-
Notifications
You must be signed in to change notification settings - Fork 4
/
models_DropPos_mae.py
500 lines (403 loc) · 21.3 KB
/
models_DropPos_mae.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
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
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
487
488
489
490
491
492
493
494
495
496
497
498
499
500
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# timm: https://github.com/rwightman/pytorch-image-models/tree/master/timm
# DeiT: https://github.com/facebookresearch/deit
# MAE: https://github.com/facebookresearch/mae
# --------------------------------------------------------
import math
from functools import partial
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from timm.models.vision_transformer import PatchEmbed, Block, DropPath, Mlp
from util.pos_embed import get_2d_sincos_pos_embed
from einops import rearrange
class DropPositionMaskedAutoEncoderViT(nn.Module):
def __init__(self, img_size=224, patch_size=16, in_chans=3,
embed_dim=1024, depth=24, num_heads=16,
decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16,
mlp_ratio=4., norm_layer=nn.LayerNorm, norm_pix_loss=True,
mask_token_type='param', shuffle=False, multi_task=False, conf_ignore=False, attn_guide=False):
super().__init__()
self.norm_pix_loss = norm_pix_loss
self.mask_token_type = mask_token_type
self.shuffle = shuffle
self.multi_task = multi_task
self.conf_ignore = conf_ignore
self.attn_guide = attn_guide
# --------------------------------------------------------------------------
# DropPos encoder specifics
self.embed_dim = embed_dim
self.patch_embed = PatchEmbed(img_size, patch_size, in_chans, embed_dim)
num_patches = self.patch_embed.num_patches
self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim),
requires_grad=False) # fixed sin-cos embedding
# mask token for position
self.mask_pos_token = nn.Parameter(torch.zeros(1, 1, embed_dim),
requires_grad=True if mask_token_type == 'param' else False)
self.blocks = nn.ModuleList([
Block(embed_dim, num_heads, mlp_ratio, qkv_bias=True, qk_scale=None, norm_layer=norm_layer)
for i in range(depth)])
self.norm = norm_layer(embed_dim)
# --------------------------------------------------------------------------
# --------------------------------------------------------------------------
# DropPos decoder specifics (w/o position embedding)
self.decoder_embed_dim = decoder_embed_dim
self.decoder_embed = nn.Linear(embed_dim, decoder_embed_dim, bias=True)
# self.decoder_pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, decoder_embed_dim),
# requires_grad=False) # fixed sin-cos embedding
self.decoder_blocks = nn.ModuleList([
Block(decoder_embed_dim, decoder_num_heads, mlp_ratio, qkv_bias=True, qk_scale=None, norm_layer=norm_layer)
for i in range(decoder_depth)])
self.decoder_norm = norm_layer(decoder_embed_dim)
self.decoder_pred = nn.Linear(decoder_embed_dim, num_patches, bias=True) # decoder to patch
# --------------------------------------------------------------------------
if multi_task:
# --------------------------------------------------------------------------
# MAE decoder specifics (w/ position embedding)
self.aux_decoder_embed = nn.Linear(embed_dim, decoder_embed_dim, bias=True)
self.aux_decoder_pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, decoder_embed_dim),
requires_grad=False) # fixed sin-cos embedding
# mask token for patches
self.mask_token = nn.Parameter(torch.zeros(1, 1, decoder_embed_dim),
requires_grad=True if mask_token_type == 'param' else False)
self.aux_decoder_blocks = nn.ModuleList([
Block(decoder_embed_dim, decoder_num_heads, mlp_ratio, qkv_bias=True, qk_scale=None,
norm_layer=norm_layer)
for i in range(decoder_depth)])
self.aux_decoder_norm = norm_layer(decoder_embed_dim)
self.aux_decoder_pred = nn.Linear(decoder_embed_dim, patch_size ** 2 * in_chans, bias=True) # decoder to patch
# --------------------------------------------------------------------------
# label smoothing for positions
# self._get_label_smoothing_map(num_patches, sigma)
self.initialize_weights()
def initialize_weights(self):
# initialization
# initialize (and freeze) pos_embed by sin-cos embedding
pos_embed = get_2d_sincos_pos_embed(self.pos_embed.shape[-1], int(self.patch_embed.num_patches ** .5),
cls_token=True)
self.pos_embed.data.copy_(torch.from_numpy(pos_embed).float().unsqueeze(0))
if self.multi_task:
decoder_pos_embed = get_2d_sincos_pos_embed(self.aux_decoder_pos_embed.shape[-1],
int(self.patch_embed.num_patches ** .5), cls_token=True)
self.aux_decoder_pos_embed.data.copy_(torch.from_numpy(decoder_pos_embed).float().unsqueeze(0))
torch.nn.init.normal_(self.mask_token, std=.02)
# initialize patch_embed like nn.Linear (instead of nn.Conv2d)
w = self.patch_embed.proj.weight.data
torch.nn.init.xavier_uniform_(w.view([w.shape[0], -1]))
