-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodels_mae_meta.py
366 lines (301 loc) · 14.4 KB
/
models_mae_meta.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
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# 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
# MetaFormer: https://github.com/dqshuai/MetaFormer
# --------------------------------------------------------
from functools import partial
import numpy as np
import torch
import torch.nn as nn
from timm.models.layers import trunc_normal_
from timm.models.vision_transformer import Block, PatchEmbed
from util.pos_embed import get_1d_sincos_pos_embed_from_grid, get_2d_sincos_pos_embed
class ResNormLayer(nn.Module):
def __init__(self, linear_size):
super(ResNormLayer, self).__init__()
self.l_size = linear_size
self.nonlin1 = nn.ReLU(inplace=True)
self.nonlin2 = nn.ReLU(inplace=True)
self.norm_fn1 = nn.LayerNorm(self.l_size)
self.norm_fn2 = nn.LayerNorm(self.l_size)
self.w1 = nn.Linear(self.l_size, self.l_size)
self.w2 = nn.Linear(self.l_size, self.l_size)
def forward(self, x):
y = self.w1(x)
y = self.nonlin1(y)
y = self.norm_fn1(y)
y = self.w2(y)
y = self.nonlin2(y)
y = self.norm_fn2(y)
out = x + y
return out
class MaskedAutoencoderViT(nn.Module):
""" Masked Autoencoder with VisionTransformer backbone
"""
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=False,
meta_dims=[4, 3]):
super().__init__()
self.embed_dim = embed_dim
# --------------------------------------------------------------------------
# MAE encoder specifics
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 + len(meta_dims), embed_dim),
requires_grad=False) # fixed sin-cos embedding
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)
# --------------------------------------------------------------------------
# --------------------------------------------------------------------------
# MAE decoder specifics
self.decoder_embed = nn.Linear(embed_dim, decoder_embed_dim, bias=True)
self.mask_token = nn.Parameter(torch.zeros(1, 1, decoder_embed_dim))
self.decoder_pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1 + len(meta_dims), 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, patch_size**2 * in_chans, bias=True) # decoder to patch
# --------------------------------------------------------------------------
self.norm_pix_loss = norm_pix_loss
# --------------------------------------------------------------------------
# add metadata
self.meta_dims = meta_dims
for ind, meta_dim in enumerate(meta_dims):
meta_head = nn.Sequential(
nn.Linear(meta_dim, embed_dim),
nn.ReLU(inplace=True),
nn.LayerNorm(embed_dim),
ResNormLayer(embed_dim),
) if meta_dim > 0 else nn.Identity()
setattr(self, f"meta_{ind+1}_head", meta_head)
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)
pos_embed = torch.from_numpy(pos_embed).float().unsqueeze(0)
meta_pos_embed = get_1d_sincos_pos_embed_from_grid(self.pos_embed.shape[-1],
np.arange(len(self.meta_dims), dtype=np.float32))
meta_pos_embed = torch.from_numpy(meta_pos_embed).float().unsqueeze(0)
pos_embed = torch.cat((pos_embed[:, :1, :], meta_pos_embed, pos_embed[:, 1:, :]), dim=1)
print('pos_embed', pos_embed.shape)
self.pos_embed.data.copy_(pos_embed)
decoder_pos_embed = get_2d_sincos_pos_embed(self.decoder_pos_embed.shape[-1],
int(self.patch_embed.num_patches**.5),
cls_token=True)
decoder_pos_embed = torch.from_numpy(decoder_pos_embed).float().unsqueeze(0)
meta_decoder_pos_embed = get_1d_sincos_pos_embed_from_grid(self.decoder_pos_embed.shape[-1],
np.arange(len(self.meta_dims), dtype=np.float32))
meta_decoder_pos_embed = torch.from_numpy(meta_decoder_pos_embed).float().unsqueeze(0)
decoder_pos_embed = torch.cat(
(decoder_pos_embed[:, :1, :], meta_decoder_pos_embed, decoder_pos_embed[:, 1:, :]), dim=1)
print('decoder_pos_embed', decoder_pos_embed.shape)
self.decoder_pos_embed.data.copy_(decoder_pos_embed)
# 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_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))
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]
x_masked = torch.gather(x, dim=1, index=ids_keep.unsqueeze(-1).repeat(1, 1, D))
# 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)
return x_masked, mask, ids_restore
def forward_encoder(self, x, mask_ratio, meta):
# forward metadata
meta_tokens = []
if len(self.meta_dims) > 1:
metas = torch.split(meta, self.meta_dims, dim=1)
else:
metas = (meta, )
for ind, cur_meta in enumerate(metas):
meta_head = getattr(self, f"meta_{ind+1}_head")
meta = meta_head(cur_meta)
meta = meta.reshape(x.shape[0], -1, self.embed_dim)
meta_tokens.append(meta)
meta_tokens = torch.cat(meta_tokens, dim=1)
meta_tokens += self.pos_embed[:, 1:(1 + len(self.meta_dims)), :]
# forward images
# embed patches
x = self.patch_embed(x)
# add pos embed w/o cls token
x = x + self.pos_embed[:, (1 + len(self.meta_dims)):, :]
# masking: length -> length * mask_ratio
x, mask, ids_restore = self.random_masking(x, mask_ratio)
# 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, meta_tokens, x), dim=1)
# apply Transformer blocks
for blk in self.blocks:
x = blk(x)
x = self.norm(x)
return x, mask, ids_restore
def forward_decoder(self, x, ids_restore):
# embed tokens
x = self.decoder_embed(x)
# append mask tokens to sequence
mask_tokens = self.mask_token.repeat(x.shape[0], ids_restore.shape[1] + 1 + len(self.meta_dims) - x.shape[1], 1)
x_ = torch.cat([x[:, (1 + len(self.meta_dims)):, :], mask_tokens], dim=1) # no cls token or meta token
x_ = torch.gather(x_, dim=1, index=ids_restore.unsqueeze(-1).repeat(1, 1, x.shape[2])) # unshuffle
x = torch.cat([x[:, :(1 + len(self.meta_dims)), :], x_], dim=1) # append cls token and meta token
# add pos embed
x = x + self.decoder_pos_embed
# apply Transformer blocks
for blk in self.decoder_blocks:
x = blk(x)
x = self.decoder_norm(x)
# predictor projection
x = self.decoder_pred(x)
# remove cls token and meta token
x = x[:, (1 + len(self.meta_dims)):, :]
return x
def forward_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(self, imgs, meta, mask_ratio=0.75):
latent, mask, ids_restore = self.forward_encoder(imgs, mask_ratio, meta)
pred = self.forward_decoder(latent, ids_restore) # [N, L, p*p*3]
loss = self.forward_loss(imgs, pred, mask)
return loss, pred, mask
def mae_vit_base_patch16_dec512d8b(**kwargs):
model = MaskedAutoencoderViT(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 mae_vit_large_patch16_dec512d8b(**kwargs):
model = MaskedAutoencoderViT(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 mae_vit_huge_patch14_dec512d8b(**kwargs):
model = MaskedAutoencoderViT(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
mae_vit_base_patch16 = mae_vit_base_patch16_dec512d8b # decoder: 512 dim, 8 blocks
mae_vit_large_patch16 = mae_vit_large_patch16_dec512d8b # decoder: 512 dim, 8 blocks
mae_vit_huge_patch14 = mae_vit_huge_patch14_dec512d8b # decoder: 512 dim, 8 blocks
def mae_vit_base_patch16_dec512d2b(**kwargs):
model = MaskedAutoencoderViT(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