-
Notifications
You must be signed in to change notification settings - Fork 1
/
backbone.py
276 lines (215 loc) · 10.4 KB
/
backbone.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
#%%
import torch
import torch.nn as nn
import torch.nn.functional as F
import math
from functools import partial
from bricks import DWConv, StochasticDepth
'''
Mix Transformer Encoder
'''
class Mlp(nn.Module):
'''
Implements Eq 3 in SegFormer Paper
'''
def __init__(self, inFeats, hidFeats=None, outFeats=None, act_layer = nn.GELU, drop=0.0):
super(Mlp, self).__init__()
outFeats = outFeats or inFeats
hidFeats = hidFeats or inFeats
self.fc1 = nn.Linear(inFeats, hidFeats)
self.dwconv = DWConv(hidFeats)
self.act = act_layer()
self.fc2 = nn.Linear(hidFeats, outFeats)
self.drop = nn.Dropout(drop)
def forward(self, x, H, W):
x = self.fc1(x)
x = self.dwconv(x, H, W)
x = self.act(x)
x = self.drop(x)
x = self.fc2(x)
x = self.drop(x)
return x
class Attention(nn.Module):
def __init__(self, dim, num_heads=8, qkv_bias=False, qkv_scale=None, attn_drop=0.0, proj_drop=0.0, sr_ratio=1):
super(Attention, self).__init__()
self.dim = dim
self.num_heads = num_heads
head_dim = self.dim // self.num_heads
self.scale = qkv_scale or head_dim ** -0.5 # as in Eq 1 -> (1/(d_head)**1/2)
self.q = nn.Linear(dim, dim, bias=qkv_bias)
self.kv = nn.Linear(dim, 2*dim, bias=qkv_bias)
self.attn_drop = nn.Dropout(attn_drop)
self.proj = nn.Linear(dim, dim)
self.proj_drop = nn.Dropout(proj_drop)
self.sr_ratio = sr_ratio
if self.sr_ratio > 1:
self.sr = nn.Conv2d(dim, dim, kernel_size=sr_ratio, stride=sr_ratio)
self.norm = nn.LayerNorm(dim)
def forward(self, x, H, W):
B, N, C = x.shape # BxNxC
q = self.q(x).reshape(B, N, self.num_heads, C//self.num_heads).permute(0, 2, 1, 3) # BxNxhx(C/h) -> BxhxNx(C/h)
if self.sr_ratio > 1: # reduction ratio in EQ 2.
x_ = x.permute(0,2,1).reshape(B, C, H, W)
x_ = self.sr(x_).reshape(B, C, -1).permute(0,2,1) # BxCxHxW -> BxCxN -> BxNxC
x_ = self.norm(x_)
kv = self.kv(x_).reshape(B, -1, 2, self.num_heads, C//self.num_heads).permute(2, 0, 3, 1, 4)
# BxNxC -> Bx-1x2xhx(C/h) -> 2xBxhx-1x(C/h) i.e. seprating Keys and values BxhxNxC/h
else:
kv = self.kv(x).reshape(B, -1, 2, self.num_heads, C//self.num_heads).permute(2, 0, 3, 1, 4)
k, v = kv[0], kv[1]
attn = (q @ k.transpose(-2,-1)) * self.scale # EQ 1 in paper
attn = attn.softmax(dim=-1)
attn = self.attn_drop(attn)
x = (attn @ v).transpose(1,2).reshape(B, N, C)
x = self.proj(x)
x = self.proj_drop(x)
return x
class Block(nn.Module):
def __init__(self, dim, num_heads, mlp_ratio=4.0, qkv_bias=False, qk_scale=None, drop=0.0, attn_drop=0.0,
drop_path=0., sr_ratio=1, act_layer=nn.GELU, norm_layer=nn.LayerNorm):
super(Block, self).__init__()
self.norm1 = norm_layer(dim)
self.attn = Attention(dim, num_heads=num_heads, qkv_bias=qkv_bias, qkv_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop, sr_ratio=sr_ratio)
self.drop_path = StochasticDepth(p=drop_path, mode='batch') if drop_path > 0. else nn.Identity()
self.norm2 = norm_layer(dim)