# timm's trunc_normal_(std=.02) is effectively normal_(std=0.02) as cutoff is too big (2.)
torch.nn.init.normal_(self.cls_token, std=.02)
torch.nn.init.normal_(self.mask_pos_token, std=.02)
# initialize nn.Linear and nn.LayerNorm
self.apply(self._init_weights)
def _init_weights(self, m):
if isinstance(m, nn.Linear):
# we use xavier_uniform following official JAX ViT:
torch.nn.init.xavier_uniform_(m.weight)
if isinstance(m, nn.Linear) and m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.LayerNorm):
nn.init.constant_(m.bias, 0)
nn.init.constant_(m.weight, 1.0)
def patchify(self, imgs):
"""
imgs: (N, 3, H, W)
x: (N, L, patch_size**2 *3)
"""
p = self.patch_embed.patch_size[0]
assert imgs.shape[2] == imgs.shape[3] and imgs.shape[2] % p == 0
h = w = imgs.shape[2] // p
x = imgs.reshape(shape=(imgs.shape[0], 3, h, p, w, p))
x = torch.einsum('nchpwq->nhwpqc', x)
x = x.reshape(shape=(imgs.shape[0], h * w, p ** 2 * 3))
# x = rearrange(imgs, 'b c (h p1) (w p2) -> b (h w) (p1 p2 c)', p1=p, p2=p)
return x
def unpatchify(self, x):
"""
x: (N, L, patch_size**2 *3)
imgs: (N, 3, H, W)
"""
p = self.patch_embed.patch_size[0]
h = w = int(x.shape[1] ** .5)
assert h * w == x.shape[1]
x = x.reshape(shape=(x.shape[0], h, w, p, p, 3))
x = torch.einsum('nhwpqc->nchpwq', x)
imgs = x.reshape(shape=(x.shape[0], 3, h * p, h * p))
return imgs
def random_masking(self, x, mask_ratio):
"""
Perform per-sample random masking by per-sample shuffling.
Per-sample shuffling is done by argsort random noise.
x: [N, L, D], sequence
"""
N, L, D = x.shape # batch, length, dim
len_keep = int(L * (1 - mask_ratio))
noise = torch.rand(N, L, device=x.device) # noise in [0, 1]
# sort noise for each sample
ids_shuffle = torch.argsort(noise, dim=1) # ascend: small is keep, large is remove
ids_restore = torch.argsort(ids_shuffle, dim=1)
# keep the first subset
ids_keep = ids_shuffle[:, :len_keep]
# remove the second subset
ids_remove = ids_shuffle[:, len_keep:]
# generate the binary mask: 0 is keep, 1 is remove
mask = torch.ones([N, L], device=x.device)
mask[:, :len_keep] = 0
# unshuffle to get the binary mask
mask = torch.gather(mask, dim=1, index=ids_restore).bool()
return ids_keep, mask, ids_restore, ids_remove
@torch.no_grad()
def get_last_attention(self, x):
# embed patches
x = self.patch_embed(x)
# add pos embed w/o cls token
x = x + self.pos_embed[:, 1:, :]
# append cls token
cls_token = self.cls_token + self.pos_embed[:, :1, :]
cls_tokens = cls_token.expand(x.shape[0], -1, -1)
x = torch.cat((cls_tokens, x), dim=1)
for i, blk in enumerate(self.blocks):
if i < len(self.blocks) - 1:
x = blk(x)
else:
# return attention of the last block, following timm
# y, attn = blk.attn(blk.norm1(x))
x = blk.norm1(x)
B, N, C = x.shape
qkv = blk.attn.qkv(x).reshape(B, N, 3, blk.attn.num_heads, C // blk.attn.num_heads).permute(2, 0, 3, 1, 4)
q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple)
attn = (q @ k.transpose(-2, -1)) * blk.attn.scale
