89 |
+
if hasattr(scheduler.config, "steps_offset") and scheduler.config.steps_offset != 1: # type: ignore
|
90 |
+
deprecation_message = (f"The configuration file of this scheduler: {scheduler} is outdated. `steps_offset`"
|
91 |
+
f" should be set to 1 instead of {scheduler.config.steps_offset}. Please make sure " # type: ignore
|
92 |
+
"to update the config accordingly as leaving `steps_offset` might led to incorrect results"
|
93 |
+
" in future versions. If you have downloaded this checkpoint from the Hugging Face Hub,"
|
94 |
+
" it would be very nice if you could open a Pull request for the `scheduler/scheduler_config.json`"
|
95 |
+
" file")
|
96 |
+
deprecate("steps_offset!=1", "1.0.0", deprecation_message, standard_warn=False)
|
|
|
|
|
|
|
|
|
|
|
97 |
new_config = dict(scheduler.config)
|
98 |
new_config["steps_offset"] = 1
|
99 |
scheduler._internal_dict = FrozenDict(new_config)
|
100 |
|
101 |
+
if hasattr(scheduler.config, "clip_sample") and scheduler.config.clip_sample is True: # type: ignore
|
102 |
+
deprecation_message = (f"The configuration file of this scheduler: {scheduler} has not set the configuration `clip_sample`."
|
103 |
+
" `clip_sample` should be set to False in the configuration file. Please make sure to update the"
|
104 |
+
" config accordingly as not setting `clip_sample` in the config might lead to incorrect results in"
|
105 |
+
" future versions. If you have downloaded this checkpoint from the Hugging Face Hub, it would be very"
|
106 |
+
" nice if you could open a Pull request for the `scheduler/scheduler_config.json` file")
|
107 |
+
deprecate("clip_sample not set", "1.0.0", deprecation_message, standard_warn=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
108 |
new_config = dict(scheduler.config)
|
109 |
new_config["clip_sample"] = False
|
110 |
scheduler._internal_dict = FrozenDict(new_config)
|
|
|
116 |
tokenizer=tokenizer,
|
117 |
text_encoder=text_encoder,
|
118 |
)
|
119 |
+
self.vae_scale_factor = 2**(len(self.vae.config.block_out_channels) - 1)
|
|
|
120 |
self.register_to_config(requires_safety_checker=False)
|
121 |
|
122 |
def enable_vae_slicing(self):
|
|
|
162 |
if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
|
163 |
from accelerate import cpu_offload
|
164 |
else:
|
165 |
+
raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
|
|
|
|
|
166 |
|
167 |
device = torch.device(f"cuda:{gpu_id}")
|
168 |
|
169 |
if self.device.type != "cpu":
|
170 |
self.to("cpu", silence_dtype_warnings=True)
|
171 |
+
torch.cuda.empty_cache() # otherwise we don't see the memory savings (but they probably exist)
|
|
|
172 |
|
173 |
for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
|
174 |
cpu_offload(cpu_offloaded_model, device)
|
|
|
180 |
method is called, and the model remains in GPU until the next model runs. Memory savings are lower than with
|
181 |
`enable_sequential_cpu_offload`, but performance is much better due to the iterative execution of the `unet`.
|
182 |
"""
|
183 |
+
if is_accelerate_available() and is_accelerate_version(">=", "0.17.0.dev0"):
|
|
|
184 |
from accelerate import cpu_offload_with_hook
|
185 |
else:
|
186 |
+
raise ImportError("`enable_model_offload` requires `accelerate v0.17.0` or higher.")
|
|
|
|
|
187 |
|
188 |
device = torch.device(f"cuda:{gpu_id}")
|
189 |
|
190 |
if self.device.type != "cpu":
|
191 |
self.to("cpu", silence_dtype_warnings=True)
|
192 |
+
torch.cuda.empty_cache() # otherwise we don't see the memory savings (but they probably exist)
|
|
|
193 |
|
194 |
hook = None
|
195 |
for cpu_offloaded_model in [self.text_encoder, self.unet, self.vae]:
|
196 |
+
_, hook = cpu_offload_with_hook(cpu_offloaded_model, device, prev_module_hook=hook)
|
|
|
|
|
197 |
|
198 |
# We'll offload the last model manually.
|
199 |
self.final_offload_hook = hook
|
|
|
208 |
if not hasattr(self.unet, "_hf_hook"):
|
209 |
return self.device
|
210 |
for module in self.unet.modules():
|
211 |
+
if (hasattr(module, "_hf_hook") and hasattr(module._hf_hook, "execution_device") and module._hf_hook.execution_device is not None):
|
|
|
|
|
212 |
return torch.device(module._hf_hook.execution_device)
|
213 |
return self.device
|
214 |
|
|
|
219 |
num_images_per_prompt,
|
220 |
do_classifier_free_guidance: bool,
|
221 |
negative_prompt=None,
|
|
|
|
|
222 |
):
|
223 |
r"""
|
224 |
Encodes the prompt into text encoder hidden states.
