Cacher is the code snippet organizer for pro developers

class ASPP_module(nn.Module):
	def __init__(self, inplanes, planes, rate):
		super(ASPP_module, self).__init__()
		if rate == 1:  # 当rate为1的时候才是1*1的卷积
			kernel_size = 1
			padding = 0
		else:
			kernel_size = 3
			padding = rate
		self.atrous_convolution = nn.Conv2d(inplanes, planes, kernel_size=kernel_size,  # 空洞卷积的padding和rate似乎是相等的
			stride=1, padding=padding, dilation=rate, bias=False)
		self.bn = nn.BatchNorm2d(planes)
		self.relu = nn.ReLU()

		self.__init_weight()

	def forward(self, x):
		x = self.atrous_convolution(x)
		x = self.bn(x)

		return self.relu(x)

	def __init_weight(self):  # 可能是因为空洞卷积也是定义在一个class里面，所以还需要一个参数初始化
		for m in self.modules():
			if isinstance(m, nn.Conv2d):
				# n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
				# m.weight.data.normal_(0, math.sqrt(2. / n))
				torch.nn.init.kaiming_normal_(m.weight)
			elif isinstance(m, nn.BatchNorm2d):
				m.weight.data.fill_(1)
				m.bias.data.zero_()

class ASPP(nn.Module):
	def __init__(self, inplanes, planes, rates):  # 根据rate的list来决定各个空洞卷积rate的
		super(ASPP, self).__init__()

		self.aspp1 = ASPP_module(inplanes, planes, rate=rates[0])
		self.aspp2 = ASPP_module(inplanes, planes, rate=rates[1])
		self.aspp3 = ASPP_module(inplanes, planes, rate=rates[2])
		self.aspp4 = ASPP_module(inplanes, planes, rate=rates[3])

		self.relu = nn.ReLU()

		self.global_avg_pool = nn.Sequential(nn.AdaptiveAvgPool2d((1, 1)),
			nn.Conv2d(inplanes, planes, 1, stride=1, bias=False),  # 一个1*1卷积来变通道
			nn.BatchNorm2d(planes),
			nn.ReLU()
		)

		self.conv1 = nn.Conv2d(planes*5, planes, 1, bias=False)
		self.bn1 = nn.BatchNorm2d(planes)

	def forward(self, x):
		x1 = self.aspp1(x)
		x2 = self.aspp2(x)
		x3 = self.aspp3(x)
		x4 = self.aspp4(x)
		x5 = self.global_avg_pool(x)
		# 上采样x5使得和x4的尺寸一样，x4的size的第二层开始模仿
		x5 = F.upsample(x5, size=x4.size()[2:], mode='bilinear', align_corners=True)

		x = torch.cat((x1, x2, x3, x4, x5), dim=1)  # 这些特征加在一起

		x = self.conv1(x)
		x = self.bn1(x)
		x = self.relu(x)
		return x