attn = attn.softmax(dim=-1)
attn = blk.attn.attn_drop(attn)
return attn
@torch.no_grad()
def get_feature_similarity(self, x):
# embed patches
x = self.patch_embed(x)
# add pos embed w/o cls token
x = x + self.pos_embed[:, 1:, :]
# append cls token
cls_token = self.cls_token + self.pos_embed[:, :1, :]
cls_tokens = cls_token.expand(x.shape[0], -1, -1)
x = torch.cat((cls_tokens, x), dim=1)
for blk in self.blocks:
x = blk(x)
x = self.norm(x)
cls_tokens = x[:, :1, :]
patch_tokens = x[:, 1:, :]
sim = F.cosine_similarity(cls_tokens, patch_tokens, dim=-1)
return F.softmax(sim / 0.1, dim=-1)
def forward_encoder(self, x, mask_ratio, pos_mask_ratio):
outs = {}
inputs = x.detach().clone()
# embed patches w/o [cls] token
x = self.patch_embed(x)
N, L, D = x.shape
# generate mask
ids_keep, mask, ids_restore, ids_remove = self.random_masking(x, mask_ratio)
outs['mask'], outs['ids_keep'], outs['ids_restore'] = mask, ids_keep, ids_restore
# gather patch embeddings and position embeddings
x = torch.gather(x, dim=1, index=ids_keep.unsqueeze(-1).repeat(1, 1, D))
pos_embed_all = self.pos_embed[:, 1:, :].data.repeat(N, 1, 1) # w/o [cls] token
pos_embed_vis = torch.gather(pos_embed_all, dim=1, index=ids_keep.unsqueeze(-1).repeat(1, 1, D)).detach()
# random masking for position embedding
ids_keep_pos, mask_pos, ids_restore_pos, ids_remove_pos = self.random_masking(x, pos_mask_ratio)
outs['mask_pos'], outs['ids_keep_pos'], outs['ids_restore_pos'] = mask_pos, ids_keep_pos, ids_restore_pos
# gather position embeddings
pos_embed = torch.gather(pos_embed_vis, dim=1, index=ids_keep_pos.unsqueeze(-1).repeat(1, 1, D))
# append mask tokens to position embeddings
mask_pos_length = mask_pos.sum().item()
if self.mask_token_type == 'param':
mask_pos_tokens = self.mask_pos_token.repeat(N, mask_pos_length, 1)
elif self.mask_token_type == 'zeros':
mask_pos_tokens = torch.zeros((N, mask_pos_length, self.embed_dim)).to(x.device)
elif self.mask_token_type == 'wrong_pos':
removed_pos_embed = torch.gather(pos_embed_vis, dim=1, index=ids_remove_pos.unsqueeze(-1).repeat(1, 1, D))
# convert to numpy, since numpy shuffles the first dimension, we have to transpose first
removed_pos_embed = removed_pos_embed.detach().cpu().permute(1, 0, 2).numpy() # [N, L, D] -> [L, N, D]
np.random.shuffle(removed_pos_embed)
# restore to torch
removed_pos_embed = torch.from_numpy(removed_pos_embed).permute(1, 0, 2) # [L, N, D] -> [N, L, D]
mask_pos_tokens = removed_pos_embed.to(x.device)
else:
raise Exception('unknown mask_token_type: {}'.format(self.mask_token_type))
pos_embed = torch.cat([pos_embed, mask_pos_tokens], dim=1)
# restore position embeddings before adding
pos_embed = torch.gather(pos_embed, dim=1, index=ids_restore_pos.unsqueeze(-1).repeat(1, 1, D))
# add position embedding w/o [cls] token
x = x + pos_embed
if self.shuffle:
# generate shuffle indexes first
ids_keep_shuffle, _, ids_restore_shuffle, _ = self.random_masking(x, 0.)
