每日Attention学习5——Multi-Scale Channel Attention Module

模块出处

[link] [code] [WACV 21] Attentional Feature Fusion

---

模块名称

Multi-Scale Channel Attention Module (MS-CAM)

---

模块作用

通道注意力

---

模块结构

---

模块代码

import torch
import torch.nn as nn


class MS_CAM(nn.Module):

    def __init__(self, channels=64, r=4):
        super(MS_CAM, self).__init__()
        inter_channels = int(channels // r)

        self.local_att = nn.Sequential(
            nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
            nn.BatchNorm2d(inter_channels),
            nn.ReLU(inplace=True),
            nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
            nn.BatchNorm2d(channels),
        )

        self.global_att = nn.Sequential(
            nn.AdaptiveAvgPool2d(1),
            nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
            nn.BatchNorm2d(inter_channels),
            nn.ReLU(inplace=True),
            nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
            nn.BatchNorm2d(channels),
        )

        self.sigmoid = nn.Sigmoid()

    def forward(self, x):
        xl = self.local_att(x)
        xg = self.global_att(x)
        xlg = xl + xg
        wei = self.sigmoid(xlg)
        return x * wei

    
if __name__ == '__main__':
    x = torch.randn([3, 256, 16, 16])
    ms_sam = MS_CAM(channels=256)
    out = ms_sam(x)
    print(out.shape)  # 3, 256, 16, 16

---

原文表述

MS-CAM的核心思想在于，通过改变空间池化的大小，可以在多个尺度上实现通道注意力。为了尽可能保持轻量级，我们只是在注意力模块内将局部上下文添加到全局上下文中。我们选择点卷积（PointWise Conv）作为局部通道上下文融合器，它只利用每个空间位置的点级通道交互。