/
models_simmim_swin.py
218 lines (182 loc) · 8.33 KB
/
models_simmim_swin.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
# 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
# SimMIM: https://github.com/microsoft/SimMIM
# --------------------------------------------------------
from functools import partial
import math
from turtle import pos
import torch
import torch.nn as nn
import torch.nn.functional as F
from timm.models.vision_transformer import Block
from timm.models.layers import DropPath, to_2tuple, trunc_normal_
from util.pos_embed import get_2d_sincos_pos_embed
from einops import rearrange
from models_mae_swin import PatchEmbed, SwinBlock, PatchMerge
class SimMIMSwin(nn.Module):
""" SimMIM with Swin Transformer backbone
"""
def __init__(self, img_size=256, patch_size=4, in_chans=3, stride=16,
embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24],
mlp_ratio=4, window_size=8, # 16 for finetune
posmlp_dim=32,
norm_layer=nn.LayerNorm, norm_pix_loss=False):
super().__init__()
self.embed_dim = embed_dim
self.stride = stride
self.mask_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
# --------------------------------------------------------------------------
# simmim encoder specifics
self.patch_embed = PatchEmbed(img_size, patch_size, in_chans, embed_dim, norm_layer)
patches_resolution = self.patch_embed.patches_resolution
self.patch_size = patch_size
self.embed_h = self.embed_w = int(self.patch_embed.num_patches ** 0.5)
self.patches_resolution = self.patch_embed.patches_resolution
self.num_layers = len(depths)
pos_h = torch.arange(0, self.embed_h)[None, :, None, None].repeat(1, 1, self.embed_w, 1).float()
pos_w = torch.arange(0, self.embed_w)[None, None, :, None].repeat(1, self.embed_h, 1, 1).float()
self.pos_hw = torch.cat((pos_h, pos_w), dim=-1) #(1, H, W, 2)
self.blocks = nn.ModuleList()
for i_layer in range(self.num_layers):
for dep in range(depths[i_layer]):
downsample_flag = (i_layer > 0) and (dep == 0)
layer = SwinBlock(dim=embed_dim*(2**i_layer),
input_resolution=(
patches_resolution[0] // (2**(i_layer)),
patches_resolution[1] // (2**(i_layer))
),
num_heads=num_heads[i_layer],
window_size=window_size,
shift_size=0 if (dep % 2 == 0) else window_size // 2,
mlp_ratio=mlp_ratio, qkv_bias=True, qk_scale=None,
posmlp_dim=posmlp_dim,
drop_path=0.,
downsample=PatchMerge(
patch_size=2,
in_chans=embed_dim*(2**(i_layer - 1)),
embed_dim=embed_dim*(2**i_layer),
norm_layer=norm_layer
) if downsample_flag else None
)
self.blocks.append(layer)
encoder_out_dim = embed_dim*(2**(self.num_layers-1))
self.norm = norm_layer(encoder_out_dim)
# --------------------------------------------------------------------------
# --------------------------------------------------------------------------
# simmim decoder specifics
self.decoder_pred = nn.Linear(encoder_out_dim, 4 * stride**2 * in_chans, bias=True)
self.decoder_shuffle = nn.PixelShuffle(2)
# --------------------------------------------------------------------------
self.norm_pix_loss = norm_pix_loss
self.initialize_weights()
def initialize_weights(self):
# initialization
# timm's trunc_normal_(std=.02) is effectively normal_(std=0.02) as cutoff is too big (2.)
torch.nn.init.normal_(self.mask_token, std=.02)
# initialize nn.Linear and nn.LayerNorm
self.apply(self._init_weights)
# 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]))
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)
elif isinstance(m, nn.Conv2d):
fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
fan_out //= m.groups
m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
if m.bias is not None:
m.bias.data.zero_()
def unpatchify(self, x, stride=16):
"""
x: (N, L, patch_size**2 *3)
imgs: (N, 3, H, W)
"""
p = stride
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 patchify(self, imgs, stride=16):
"""
imgs: (N, 3, H, W)
x: (N, L, patch_size**2 *3)
"""
p = stride
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 forward_encoder(self, x, mask):
N, _, H, W = x.size()
# embed patches
x = self.patch_embed(x)
H, W = H//self.patch_size, W//self.patch_size
L = mask.size(1)
M = int(L**0.5)
scale = self.embed_h // M
mask = mask.reshape(N, M, M)
mask = mask.repeat_interleave(scale, 1).repeat_interleave(scale, 2)
N, L, _ = x.size()
mask_tokens = self.mask_token.expand(N, L, -1)
w = mask.flatten(1).unsqueeze(-1).type_as(mask_tokens)
x = x * (1. - w) + mask_tokens * w
pos_hw = self.pos_hw.to(x.device)
# apply Transformer blocks
for blk in self.blocks:
x, pos_hw = blk(x, pos_hw)
x = self.norm(x)
return x
def forward_decoder(self, x):
x = self.decoder_pred(x) # 4 * p^2*3
N, L, D = x.shape
M = int(L**0.5)
x = self.decoder_shuffle(x.permute(0, 2, 1).reshape(N, D, M, M)).flatten(2)
x = x.permute(0, 2, 1)
return x
def forward_loss(self, imgs, pred, mask):
"""
imgs: [N, 3, H, W]
pred: [N, mask, p*p*3]
mask: [N, L], 0 is keep, 1 is remove,
"""
target = self.patchify(imgs, self.stride)
N, _, D = target.shape
target = target[mask].reshape(N, -1, D)
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 # (N, L, p*p*3)
# add new type
pred = pred[mask].reshape(N, -1, D)
# Following SimMIM, use L1 loss for reconstruction
loss = (pred - target).abs()
loss = loss.mean()
return loss
def forward(self, imgs, mask):
latent = self.forward_encoder(imgs, mask) # returned mask may change
pred = self.forward_decoder(latent) # [N, L, p*p*3]
loss = self.forward_loss(imgs, pred, mask)
return loss, pred, mask
def simmim_swin_tiny_256(**kwargs):
model = SimMIMSwin(
img_size=256, patch_size=4, in_chans=3, stride=16,
embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24],
mlp_ratio=4, window_size=16,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model