ml_hid_dim = int(dim * mlp_ratio)
self.mlp = Mlp(dim, ml_hid_dim, act_layer=act_layer, drop=drop)
def forward(self, x, H, W):
x = x + self.drop_path(self.attn(self.norm1(x), H, W))
x = x + self.drop_path(self.mlp(self.norm2(x), H, W))
return x
class OverlapPatchEmbed(nn.Module):
def __init__(self, img_size=224, patch_size=7, stride=4, inChannels=3, embed_dim=768):
super(OverlapPatchEmbed, self).__init__()
img_size = (img_size, img_size)
patch_size = (patch_size, patch_size)
self.img_size = img_size
self.patch_size = patch_size
self.H, self.W = img_size[0] // patch_size[0], img_size[1] // patch_size[1]
self.num_patches = self.H * self.W
self.proj = nn.Conv2d(inChannels, embed_dim, kernel_size=patch_size, stride=stride, padding=(patch_size[0]//2, patch_size[1]//2))
self.norm = nn.LayerNorm(embed_dim)
def forward(self, x):
x = self.proj(x)
_, _, H, W = x.shape
x = x.flatten(2).transpose(1,2) # B*C*H*W -> B*C*HW -> B*HW*C
x = self.norm(x)
return x, H, W
class MixVisionTransformer(nn.ModuleDict):
def __init__(self, img_size=224, patch_size=16, inChannels=3, num_classes=1000, embed_dims=[64, 128, 256, 512],
num_heads=[1,2,4,8], mlp_ratio=[4,4,4,4], qkv_bias=False, qk_scale=None, drop_rate=0.0, attn_drop_rate=0.0,
drop_path_rate=0.0, depths=[3,4,6,3], sr_ratio=[8,4,2,1], norm_layer=nn.LayerNorm):
super(MixVisionTransformer, self).__init__()
self.num_classes = num_classes
self.depths = depths
self.patch_embed1 = OverlapPatchEmbed(img_size=img_size, patch_size=7, stride=4, inChannels=inChannels, embed_dim=embed_dims[0])
self.patch_embed2 = OverlapPatchEmbed(img_size=img_size // 4, patch_size=3, stride=2, inChannels=embed_dims[0], embed_dim=embed_dims[1])
self.patch_embed3 = OverlapPatchEmbed(img_size=img_size // 8, patch_size=3, stride=2, inChannels=embed_dims[1], embed_dim=embed_dims[2])
self.patch_embed4 = OverlapPatchEmbed(img_size=img_size // 16, patch_size=3, stride=2, inChannels=embed_dims[2], embed_dim=embed_dims[3])
# stochastic depth decay rule (similar to linear decay) / just like matplot linspace
dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]
cur = 0
self.block1 = nn.ModuleList([Block(
dim=embed_dims[0], num_heads=num_heads[0], mlp_ratio=mlp_ratio[0], qkv_bias=qkv_bias, qk_scale=qk_scale, drop=drop_rate,
attn_drop=attn_drop_rate, drop_path=dpr[cur+i], sr_ratio=sr_ratio[0], norm_layer=norm_layer
) for i in range(depths[0])])
self.norm1 = norm_layer(embed_dims[0])
cur += depths[0]
self.block2 = nn.ModuleList([Block(
dim=embed_dims[1], num_heads=num_heads[1], mlp_ratio=mlp_ratio[1], qkv_bias=qkv_bias, qk_scale=qk_scale, drop=drop_rate,
attn_drop=attn_drop_rate, drop_path=dpr[cur+i], sr_ratio=sr_ratio[1], norm_layer=norm_layer
) for i in range(depths[1])])
self.norm2 = norm_layer(embed_dims[1])
cur += depths[1]
self.block3 = nn.ModuleList([Block(
dim=embed_dims[2], num_heads=num_heads[2], mlp_ratio=mlp_ratio[2], qkv_bias=qkv_bias, qk_scale=qk_scale, drop=drop_rate,
attn_drop=attn_drop_rate, drop_path=dpr[cur+i], sr_ratio=sr_ratio[2], norm_layer=norm_layer
