forked from facebookresearch/deit
-
Notifications
You must be signed in to change notification settings - Fork 0
/
models.py
408 lines (341 loc) · 14.9 KB
/
models.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
# Copyright (c) 2015-present, Facebook, Inc.
# All rights reserved.
import torch
import torch.nn as nn
from functools import partial
import numpy as np
from timm.models.vision_transformer import VisionTransformer, _cfg
from timm.models.registry import register_model
from timm.models.layers import trunc_normal_
__all__ = [
'deit_tiny_patch16_224', 'deit_small_patch16_224', 'deit_base_patch16_224',
'deit_tiny_distilled_patch16_224', 'deit_small_distilled_patch16_224',
'deit_base_distilled_patch16_224', 'deit_base_patch16_384',
'deit_base_distilled_patch16_384', 'deit_small_patch16_160',
'masked_deit_small_patch16_224', 'unstructured_masked_deit_small_patch16_224', 'structured_masked_deit_small_patch16_224'
]
try:
from torch import _assert
except ImportError:
def _assert(condition: bool, message: str):
assert condition, message
class StructuredMaskedVisionTransformer(VisionTransformer):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.force_mask = False
self.keep_ratio = None
self.keep_index = None
def set_force_mask(self, force):
self.force_mask = force
def set_keep_ratio(self, kr):
_assert(kr > 0 and kr <= 1.0, f"Keep_ratio must be within (0, 1]")
self.keep_ratio = kr
num_patches = self.patch_embed.num_patches
_assert(kr==0.5, f"keep ration must be 0.5 for now")
keep_index = torch.arange(int(num_patches * self.keep_ratio))
_assert(num_patches == 14*14, f"num_tokens must be 14*14 for now")
index = 0
for i in range(14):
for j in range(14):
if (i+j)%2 == 0:
keep_index[index] = i*14+j
index += 1
if self.dist_token is None:
num_tokens = 1
else:
num_tokens = 2
keep_index = keep_index + num_tokens
keep_index = torch.cat((torch.arange(num_tokens), keep_index))
self.keep_index = keep_index
return keep_index
def forward_features(self, x):
x = self.patch_embed(x)
# x = self.pos_drop(x + self.pos_embed)
if not self.use_learnable_pos_emb:
x = x + self.pos_embed
cls_token = self.cls_token.expand(x.shape[0], -1, -1) # stole cls_tokens impl from Phil Wang, thanks
if self.dist_token is None:
x = torch.cat((cls_token, x), dim=1)
else:
x = torch.cat((cls_token, self.dist_token.expand(x.shape[0], -1, -1), x), dim=1)
if self.use_learnable_pos_emb:
x = x + self.pos_embed
if (self.force_mask or self.training) and self.keep_ratio < 1.0:
x = x[:, self.keep_index]
x = self.blocks(x)
x = self.norm(x)
if self.dist_token is None:
return self.pre_logits(x[:, 0])
else:
return x[:, 0], x[:, 1]
class UnstructuredMaskedVisionTransformer(VisionTransformer):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.force_mask = False
self.keep_ratio = None
self.keep_index = None
def set_force_mask(self, force):
self.force_mask = force
def set_keep_ratio(self, kr, index=None):
_assert(kr > 0 and kr <= 1.0, f"Keep_ratio must be within (0, 1]")
self.keep_ratio = kr
if index is None:
num_patches = self.patch_embed.num_patches
keep_index = torch.randperm(num_patches)[:int(num_patches * self.keep_ratio)]
if self.dist_token is None:
num_tokens = 1
else:
num_tokens = 2
keep_index = keep_index + num_tokens
keep_index = torch.cat((torch.arange(num_tokens), keep_index))
self.keep_index = keep_index
else:
self.keep_index = torch.from_numpy(index).cuda()
return self.keep_index
def forward_features(self, x):
x = self.patch_embed(x)
# x = self.pos_drop(x + self.pos_embed)
if not self.use_learnable_pos_emb:
x = x + self.pos_embed
cls_token = self.cls_token.expand(x.shape[0], -1, -1) # stole cls_tokens impl from Phil Wang, thanks
if self.dist_token is None:
x = torch.cat((cls_token, x), dim=1)
else:
x = torch.cat((cls_token, self.dist_token.expand(x.shape[0], -1, -1), x), dim=1)
if self.use_learnable_pos_emb:
x = x + self.pos_embed
if (self.force_mask or self.training) and self.keep_ratio < 1.0:
x = x[:, self.keep_index]
x = self.blocks(x)
x = self.norm(x)
if self.dist_token is None:
return self.pre_logits(x[:, 0])
else:
return x[:, 0], x[:, 1]
class MaskedVisionTransformer(VisionTransformer):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.force_mask = False
self.keep_ratio = 1.0
def set_force_mask(self, force):
self.force_mask = force
def set_keep_ratio(self, kr):
_assert(kr > 0 and kr <= 1.0, f"Keep_ratio must be within (0, 1]")
self.keep_ratio = kr
def forward_features(self, x):
x = self.patch_embed(x)
# x = self.pos_drop(x + self.pos_embed)
if not self.use_learnable_pos_emb:
x = x + self.pos_embed
cls_token = self.cls_token.expand(x.shape[0], -1, -1) # stole cls_tokens impl from Phil Wang, thanks
if self.dist_token is None:
x = torch.cat((cls_token, x), dim=1)
else:
x = torch.cat((cls_token, self.dist_token.expand(x.shape[0], -1, -1), x), dim=1)
if self.use_learnable_pos_emb:
x = x + self.pos_embed
if (self.force_mask or self.training) and self.keep_ratio < 1.0:
num_patches = self.patch_embed.num_patches
keep_index = torch.randperm(num_patches)[:int(num_patches * self.keep_ratio)]
if self.dist_token is None:
num_tokens = 1
else:
num_tokens = 2