# gather
x = torch.gather(x, dim=1, index=ids_keep_shuffle.unsqueeze(-1).repeat(1, 1, D))
outs['ids_restore_shuffle'] = ids_restore_shuffle
# append cls token
cls_token = self.cls_token + self.pos_embed[:, :1, :]
cls_tokens = cls_token.expand(x.shape[0], -1, -1)
x = torch.cat((cls_tokens, x), dim=1)
# get last self-attention
if self.attn_guide:
# get attentions
# attn = self.get_last_attention(inputs)
# attn = attn[:, :, 0, 1:].mean(1) # [N, num_patches]
# outs['attn_full'] = attn
# get similarities
attn = self.get_feature_similarity(inputs)
outs['attn_full'] = attn
# gather visible patches
attn = torch.gather(attn, dim=1, index=ids_keep)
outs['attn'] = attn / attn.sum(-1, keepdims=True)
# apply Transformer blocks
for blk in self.blocks:
x = blk(x)
outs['x'] = self.norm(x)
return outs
def forward_decoder(self, outs):
x = outs['x']
# --------------------------------------------------------------------------
# DropPos decoder forward
x = self.decoder_embed(x)
for blk in self.decoder_blocks:
x = blk(x)
x = self.decoder_norm(x)
x = self.decoder_pred(x)
x = x[:, 1:, :]
if self.shuffle:
ids_restore_shuffle = outs['ids_restore_shuffle']
x = torch.gather(x, dim=1, index=ids_restore_shuffle.unsqueeze(-1).repeat(1, 1, x.shape[-1]))
# --------------------------------------------------------------------------
# --------------------------------------------------------------------------
# MAE decoder forward
if self.multi_task:
x_mae = outs['x'].clone()
x_mae = self.aux_decoder_embed(x_mae)
# append mask tokens to sequence
mask_tokens = self.mask_token.repeat(x_mae.shape[0], outs['ids_restore'].shape[1] + 1 - x_mae.shape[1], 1)
x_ = torch.cat([x_mae[:, 1:, :], mask_tokens], dim=1) # no cls token
x_ = torch.gather(x_, dim=1, index=outs['ids_restore'].unsqueeze(-1).repeat(1, 1, x_mae.shape[2])) # unshuffle
x_mae = torch.cat([x_mae[:, :1, :], x_], dim=1) # append cls token
# add pos embed
x_mae = x_mae + self.aux_decoder_pos_embed
# apply Transformer blocks
for blk in self.aux_decoder_blocks:
x_mae = blk(x_mae)
x_mae = self.aux_decoder_norm(x_mae)
# predictor projection
x_mae = self.aux_decoder_pred(x_mae)
# remove cls token
x_mae = x_mae[:, 1:, :]
return x, x_mae
# --------------------------------------------------------------------------
return x
def forward_mae_loss(self, imgs, pred, mask):
"""
imgs: [N, 3, H, W]
pred: [N, L, p*p*3]
mask: [N, L], 0 is keep, 1 is remove,
"""
target = self.patchify(imgs)
if self.norm_pix_loss:
mean = target.mean(dim=-1, keepdim=True)
var = target.var(dim=-1, keepdim=True)
target = (target - mean) / (var + 1.e-6) ** .5
loss = (pred - target) ** 2
loss = loss.mean(dim=-1) # [N, L], mean loss per patch
loss = (loss * mask).sum() / mask.sum() # mean loss on removed patches
return loss
def forward_drop_pos_loss(self, pred, mask, ids_keep, mask_pos, smooth, attn):
smooth = smooth.to(pred.device).detach()
N, L = mask.shape
num_vis = pred.shape[1]
labels = torch.arange(L).repeat(N, 1).to(pred.device).detach()
labels = torch.gather(labels, dim=1, index=ids_keep)
labels_smooth = torch.gather(smooth.repeat(N, 1, 1), dim=1, index=ids_keep.unsqueeze(-1).repeat(1, 1, L))
log_prob = F.log_softmax(pred, dim=-1) # [N, x, L]
if self.conf_ignore:
# ignore if confidence > smooth
conf_thresh = torch.diag(smooth).cuda().detach().unsqueeze(0).repeat(N, 1) # [N, L]
conf_thresh = torch.gather(conf_thresh, dim=1, index=ids_keep) # [N, x]
with torch.no_grad():
prob = F.softmax(pred, dim=-1) # [N, x, L]
conf = torch.gather(prob, dim=2, index=ids_keep.unsqueeze(1).repeat(1, num_vis, 1)) # [N, x, x]
