-
Notifications
You must be signed in to change notification settings - Fork 10
/
CTformer.py
197 lines (159 loc) · 8.53 KB
/
CTformer.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
"""
CTformer
"""
import torch
import torch.nn as nn
import torch.nn.functional as F
from einops import rearrange
from timm.models.helpers import load_pretrained
from timm.models.registry import register_model
from timm.models.layers import trunc_normal_
import numpy as np
from token_transformer import Token_transformer
from token_performer import Token_performer
from T2T_transformer_block import Block, get_sinusoid_encoding
class MultiHeadDense(nn.Module):
def __init__(self, in_ch, out_ch):
super(MultiHeadDense, self).__init__()
self.weight = nn.Parameter(torch.Tensor(in_ch, out_ch))
def forward(self, x):
# x:[b, h*w, d]
# x = torch.bmm(x, self.weight)
x = F.linear(x, self.weight)
return x
class T2T_module(nn.Module):
"""
CTformer encoding module
"""
def __init__(self, img_size=64, tokens_type='performer', in_chans=1, embed_dim=256, token_dim=64, kernel=32, stride=32):
super().__init__()
if tokens_type == 'transformer':
print('adopt transformer encoder for tokens-to-token')
self.soft_split0 = nn.Unfold(kernel_size=(7, 7), stride=(2, 2))
self.soft_split1 = nn.Unfold(kernel_size=(3, 3), stride=(1, 1),dilation=(2,2))
self.soft_split2 = nn.Unfold(kernel_size=(3, 3), stride=(1, 1))
self.attention1 = Token_transformer(dim=in_chans*7*7, in_dim=token_dim, num_heads=1, mlp_ratio=1.0)
self.attention2 = Token_transformer(dim=token_dim*3*3, in_dim=token_dim, num_heads=1, mlp_ratio=1.0)
self.project = nn.Linear(token_dim * 3 * 3, embed_dim)
elif tokens_type == 'performer':
#print('adopt performer encoder for tokens-to-token')
self.soft_split0 = nn.Unfold(kernel_size=(7, 7), stride=(2, 2))
self.soft_split1 = nn.Unfold(kernel_size=(3, 3), stride=(1, 1),dilation=(2,2))
self.soft_split2 = nn.Unfold(kernel_size=(3, 3), stride=(1, 1))
self.attention1 = Token_performer(dim=in_chans*7*7, in_dim=token_dim, kernel_ratio=0.5)
self.attention2 = Token_performer(dim=token_dim*3*3, in_dim=token_dim, kernel_ratio=0.5)
self.project = nn.Linear(token_dim * 3 * 3, embed_dim)
#self.num_patches = (img_size // (1 * 2 * 2)) * (img_size // (1 * 2 * 2)) # there are 3 sfot split, stride are 4,2,2 seperately
self.num_patches = 529 ## calculate myself
def forward(self, x):
# Tokenization
x = self.soft_split0(x) ## [1, 128, 64, 128])
# CTformer module A
x = self.attention1(x.transpose(1, 2))
res_11 = x
B, new_HW, C = x.shape
x = x.transpose(1,2).reshape(B, C, int(np.sqrt(new_HW)), int(np.sqrt(new_HW)))
x = torch.roll(x, shifts=(2, 2), dims=(2, 3)) ## shift some position
x = self.soft_split1(x)
# CTformer module B
x = self.attention2(x.transpose(1, 2))
res_22 = x
B, new_HW, C = x.shape
x = x.transpose(1, 2).reshape(B, C, int(np.sqrt(new_HW)), int(np.sqrt(new_HW)))
x = torch.roll(x, shifts=(2, 2), dims=(2, 3)) ## shift back position
x = self.soft_split2(x)
x = self.project(x.transpose(1, 2)) ## no projection
return x,res_11,res_22 #,res0,res2
class Token_back_Image(nn.Module):
"""
CTformer decoding module
"""
def __init__(self, img_size=64, tokens_type='performer', in_chans=1, embed_dim=256, token_dim=64, kernel=32, stride=32):
super().__init__()
if tokens_type == 'transformer':
print('adopt transformer encoder for tokens-to-token')
self.soft_split0 = nn.Fold((64,64),kernel_size=(7, 7), stride=(2, 2))
self.soft_split1 = nn.Fold((29,29),kernel_size=(3, 3), stride=(1, 1),dilation=(2,2))
self.soft_split2 = nn.Fold((25,25),kernel_size=(3, 3), stride=(1, 1))
self.attention1 = Token_transformer(dim=token_dim, in_dim=in_chans*7*7, num_heads=1, mlp_ratio=1.0)