|
|
|
249 |
elif prompt is not None and isinstance(prompt, list):
|
250 |
batch_size = len(prompt)
|
251 |
else:
|
252 |
+
raise ValueError(f"`prompt` should be either a string or a list of strings, but got {type(prompt)}.")
|
253 |
+
|
254 |
+
text_inputs = self.tokenizer(
|
255 |
+
prompt,
|
256 |
+
padding="max_length",
|
257 |
+
max_length=self.tokenizer.model_max_length,
|
258 |
+
truncation=True,
|
259 |
+
return_tensors="pt",
|
260 |
+
)
|
261 |
+
text_input_ids = text_inputs.input_ids
|
262 |
+
untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
|
263 |
|
264 |
+
if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
|
265 |
+
removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer.model_max_length - 1:-1])
|
266 |
+
logger.warning("The following part of your input was truncated because CLIP can only handle sequences up to"
|
267 |
+
f" {self.tokenizer.model_max_length} tokens: {removed_text}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
268 |
|
269 |
+
if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
|
270 |
+
attention_mask = text_inputs.attention_mask.to(device)
|
271 |
+
else:
|
272 |
+
attention_mask = None
|
|
|
273 |
|
274 |
+
prompt_embeds = self.text_encoder(
|
275 |
+
text_input_ids.to(device),
|
276 |
+
attention_mask=attention_mask,
|
277 |
+
)
|
278 |
+
prompt_embeds = prompt_embeds[0]
|
279 |
|
280 |
+
prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)
|
|
|
281 |
|
282 |
bs_embed, seq_len, _ = prompt_embeds.shape
|
283 |
# duplicate text embeddings for each generation per prompt, using mps friendly method
|
284 |
prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
|
285 |
+
prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
|
|
|
286 |
|
287 |
# get unconditional embeddings for classifier free guidance
|
288 |
+
if do_classifier_free_guidance:
|
289 |
uncond_tokens: List[str]
|
290 |
if negative_prompt is None:
|
291 |
uncond_tokens = [""] * batch_size
|
292 |
elif type(prompt) is not type(negative_prompt):
|
293 |
+
raise TypeError(f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
|
294 |
+
f" {type(prompt)}.")
|
|
|
295 |
elif isinstance(negative_prompt, str):
|
296 |
uncond_tokens = [negative_prompt]
|
297 |
elif batch_size != len(negative_prompt):
|
298 |
+
raise ValueError(f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
|
299 |
+
f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
|
300 |
+
" the batch size of `prompt`.")
|
|
|
301 |
else:
|
302 |
uncond_tokens = negative_prompt
|
303 |
|
|
|
310 |
return_tensors="pt",
|
311 |
)
|
312 |
|
313 |
+
if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
|
|
|
314 |
attention_mask = uncond_input.attention_mask.to(device)
|
315 |
else:
|
316 |
attention_mask = None
|
|
|
321 |
)
|
322 |
negative_prompt_embeds = negative_prompt_embeds[0]
|
323 |
|
|
|
324 |
# duplicate unconditional embeddings for each generation per prompt, using mps friendly method
|
325 |
seq_len = negative_prompt_embeds.shape[1]
|
326 |
|
327 |
+
negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)
|
|
|
328 |
|
329 |
+
negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
|
330 |
+
negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
|
|
|
|
|
331 |
|
332 |
# For classifier free guidance, we need to do two forward passes.
|
333 |
# Here we concatenate the unconditional and text embeddings into a single batch
|
|
|
350 |
# eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
|
351 |
# and should be between [0, 1]
|
352 |
|
353 |
+
accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
|
|
|
354 |
extra_step_kwargs = {}
|
355 |
if accepts_eta:
|
356 |
extra_step_kwargs["eta"] = eta
|
357 |
|
358 |
# check if the scheduler accepts generator
|
359 |
+
accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
|
|
|
360 |
if accepts_generator:
|
361 |
extra_step_kwargs["generator"] = generator
|
362 |
return extra_step_kwargs
|
363 |
|
364 |
+
def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
|
365 |
+
shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
366 |
if isinstance(generator, list) and len(generator) != batch_size:
|
367 |
+
raise ValueError(f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
|
368 |
+
f" size of {batch_size}. Make sure the batch size matches the length of the generators.")
