class PRM(nn.Module):
def __init__(self, output_chl_num, efficient=False):
super(PRM, self).__init__()
self.output_chl_num = output_chl_num
self.conv_bn_relu_prm_1 = conv_bn_relu(self.output_chl_num, self.output_chl_num, kernel_size=3,
stride=1, padding=1, has_bn=True, has_relu=True,
efficient=efficient)
self.conv_bn_relu_prm_2_1 = conv_bn_relu(self.output_chl_num, self.output_chl_num, kernel_size=1,
stride=1, padding=0, has_bn=True, has_relu=True,
efficient=efficient)
self.conv_bn_relu_prm_2_2 = conv_bn_relu(self.output_chl_num, self.output_chl_num, kernel_size=1,
stride=1, padding=0, has_bn=True, has_relu=True,
efficient=efficient)
self.sigmoid2 = nn.Sigmoid()
self.conv_bn_relu_prm_3_1 = conv_bn_relu(self.output_chl_num, self.output_chl_num, kernel_size=1,
stride=1, padding=0, has_bn=True, has_relu=True,
efficient=efficient)
self.conv_bn_relu_prm_3_2 = conv_bn_relu(self.output_chl_num, self.output_chl_num, kernel_size=9,
stride=1, padding=4, has_bn=True, has_relu=True,
efficient=efficient,groups=self.output_chl_num)
self.sigmoid3 = nn.Sigmoid()
def forward(self, x):
out = self.conv_bn_relu_prm_1(x)
out_1 = out
out_2 = torch.nn.functional.adaptive_avg_pool2d(out_1, (1,1))
out_2 = self.conv_bn_relu_prm_2_1(out_2)
out_2 = self.conv_bn_relu_prm_2_2(out_2)
out_2 = self.sigmoid2(out_2)
out_3 = self.conv_bn_relu_prm_3_1(out_1)
out_3 = self.conv_bn_relu_prm_3_2(out_3)
out_3 = self.sigmoid3(out_3)
out = out_1.mul(1 + out_2.mul(out_3))
return out
class CAM_Module(Module):
""" Channel attention module"""
def __init__(self, in_dim):
super(CAM_Module, self).__init__()
self.chanel_in = in_dim
self.gamma = Parameter(torch.zeros(1))
self.softmax = Softmax(dim=-1)
def forward(self,x):
"""
inputs :
x : input feature maps( B X C X H X W)
returns :
out : attention value + input feature
attention: B X C X C
"""
m_batchsize, C, height, width = x.size()
proj_query = x.view(m_batchsize, C, -1)
proj_key = x.view(m_batchsize, C, -1).permute(0, 2, 1)
energy = torch.bmm(proj_query, proj_key)
energy_new = torch.max(energy, -1, keepdim=True)[0].expand_as(energy)-energy
attention = self.softmax(energy_new)
proj_value = x.view(m_batchsize, C, -1)
out = torch.bmm(attention, proj_value)
out = out.view(m_batchsize, C, height, width)
out = self.gamma*out + x
return out
|
