-
Notifications
You must be signed in to change notification settings - Fork 19
/
Copy pathmodel.py
198 lines (156 loc) · 9.51 KB
/
model.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
#-------------------------------------------------------------------#
# Author : 章鑫
# Date : 2023-10-11 20:22:56
# LastEditTime : 2023-10-12 17:02:50
# Description : RFAConv系列
#-------------------------------------------------------------------#
import torch
from torch import nn
from einops import rearrange
class RFAConv(nn.Module): # 基于Group Conv实现的RFAConv
def __init__(self,in_channel,out_channel,kernel_size,stride=1):
super().__init__()
self.kernel_size = kernel_size
self.get_weight = nn.Sequential(nn.AvgPool2d(kernel_size=kernel_size, padding=kernel_size // 2, stride=stride),
nn.Conv2d(in_channel, in_channel * (kernel_size ** 2), kernel_size=1, groups=in_channel,bias=False))
self.generate_feature = nn.Sequential(
nn.Conv2d(in_channel, in_channel * (kernel_size ** 2), kernel_size=kernel_size,padding=kernel_size//2,stride=stride, groups=in_channel, bias=False),
nn.BatchNorm2d(in_channel * (kernel_size ** 2)),
nn.ReLU())
self.conv = nn.Sequential(nn.Conv2d(in_channel, out_channel, kernel_size=kernel_size, stride=kernel_size),
nn.BatchNorm2d(out_channel),
nn.ReLU())
def forward(self,x):
b,c = x.shape[0:2]
weight = self.get_weight(x)
h,w = weight.shape[2:]
weighted = weight.view(b, c, self.kernel_size ** 2, h, w).softmax(2) # b c*kernel**2,h,w -> b c k**2 h w
feature = self.generate_feature(x).view(b, c, self.kernel_size ** 2, h, w) #b c*kernel**2,h,w -> b c k**2 h w 获得感受野空间特征
weighted_data = feature * weighted
conv_data = rearrange(weighted_data, 'b c (n1 n2) h w -> b c (h n1) (w n2)', n1=self.kernel_size, # b c k**2 h w -> b c h*k w*k
n2=self.kernel_size)
return self.conv(conv_data)
class RFAConv(nn.Module): # 基于Unfold实现的RFAConv
def __init__(self, in_channel, out_channel, kernel_size=3):
super().__init__()
self.kernel_size = kernel_size
self.unfold = nn.Unfold(kernel_size=(kernel_size, kernel_size), padding=kernel_size // 2)
self.get_weights = nn.Sequential(
nn.Conv2d(in_channel * (kernel_size ** 2), in_channel * (kernel_size ** 2), kernel_size=1,
groups=in_channel),
nn.BatchNorm2d(in_channel * (kernel_size ** 2)))
self.conv = nn.Conv2d(in_channel, out_channel, kernel_size=kernel_size, padding=0, stride=kernel_size)
self.bn = nn.BatchNorm2d(out_channel)
self.act = nn.ReLU()
def forward(self, x):
b, c, h, w = x.shape
unfold_feature = self.unfold(x) # 获得感受野空间特征 b c*kernel**2,h*w
x = unfold_feature
data = unfold_feature.unsqueeze(-1)
weight = self.get_weights(data).view(b, c, self.kernel_size ** 2, h, w).permute(0, 1, 3, 4, 2).softmax(-1)
weight_out = rearrange(weight, 'b c h w (n1 n2) -> b c (h n1) (w n2)', n1=self.kernel_size, n2=self.kernel_size) # b c h w k**2 -> b c h*k w*k
receptive_field_data = rearrange(x, 'b (c n1) l -> b c n1 l', n1=self.kernel_size ** 2).permute(0, 1, 3, 2).reshape(b, c, h, w, self.kernel_size ** 2) # b c*kernel**2,h*w -> b c h w k**2
data_out = rearrange(receptive_field_data, 'b c h w (n1 n2) -> b c (h n1) (w n2)', n1=self.kernel_size,n2=self.kernel_size) # b c h w k**2 -> b c h*k w*k
conv_data = data_out * weight_out
conv_out = self.conv(conv_data)
return self.act(self.bn(conv_out))
# 在Visdrone数据集中，基于Unfold和Group Conv的RFAConv进行了对比实验，检测模型为YOLOv5n，学习率为0.1，batch-size为8，epoch为300，其他超参数为默认参数。 其中RFAConv替换了C3瓶颈中的3*3卷积运算。
# 实验表明，基于Group Conv的RFAConv性能更好，因为Unfold提取感受野空间特征时，一定程度上消耗时间比较严重。因此全文选择了Group Conv的方法进行实验，并通过这种方式对CBAM和CA进行改进。
""" 一个小白写给读者的一些启发：
（1）基于局部窗口的自注意力，最后通过softmax进行加权，然后进行sum融合特征。以这种角度理解RFAConv，同样通过Softmax进行加权，然后通过卷积核参数进行sum融合局部窗口的信息。
那么是否可以将局部窗口的自注意力最后的sum也通过高效的卷积参数或者全连接参数进行融合。
（2）除去论文外的其他的空间注意力是否可以把关注度放到感受野空间特征中呢，我觉得这是可行的。
"""
class SE(nn.Module):
def __init__(self, in_channel, ratio=16):
super(SE, self).__init__()
self.gap = nn.AdaptiveAvgPool2d((1, 1))
self.fc = nn.Sequential(
nn.Linear(in_channel, ratio, bias=False), # 从 c -> c/r
nn.ReLU(),
nn.Linear(ratio, in_channel, bias=False), # 从 c/r -> c
nn.Sigmoid()
)
def forward(self, x):
b, c= x.shape[0:2]
y = self.gap(x).view(b, c)
y = self.fc(y).view(b, c,1, 1)
return y
class RFCBAMConv(nn.Module):
def __init__(self,in_channel,out_channel,kernel_size=3,stride=1,dilation=1):
super().__init__()
if kernel_size % 2 == 0:
assert("the kernel_size must be odd.")