) for i in range(depths[2])])
self.norm3 = norm_layer(embed_dims[2])
cur += depths[2]
self.block4 = nn.ModuleList([Block(
dim=embed_dims[3], num_heads=num_heads[3], mlp_ratio=mlp_ratio[3], qkv_bias=qkv_bias, qk_scale=qk_scale, drop=drop_rate,
attn_drop=attn_drop_rate, drop_path=dpr[cur+i], sr_ratio=sr_ratio[3], norm_layer=norm_layer
) for i in range(depths[3])])
self.norm4 = norm_layer(embed_dims[3])
def forward_features(self, x):
B = x.shape[0]
outs = []
# stage 1
x, H, W = self.patch_embed1(x)
for i, blk in enumerate(self.block1):
x = blk(x, H, W)
x = self.norm1(x)
x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous()
outs.append(x)
# stage 2
x, H, W = self.patch_embed2(x)
for i, blk in enumerate(self.block2):
x = blk(x, H, W)
x = self.norm2(x)
x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous()
outs.append(x)
# stage 3
x, H, W = self.patch_embed3(x)
for i, blk in enumerate(self.block3):
x = blk(x, H, W)
x = self.norm3(x)
x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous()
outs.append(x)
# stage 4
x, H, W = self.patch_embed4(x)
for i, blk in enumerate(self.block4):
x = blk(x, H, W)
x = self.norm4(x)
x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous()
outs.append(x)
return outs
def forward(self, x):
x = self.forward_features(x)
# x = self.head(x)
return x
#%%
from torchsummary import summary
model = MixVisionTransformer(patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratio=[4, 4, 4, 4],
qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 6, 40, 3], sr_ratio=[8, 4, 2, 1],
drop_rate=0.0, drop_path_rate=0.1)
# # summary(model, (3,224,224), depth=2)
x = torch.randn((1,3,224,224))
y = model.forward(x)
# for i in range(len(y)):
# print(y[i].shape)
#%%
# ckpt = '/home/user01/data/talha/segformer/chkpts/mit_b5.pth'
# def load_pretrained_chkpt(model, pretrained_path=None):
# if pretrained_path is not None:
# chkpt = torch.load(pretrained_path,
# map_location='cuda' if torch.cuda.is_available() else 'cpu')
# try:
# # load pretrained
# pretrained_dict = chkpt#['model_state_dict']
# # load model state dict
# state = model.state_dict()
# # loop over both dicts and make a new dict where name and the shape of new state match
# # with the pretrained state dict.
# matched, unmatched = [], []
# new_dict = {}
# for i, j in zip(pretrained_dict.items(), state.items()):
# pk, pv = i # pretrained state dictionary
# nk, nv = j # new state dictionary
# # if name and weight shape are same
# if pk.strip('module.') == nk.strip('module.') and pv.shape == nv.shape:
# new_dict[nk] = pv
# matched.append(pk)
# else:
# unmatched.append(pk)
# state.update(new_dict)
# model.load_state_dict(state)
# print('Pre-trained state loaded successfully...')
# print(f'Mathed kyes: {len(matched)}, Unmatched Keys: {len(unmatched)}')
# except:
# print(f'ERROR in pretrained_dict @ {pretrained_path}')
# else:
# print('Enter pretrained_dict path.')
# return matched, unmatched
# matched, unmatched = load_pretrained_chkpt(model, pretrained_path=ckpt)
# print(unmatched)