### 文章目录

2021 BoTNet 更好的backbone

## 1. 简介

### 1.1 简介

UC Berkeley 和 谷歌2021发表的一篇论文，属于 早期的结合CNN+Transformer 的工作。基于 Transformer的骨干网络 ，同时使用卷积与自注意力机制来保持全局性和局部性。模型在ResNet最后三个BottleNeck中使用了 MHSA替换3x3卷积 。简单来讲Non-Local+Self Attention+BottleNeck = BoTNet

### 1.2 摘要

(1) 使用卷积提取有效的局部特征，降低分辨率

(2) 使用self-attention聚合全局信息（操作对象是feature map)

## 2. 网络

### 2.2 MHSA模块

y i = s o f t m a x ( θ ( x i ) T ϕ ( x j ) ) g ( x j ) = 1 ∑ ∀ j e θ ( x i ) T ϕ ( x j ) e θ ( x i ) T ϕ ( x j ) W g x j y_i=softmax(\theta(x_i)^T\phi(x_j))g(x_j)=\frac{1}{\sum_{\forall j}e^{\theta(x_i)^T\phi(x_j)}}e^{\theta(x_i)^T\phi(x_j)}W_gx_j s o f t m a x ( θ ( x i ​ ) T ϕ ( x j ​ ) ) g ( x j ​ ) = ∑ ∀ j ​ e θ ( x i ​ ) T ϕ ( x j ​ ) 1 ​ e θ ( x i ​ ) T ϕ ( x j ​ ) W g ​ x j ​

## 3. 代码

import torch
import torch.nn as nn
import torch.nn.functional as F
def get_n_params(model):
pp=0
for p in list(model.parameters()):
nn=1
for s in list(p.size()):
nn = nn*s
pp += nn
return pp
class MHSA(nn.Module):
def __init__(self, n_dims, width=14, height=14, heads=4):
super(MHSA, self).__init__()
self.query = nn.Conv2d(n_dims, n_dims, kernel_size=1)
self.key = nn.Conv2d(n_dims, n_dims, kernel_size=1)
self.value = nn.Conv2d(n_dims, n_dims, kernel_size=1)
self.softmax = nn.Softmax(dim=-1)
def forward(self, x):
n_batch, C, width, height = x.size()
content_content = torch.matmul(q.permute(0, 1, 3, 2), k)
content_position = (self.rel_h + self.rel_w).view(1, self.heads, C // self.heads, -1).permute(0, 1, 3, 2)
content_position = torch.matmul(content_position, q)
energy = content_content + content_position
attention = self.softmax(energy)
out = torch.matmul(v, attention.permute(0, 1, 3, 2))
out = out.view(n_batch, C, width, height)
return out
class Bottleneck(nn.Module):
expansion = 4
def __init__(self, in_planes, planes, stride=1, heads=4, mhsa=False, resolution=None):
super(Bottleneck, self).__init__()
self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=1, bias=False)
self.bn1 = nn.BatchNorm2d(planes)
if not mhsa:
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, padding=1, stride=stride, bias=False)
else:
self.conv2 = nn.ModuleList()
if stride == 2:
self.conv2.append(nn.AvgPool2d(2, 2))
self.conv2 = nn.Sequential(*self.conv2)
self.bn2 = nn.BatchNorm2d(planes)
self.conv3 = nn.Conv2d(planes, self.expansion * planes, kernel_size=1, bias=False)
self.bn3 = nn.BatchNorm2d(self.expansion * planes)
self.shortcut = nn.Sequential()
if stride != 1 or in_planes != self.expansion*planes:
self.shortcut = nn.Sequential(
nn.Conv2d(in_planes, self.expansion*planes, kernel_size=1, stride=stride),
nn.BatchNorm2d(self.expansion*planes)
)
def forward(self, x):
out = F.relu(self.bn1(self.conv1(x)))
out = F.relu(self.bn2(self.conv2(out)))
out = self.bn3(self.conv3(out))
out += self.shortcut(x)
out = F.relu(out)
return out
# reference
# https://github.com/kuangliu/pytorch-cifar/blob/master/models/resnet.py
class ResNet(nn.Module):
def __init__(self, block, num_blocks, num_classes=1000, resolution=(224, 224), heads=4):
super(ResNet, self).__init__()
self.in_planes = 64
self.resolution = list(resolution)
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
# self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False)
if self.conv1.stride[0] == 2:
self.resolution[0] /= 2
if self.conv1.stride[1] == 2:
self.resolution[1] /= 2
self.bn1 = nn.BatchNorm2d(64)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) # for ImageNet
if self.maxpool.stride == 2:
self.resolution[0] /= 2
self.resolution[1] /= 2
self.layer1 = self._make_layer(block, 64, num_blocks[0], stride=1)
self.layer2 = self._make_layer(block, 128, num_blocks[1], stride=2)
self.layer3 = self._make_layer(block, 256, num_blocks[2], stride=2)
self.fc = nn.Sequential(
nn.Dropout(0.3), # All architecture deeper than ResNet-200 dropout_rate: 0.2
nn.Linear(512 * block.expansion, num_classes)
)
def _make_layer(self, block, planes, num_blocks, stride=1, heads=4, mhsa=False):
strides = [stride] + [1]*(num_blocks-1)
layers = []
for idx, stride in enumerate(strides):
layers.append(block(self.in_planes, planes, stride, heads, mhsa, self.resolution))
if stride == 2:
self.resolution[0] /= 2
self.resolution[1] /= 2
self.in_planes = planes * block.expansion
return nn.Sequential(*layers)
def forward(self, x):
out = self.relu(self.bn1(self.conv1(x)))
out = self.maxpool(out) # for ImageNet
out = self.layer1(out)
out = self.layer2(out)
out = self.layer3(out)
out = self.layer4(out)
out = self.avgpool(out)
out = torch.flatten(out, 1)
out = self.fc(out)
return out
def main():
x = torch.randn([2, 3, 224, 224])
print(model(x).size())
print(get_n_params(model))
# if __name__ == '__main__':
#     main()

BoTNet:Bottleneck Transformers for Visual Recognition_ pprp 的博客-CSDN博客

Bottleneck Transformers for Visual Recognition 阅读 – 知乎 (zhihu.com)