self.kernel_size = kernel_size
self.generate = nn.Sequential(nn.Conv2d(in_channel,in_channel * (kernel_size**2),kernel_size,padding=kernel_size//2,
stride=stride,groups=in_channel,bias =False),
nn.BatchNorm2d(in_channel * (kernel_size**2)),
nn.ReLU()
)
self. get_weight = nn.Sequential(nn.Conv2d(2,1,kernel_size=3,padding=1,bias=False),nn.Sigmoid())
self.se = SE(in_channel)
self.conv = nn.Sequential(nn.Conv2d(in_channel,out_channel,kernel_size,stride=kernel_size),nn.BatchNorm2d(out_channel),nn.ReLu())
def forward(self,x):
b,c = x.shape[0:2]
channel_attention = self.se(x)
generate_feature = self.generate(x)
h,w = generate_feature.shape[2:]
generate_feature = generate_feature.view(b,c,self.kernel_size**2,h,w)
generate_feature = rearrange(generate_feature, 'b c (n1 n2) h w -> b c (h n1) (w n2)', n1=self.kernel_size,
n2=self.kernel_size)
unfold_feature = generate_feature * channel_attention
max_feature,_ = torch.max(generate_feature,dim=1,keepdim=True)
mean_feature = torch.mean(generate_feature,dim=1,keepdim=True)
receptive_field_attention = self.get_weight(torch.cat((max_feature,mean_feature),dim=1))
conv_data = unfold_feature * receptive_field_attention
return self.conv(conv_data)
class h_sigmoid(nn.Module):
def __init__(self, inplace=True):
super(h_sigmoid, self).__init__()
self.relu = nn.ReLU6(inplace=inplace)
def forward(self, x):
return self.relu(x + 3) / 6
class h_swish(nn.Module):
def __init__(self, inplace=True):
super(h_swish, self).__init__()
self.sigmoid = h_sigmoid(inplace=inplace)
def forward(self, x):
return x * self.sigmoid(x)
class RFCAConv(nn.Module):
def __init__(self, inp, oup,kernel_size,stride, reduction=32):
super(RFCAConv, self).__init__()
self.kernel_size = kernel_size
self.generate = nn.Sequential(nn.Conv2d(inp,inp * (kernel_size**2),kernel_size,padding=kernel_size//2,
stride=stride,groups=inp,
bias =False),
nn.BatchNorm2d(inp * (kernel_size**2)),
nn.ReLU()
)
self.pool_h = nn.AdaptiveAvgPool2d((None, 1))
self.pool_w = nn.AdaptiveAvgPool2d((1, None))
mip = max(8, inp // reduction)
self.conv1 = nn.Conv2d(inp, mip, kernel_size=1, stride=1, padding=0)
self.bn1 = nn.BatchNorm2d(mip)
self.act = h_swish()
self.conv_h = nn.Conv2d(mip, inp, kernel_size=1, stride=1, padding=0)
self.conv_w = nn.Conv2d(mip, inp, kernel_size=1, stride=1, padding=0)
self.conv = nn.Sequential(nn.Conv2d(inp,oup,kernel_size,stride=kernel_size))
def forward(self, x):
b,c = x.shape[0:2]
generate_feature = self.generate(x)
h,w = generate_feature.shape[2:]
generate_feature = generate_feature.view(b,c,self.kernel_size**2,h,w)
generate_feature = rearrange(generate_feature, 'b c (n1 n2) h w -> b c (h n1) (w n2)', n1=self.kernel_size,
n2=self.kernel_size)
x_h = self.pool_h(generate_feature)
x_w = self.pool_w(generate_feature).permute(0, 1, 3, 2)
y = torch.cat([x_h, x_w], dim=2)
y = self.conv1(y)
y = self.bn1(y)
y = self.act(y)
h,w = generate_feature.shape[2:]
x_h, x_w = torch.split(y, [h, w], dim=2)
x_w = x_w.permute(0, 1, 3, 2)
a_h = self.conv_h(x_h).sigmoid()
a_w = self.conv_w(x_w).sigmoid()
return self.conv(generate_feature * a_w * a_h)