self.attention2 = Token_transformer(dim=token_dim, in_dim=token_dim*3*3, num_heads=1, mlp_ratio=1.0)
self.project = nn.Linear(embed_dim,token_dim * 3 * 3)
elif tokens_type == 'performer':
#print('adopt performer encoder for tokens-to-token')
self.soft_split0 = nn.Fold((64,64),kernel_size=(7, 7), stride=(2, 2))
self.soft_split1 = nn.Fold((29,29),kernel_size=(3, 3), stride=(1, 1),dilation=(2,2))
self.soft_split2 = nn.Fold((25,25),kernel_size=(3, 3), stride=(1, 1))
self.attention1 = Token_performer(dim=token_dim, in_dim=in_chans*7*7, kernel_ratio=0.5)
self.attention2 = Token_performer(dim=token_dim, in_dim=token_dim*3*3, kernel_ratio=0.5)
self.project = nn.Linear(embed_dim,token_dim * 3 * 3)
self.num_patches = (img_size // (1 * 2 * 2)) * (img_size // (1 * 2 * 2)) # there are 3 sfot split, stride are 4,2,2 seperately
def forward(self, x, res_11,res_22):
x = self.project(x).transpose(1, 2)
# CTformer module C
x = self.soft_split2(x)
x = torch.roll(x, shifts=(-2, -2), dims=(-1, -2))
x = rearrange(x,'b c h w -> b c (h w)').transpose(1,2)
x = x + res_22
x = self.attention2(x).transpose(1, 2)
# CTformer module D
x = self.soft_split1(x)
x = torch.roll(x, shifts=(-2, -2), dims=(-1, -2))
x = rearrange(x,'b c h w -> b c (h w)').transpose(1,2)
x = x + res_11
x = self.attention1(x).transpose(1, 2)
# Detokenization
x = self.soft_split0(x)
return x
class CTformer(nn.Module):
def __init__(self, img_size=512, tokens_type='convolution', in_chans=1, num_classes=1000, embed_dim=768, depth=12, ## transformer depth 12
num_heads=12, kernel=32, stride=32, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop_rate=0.1, attn_drop_rate=0.1,
drop_path_rate=0.1, norm_layer=nn.LayerNorm, token_dim=1024):
super().__init__()
self.num_classes = num_classes
self.num_features = self.embed_dim = embed_dim # num_features for consistency with other models
self.tokens_to_token = T2T_module( ## use module 2
img_size=img_size, tokens_type=tokens_type, in_chans=in_chans, embed_dim=embed_dim, token_dim=token_dim,kernel=kernel, stride=stride)
num_patches = self.tokens_to_token.num_patches
self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
self.pos_embed = nn.Parameter(data=get_sinusoid_encoding(n_position=num_patches, d_hid=embed_dim), requires_grad=False)
self.pos_drop = nn.Dropout(p=drop_rate)
dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)] # stochastic depth decay rule
self.blocks = nn.ModuleList([
Block(
dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer)
for i in range(depth)])
self.norm = norm_layer(embed_dim)
# CTformer decoder
self.dconv1 = Token_back_Image(img_size=img_size, tokens_type=tokens_type, in_chans=in_chans, embed_dim=embed_dim, token_dim=token_dim, kernel=kernel, stride=stride)
self.head = nn.Linear(embed_dim, num_classes) if num_classes > 0 else nn.Identity()
trunc_normal_(self.cls_token, std=.02)
self.apply(self._init_weights)
def _init_weights(self, m):
if isinstance(m, nn.Linear):
trunc_normal_(m.weight, std=.02)
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)
@torch.jit.ignore
def no_weight_decay(self):
return {'cls_token'}
def get_classifier(self):
return self.head
def reset_classifier(self, num_classes, global_pool=''):
self.num_classes = num_classes
self.head = nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity()
def forward(self, x):
res1 = x
x, res_11, res_22 = self.tokens_to_token(x)
x = x + self.pos_embed
x = self.pos_drop(x)
i = 0
for blk in self.blocks: ## only one intermediate transformer block
i += 1
x = blk(x)
x = self.norm(x) #+ res_0 ## do not use 0,2,4
out = res1 - self.dconv1(x,res_11,res_22)
return out