conf = torch.diagonal(conf, dim1=1, dim2=2)
conf_mask = (conf < conf_thresh).bool()
mask_pos = mask_pos * conf_mask
if attn is not None:
mask_pos = mask_pos * attn
# loss = criterion(pred.permute(0, 2, 1), labels) * mask_pos
loss = (-labels_smooth * log_prob).sum(-1) * mask_pos
loss = loss.sum() / mask_pos.sum()
# evaluate position acc
with torch.no_grad():
pred_position = torch.argmax(pred.detach(), dim=-1)
acc1 = (pred_position == labels) * mask_pos
acc1 = acc1.sum() / mask_pos.sum()
acc1 = acc1.item()
return acc1, loss
def forward(self, imgs, mask_ratio, pos_mask_ratio, smooth):
outs = self.forward_encoder(imgs, mask_ratio, pos_mask_ratio)
pred = self.forward_decoder(outs)
if self.multi_task:
acc1, loss_drop_pos = self.forward_drop_pos_loss(pred[0], outs['mask'], outs['ids_keep'], outs['mask_pos'], smooth,
attn=outs['attn'] if self.attn_guide else None)
loss_mae = self.forward_mae_loss(imgs, pred[1], outs['mask'])
return acc1, loss_drop_pos, loss_mae
return self.forward_drop_pos_loss(pred, outs['mask'], outs['ids_keep'], outs['mask_pos'], smooth,
attn=outs['attn'] if self.attn_guide else None)
def DropPos_mae_vit_small_patch16_dec512d2b(**kwargs):
model = DropPositionMaskedAutoEncoderViT(
patch_size=16, embed_dim=384, depth=12, num_heads=6,
decoder_embed_dim=512, decoder_depth=2, decoder_num_heads=16,
mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
def DropPos_mae_vit_small_patch16_dec512d8b(**kwargs):
model = DropPositionMaskedAutoEncoderViT(
patch_size=16, embed_dim=384, depth=12, num_heads=6,
decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16,
mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
def DropPos_mae_vit_base_patch16_dec512d2b(**kwargs):
model = DropPositionMaskedAutoEncoderViT(
patch_size=16, embed_dim=768, depth=12, num_heads=12,
decoder_embed_dim=512, decoder_depth=2, decoder_num_heads=16,
mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
def DropPos_mae_vit_base_patch32_dec512d2b(**kwargs):
model = DropPositionMaskedAutoEncoderViT(
patch_size=32, embed_dim=768, depth=12, num_heads=12,
decoder_embed_dim=512, decoder_depth=2, decoder_num_heads=16,
mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
def DropPos_mae_vit_base_patch16_dec512d8b(**kwargs):
model = DropPositionMaskedAutoEncoderViT(
patch_size=16, embed_dim=768, depth=12, num_heads=12,
decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16,
mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
def DropPos_mae_vit_large_patch16_dec512d2b(**kwargs):
model = DropPositionMaskedAutoEncoderViT(
patch_size=16, embed_dim=1024, depth=24, num_heads=16,
decoder_embed_dim=512, decoder_depth=2, decoder_num_heads=16,
mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
def DropPos_mae_vit_large_patch16_dec512d8b(**kwargs):
model = DropPositionMaskedAutoEncoderViT(
patch_size=16, embed_dim=1024, depth=24, num_heads=16,
decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16,
mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
def DropPos_mae_vit_huge_patch14_dec512d8b(**kwargs):
model = DropPositionMaskedAutoEncoderViT(
patch_size=14, embed_dim=1280, depth=32, num_heads=16,
decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16,
mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
# set recommended archs
DropPos_mae_vit_base_patch16 = DropPos_mae_vit_base_patch16_dec512d8b # decoder: 512 dim, 8 blocks
DropPos_mae_vit_large_patch16 = DropPos_mae_vit_large_patch16_dec512d8b # decoder: 512 dim, 8 blocks
DropPos_mae_vit_huge_patch14 = DropPos_mae_vit_huge_patch14_dec512d8b # decoder: 512 dim, 8 blocks