demo / model /CoordAttention.py
amazinghaha's picture
Upload 106 files
b4d6f1e verified
raw
history blame
3.31 kB
import torch
import torch.nn as nn
import torch.nn.functional as F
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 CoordAtt(nn.Module):
def __init__(self, inp, oup, reduction=32):
super(CoordAtt, self).__init__()
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.bn2 = nn.BatchNorm2d(1)
self.bn3 = nn.BatchNorm2d(1)
self.act = h_swish()
self.bn4 = nn.BatchNorm2d(mip)
self.bn5 = nn.BatchNorm2d(mip)
self.bn6 = nn.BatchNorm2d(1)
self.bn7 = nn.BatchNorm2d(1)
self.conv_h = nn.Conv2d(mip, oup, kernel_size=1, stride=1, padding=0)
self.conv_w = nn.Conv2d(mip, oup, kernel_size=1, stride=1, padding=0)
def forward(self, x):
x = torch.unsqueeze(x, 1) #2 1 2304 196
identity = x
n, c, h, w = x.size()#2 1 2304 196
x_h = self.bn2(self.pool_h(x))#2 1 2304 1
x_w = self.bn3(self.pool_w(x).permute(0, 1, 3, 2)) #2 1 196 1
identity_x_w = x_w
identity_x_h = x_h
y = torch.cat([x_h, x_w], dim=2)
y = self.conv1(y) #2 8 2500 1
y = self.bn1(y)
y = self.act(y)
x_h, x_w = torch.split(y, [h, w], dim=2) #2 8 2304 1 | 2 8 196 1
x_h = self.bn4(x_h)+identity_x_h
x_w = self.bn5(x_w)+identity_x_w
x_w = x_w.permute(0, 1, 3, 2)
a_h = self.bn6(self.conv_h(x_h)).sigmoid() #2 1 2304 1
a_w = self.bn7(self.conv_w(x_w)).sigmoid() #24 1 1 196
out = identity * a_w * a_h #点×
out = torch.squeeze(out, 1)
return out
class CoordAtt_ori(nn.Module):
def __init__(self, inp, oup, reduction=32):
super(CoordAtt_ori, self).__init__()
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, oup, kernel_size=1, stride=1, padding=0)
self.conv_w = nn.Conv2d(mip, oup, kernel_size=1, stride=1, padding=0)
def forward(self, x):
x = torch.unsqueeze(x, 1)
identity = x
n, c, h, w = x.size()
x_h = self.pool_h(x)
x_w = self.pool_w(x).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)
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()
out = identity * a_w * a_h
out = torch.squeeze(out, 1)
return out