|
|
|
|
|
369 |
|
370 |
if latents is None:
|
371 |
+
latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
|
|
|
|
|
|
|
372 |
else:
|
373 |
latents = latents.to(device)
|
374 |
|
|
|
377 |
return latents
|
378 |
|
379 |
@torch.no_grad()
|
|
|
380 |
def __call__(
|
381 |
self,
|
382 |
height: int = 256,
|
|
|
387 |
negative_prompt: str = "bad quality",
|
388 |
num_images_per_prompt: int = 1,
|
389 |
eta: float = 0.0,
|
390 |
+
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
|
|
|
391 |
output_type: Optional[str] = "pil",
|
392 |
+
callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
|
|
|
|
|
393 |
callback_steps: int = 1,
|
|
|
|
|
394 |
):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
395 |
batch_size = 4
|
396 |
device = torch.device("cuda:0")
|
397 |
|
|
|
402 |
# corresponds to doing no classifier free guidance.
|
403 |
do_classifier_free_guidance = guidance_scale > 1.0
|
404 |
|
405 |
+
# Prepare timesteps
|
406 |
self.scheduler.set_timesteps(num_inference_steps, device=device)
|
407 |
timesteps = self.scheduler.timesteps
|
408 |
|
|
|
412 |
num_images_per_prompt=num_images_per_prompt,
|
413 |
do_classifier_free_guidance=do_classifier_free_guidance,
|
414 |
negative_prompt=negative_prompt,
|
415 |
+
) # type: ignore
|
416 |
prompt_embeds_neg, prompt_embeds_pos = _prompt_embeds.chunk(2)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
417 |
|
418 |
+
# Prepare latent variables
|
419 |
latents: torch.Tensor = self.prepare_latents(
|
420 |
batch_size * num_images_per_prompt,
|
421 |
4,
|
|
|
427 |
None,
|
428 |
)
|
429 |
|
430 |
+
# Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
|
431 |
extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
|
432 |
|
433 |
+
# Denoising loop
|
434 |
+
num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
|
|
|
435 |
with self.progress_bar(total=num_inference_steps) as progress_bar:
|
436 |
for i, t in enumerate(timesteps):
|
437 |
# expand the latents if we are doing classifier free guidance
|
438 |
multiplier = 2 if do_classifier_free_guidance else 1
|
439 |
latent_model_input = torch.cat([latents] * multiplier)
|
440 |
+
latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
|
|
|
441 |
|
442 |
# predict the noise residual
|
|
|
|
|
|
|
|
|
|
|
|
|
443 |
noise_pred = self.unet.forward(
|
444 |
x=latent_model_input,
|
445 |
+
timesteps=torch.tensor([t] * 4 * multiplier, device=device),
|
446 |
+
context=torch.cat([prompt_embeds_neg] * 4 + [prompt_embeds_pos] * 4),
|
|
|
|
|
447 |
num_frames=4,
|
448 |
camera=torch.cat([camera] * multiplier),
|
449 |
)
|
|
|
451 |
# perform guidance
|
452 |
if do_classifier_free_guidance:
|
453 |
noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
|
454 |
+
noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
|
|
|
455 |
|
456 |
# compute the previous noisy sample x_t -> x_t-1
|
457 |
# latents = self.scheduler.step(noise_pred.to(dtype=torch.float32), t, latents.to(dtype=torch.float32)).prev_sample.to(prompt_embeds.dtype)
|
458 |
+
latents: torch.Tensor = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
|
|
|
|
|
|
|
|
|
|
|
459 |
|
460 |
# call the callback, if provided
|
461 |
+
if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
|
|
|
|
|
462 |
progress_bar.update()
|
463 |
if callback is not None and i % callback_steps == 0:
|
464 |
+
callback(i, t, latents) # type: ignore
|
465 |
|
466 |
+
# Post-processing
|
467 |
if output_type == "latent":
|
468 |
image = latents
|
469 |
elif output_type == "pil":
|
|
|
470 |
image = self.decode_latents(latents)
|
|
|
471 |
image = self.numpy_to_pil(image)
|
472 |
else:
|
|
|
473 |
image = self.decode_latents(latents)
|
474 |
|
475 |
# Offload last model to CPU
|
476 |
+
if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
|
|
|
|
|
477 |
self.final_offload_hook.offload()
|
478 |
|
479 |
+
return image
|
|
|
|
|
|
|
|
vae/diffusion_pytorch_model.bin
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 334716034
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:660d2d3c357697e87aded9b7d821dd726977291c049be64489132cd442ce6477
|
3 |
size 334716034
|