keep_index = keep_index + num_tokens
keep_index = torch.cat((torch.arange(num_tokens), keep_index))
x = x[:, keep_index]
x = self.blocks(x)
x = self.norm(x)
if self.dist_token is None:
return self.pre_logits(x[:, 0])
else:
return x[:, 0], x[:, 1]
class DistilledVisionTransformer(VisionTransformer):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.dist_token = nn.Parameter(torch.zeros(1, 1, self.embed_dim))
num_patches = self.patch_embed.num_patches
self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 2, self.embed_dim))
self.head_dist = nn.Linear(self.embed_dim, self.num_classes) if self.num_classes > 0 else nn.Identity()
trunc_normal_(self.dist_token, std=.02)
trunc_normal_(self.pos_embed, std=.02)
self.head_dist.apply(self._init_weights)
def forward_features(self, x):
# taken from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
# with slight modifications to add the dist_token
B = x.shape[0]
x = self.patch_embed(x)
cls_tokens = self.cls_token.expand(B, -1, -1) # stole cls_tokens impl from Phil Wang, thanks
dist_token = self.dist_token.expand(B, -1, -1)
x = torch.cat((cls_tokens, dist_token, x), dim=1)
x = x + self.pos_embed
x = self.pos_drop(x)
for blk in self.blocks:
x = blk(x)
x = self.norm(x)
return x[:, 0], x[:, 1]
def forward(self, x):
x, x_dist = self.forward_features(x)
x = self.head(x)
x_dist = self.head_dist(x_dist)
if self.training:
return x, x_dist
else:
# during inference, return the average of both classifier predictions
return (x + x_dist) / 2
@register_model
def deit_tiny_patch16_224(pretrained=False, **kwargs):
model = VisionTransformer(
patch_size=16, embed_dim=192, depth=12, num_heads=3, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/deit_tiny_patch16_224-a1311bcf.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model
@register_model
def deit_small_patch16_160(pretrained=False, **kwargs):
model = VisionTransformer(
img_size=160, patch_size=16, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/deit_small_patch16_160-cd65a155.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model
@register_model
def deit_small_patch16_224(pretrained=False, **kwargs):
model = VisionTransformer(
patch_size=16, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/deit_small_patch16_224-cd65a155.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model
@register_model
def masked_deit_small_patch16_224(pretrained=False, **kwargs):
model = MaskedVisionTransformer(
patch_size=16, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/masked_deit_small_patch16_224-cd65a155.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model
@register_model
def unstructured_masked_deit_small_patch16_224(pretrained=False, **kwargs):
model = UnstructuredMaskedVisionTransformer(
patch_size=16, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/unstructured_masked_deit_small_patch16_224-cd65a155.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model
@register_model
def structured_masked_deit_small_patch16_224(pretrained=False, **kwargs):
model = StructuredMaskedVisionTransformer(
patch_size=16, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/structured_masked_deit_small_patch16_224-cd65a155.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model
@register_model
def deit_base_patch16_224(pretrained=False, **kwargs):
model = VisionTransformer(
patch_size=16, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/deit_base_patch16_224-b5f2ef4d.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model
@register_model
def deit_tiny_distilled_patch16_224(pretrained=False, **kwargs):
model = DistilledVisionTransformer(
patch_size=16, embed_dim=192, depth=12, num_heads=3, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/deit_tiny_distilled_patch16_224-b40b3cf7.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model
@register_model
def deit_small_distilled_patch16_224(pretrained=False, **kwargs):
model = DistilledVisionTransformer(
patch_size=16, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/deit_small_distilled_patch16_224-649709d9.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model
@register_model
def deit_base_distilled_patch16_224(pretrained=False, **kwargs):
model = DistilledVisionTransformer(
patch_size=16, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/deit_base_distilled_patch16_224-df68dfff.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model
@register_model
def deit_base_patch16_384(pretrained=False, **kwargs):
model = VisionTransformer(
img_size=384, patch_size=16, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/deit_base_patch16_384-8de9b5d1.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model
@register_model
def deit_base_distilled_patch16_384(pretrained=False, **kwargs):
model = DistilledVisionTransformer(
img_size=384, patch_size=16, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
model.default_cfg = _cfg()
if pretrained:
checkpoint = torch.hub.load_state_dict_from_url(
url="https://dl.fbaipublicfiles.com/deit/deit_base_distilled_patch16_384-d0272ac0.pth",
map_location="cpu", check_hash=True
)
model.load_state_dict(checkpoint["model"])
return model