-
Notifications
You must be signed in to change notification settings - Fork 8
/
register.py
194 lines (157 loc) · 6.16 KB
/
register.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
from types import MethodType
import torch
import torch.nn.functional as F
from timm.models.beit import Beit
from timm.models.resnet import ResNet
class Config:
_feat_dim = {
'resnet50': (
(64, 112, 112), (256, 56, 56), (512, 28, 28), (1024, 14, 14), (2048, 7, 7), (2048, None, None)),
'resnet152': (
(64, 112, 112), (256, 56, 56), (512, 28, 28), (1024, 14, 14), (2048, 7, 7), (2048, None, None)),
'swin_small_patch4_window7_224': (
(96, 56, 56), (192, 28, 28), (384, 14, 14), (768, 7, 7), (768, 7, 7), (768, None, None)),
'swin_base_patch4_window7_224': (
(128, 56, 56), (256, 28, 28), (512, 14, 14), (1024, 7, 7), (1024, 7, 7), (1024, None, None)),
'swin_large_patch4_window7_224': (
(192, 56, 56), (384, 28, 28), (768, 14, 14), (1536, 7, 7), (1536, 7, 7), (1536, None, None)),
'beitv2_large_patch16_224': (
(64, 56, 56), (64, 56, 56), (256, 28, 28), (1024, 14, 14), (1024, 7, 7), (1024, None, None)),
'bit_r152x2': (
(128, 112, 112), (512, 56, 56), (1024, 28, 28), (2048, 14, 14), (4096, 7, 7), (4096, 1, 1)),
}
_kd_feat_index = {
'resnet50': (1, 2, 3, 4),
'resnet152': (1, 2, 3, 4),
'swin_small_patch4_window7_224': (0, 1, 2, 4),
'swin_base_patch4_window7_224': (0, 1, 2, 4),
'swin_large_patch4_window7_224': (0, 1, 2, 4),
'beitv2_large_patch16_224': (1, 2, 3, 4),
'bit_r152x2': (1, 2, 3, 4),
}
def get_pre_logit_dim(self, model):
feat_sizes = self._feat_dim[model]
if isinstance(feat_sizes, tuple):
return feat_sizes[-1][0]
else:
return feat_sizes
def get_used_feature_index(self, model):
index = self._kd_feat_index[model]
if index is None:
raise NotImplementedError(f'undefined feature kd for model {model}')
return index
def get_feature_size_by_index(self, model, index):
valid_index = self.get_used_feature_index(model)
feat_sizes = self._feat_dim[model]
assert index in valid_index
return feat_sizes[index]
config = Config()
def register_forward(model): # only resnet have implemented pre_act feat
if isinstance(model, ResNet): # ResNet
model.forward = MethodType(ResNet_forward, model)
model.forward_features = MethodType(ResNet_forward_features, model)
elif isinstance(model, Beit): # Beit
model.forward = MethodType(Beitv2_forward, model)
model.forward_features = MethodType(Beitv2_forward_features, model)
else:
raise NotImplementedError('undefined forward method to get feature, check the exp setting carefully!')
def _unpatchify(x, p, remove_token=0):
"""
x: (N, L, patch_size**2 *C)
imgs: (N, C, H, W)
"""
# p = self.patch_embed.patch_size[0]
x = x[:, remove_token:, :]
h = w = int(x.shape[1] ** .5)
assert h * w == x.shape[1]
x = x.reshape(shape=(x.shape[0], h, w, p, p, -1))
x = torch.einsum('nhwpqc->nchpwq', x)
imgs = x.reshape(shape=(x.shape[0], -1, h * p, h * p))
return imgs
# ResNet
def bottleneck_forward(self, x):
shortcut = x
x = self.conv1(x)
x = self.bn1(x)
x = self.act1(x)
x = self.conv2(x)
x = self.bn2(x)
x = self.drop_block(x)
x = self.act2(x)
x = self.aa(x)
x = self.conv3(x)
x = self.bn3(x)
if self.se is not None:
x = self.se(x)
if self.drop_path is not None:
x = self.drop_path(x)
if self.downsample is not None:
shortcut = self.downsample(shortcut)
x += shortcut
pre_act_x = x
x = self.act3(pre_act_x)
return x, pre_act_x
def ResNet_forward_features(self, x, requires_feat):
pre_act_feat = []
feat = []
x = self.conv1(x)
x = self.bn1(x)
pre_act_feat.append(x)
x = self.act1(x)
feat.append(x)
x = self.maxpool(x)
for layer in [self.layer1, self.layer2, self.layer3, self.layer4]:
for bottleneck in layer:
x, pre_act_x = bottleneck_forward(bottleneck, x)
pre_act_feat.append(pre_act_x)
feat.append(x)
return (x, (pre_act_feat, feat)) if requires_feat else x
def ResNet_forward(self, x, requires_feat=False):
if requires_feat:
x, (pre_act_feat, feat) = self.forward_features(x, requires_feat=True)
x = self.forward_head(x, pre_logits=True)
feat.append(x)
pre_act_feat.append(x)
x = self.fc(x)
return x, (pre_act_feat, feat)
else:
x = self.forward_features(x, requires_feat=False)
x = self.forward_head(x)
return x
def Beitv2_forward_features(self, x, requires_feat):
x = self.patch_embed(x)
x = torch.cat((self.cls_token.expand(x.shape[0], -1, -1), x), dim=1)
if self.pos_embed is not None:
x = x + self.pos_embed
x = self.pos_drop(x)
pre_act_feat = [_unpatchify(x, 4, 1)] # stem
feat = [_unpatchify(x, 4, 1)] # stem
rel_pos_bias = self.rel_pos_bias() if self.rel_pos_bias is not None else None
for i, blk in enumerate(self.blocks): # fixme: curremt for beitv2L only
x = blk(x, shared_rel_pos_bias=rel_pos_bias)
f = None
if i == 1:
f = _unpatchify(x, 4, 1)
elif i == 3:
f = _unpatchify(x, 2, 1)
elif i == 21:
f = _unpatchify(x, 1, 1)
elif i == 23:
f = F.adaptive_avg_pool2d(_unpatchify(x, 1, 1), (7, 7))
if f is not None:
pre_act_feat.append(f)
feat.append(f)
x = self.norm(x)
return (x, (pre_act_feat, feat)) if requires_feat else x
def Beitv2_forward(self, x, requires_feat=False):
if requires_feat:
x, (pre_act_feat, feat) = self.forward_features(x, requires_feat=True)
x = self.forward_head(x, pre_logits=True)
feat.append(x)
pre_act_feat.append(x)
x = self.head(x)
return x, (pre_act_feat, feat)
else:
x = self.forward_features(x, requires_feat=False)
x = self.forward_head(x)
return x