针对 BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation.该论文提出的语义分割网络，根据第三方实现提供的pytorch源码，进行了详细分析解读。论文中的网络框架如下图：
源码中网络设计
对照上面的网络框架，下面的代码很好理解。其中在Context path部分，代码中使用的是res18和res101。
class ConvBlock(torch.nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=3, stride=2,padding=1):
super().__init__()
self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, stride=stride, padding=padding)
self.bn = nn.BatchNorm2d(out_channels)
self.relu = nn.ReLU()
def forward(self, input):
x = self.conv1(input)
return self.relu(self.bn(x))
class Spatial_path(torch.nn.Module):
def __init__(self):
super().__init__()
self.convblock1 = ConvBlock(in_channels=3, out_channels=64)
self.convblock2 = ConvBlock(in_channels=64, out_channels=128)
self.convblock3 = ConvBlock(in_channels=128, out_channels=256)
def forward(self, input):
x = self.convblock1(input)
x = self.convblock2(x)
x = self.convblock3(x)
return x
class AttentionRefinementModule(torch.nn.Module):
def __init__(self, in_channels, out_channels):
super().__init__()
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=1)
self.bn = nn.BatchNorm2d(out_channels)
self.sigmoid = nn.Sigmoid()
self.in_channels = in_channels
def forward(self, input):
# global average pooling
x = torch.mean(input, 3, keepdim=True)
x = torch.mean(x, 2, keepdim=True)
assert self.in_channels == x.size(1), 'in_channels {} and out_channels {} should all be {}'.format(self.in_channels,x.size(1),x.size(1))
x = self.conv(x)
# x = self.sigmoid(self.bn(x))
x = self.sigmoid(x)
# channels of input and x should be same
x = torch.mul(input, x)
return x
class FeatureFusionModule(torch.nn.Module):
def __init__(self, num_classes,in_channels=1024):
super().__init__()
self.in_channels = in_channels
self.convblock = ConvBlock(in_channels=self.in_channels, out_channels=num_classes, stride=1)
self.conv1 = nn.Conv2d(num_classes, num_classes, kernel_size=1)
self.relu = nn.ReLU()
self.conv2 = nn.Conv2d(num_classes, num_classes, kernel_size=1)
self.sigmoid = nn.Sigmoid()
def forward(self, input_1, input_2):
x = torch.cat((input_1, input_2), dim=1)
assert self.in_channels == x.size(1), 'in_channels {} of ConvBlock should be {}'.format(self.in_channels,x.size(1))
feature = self.convblock(x)
x = torch.mean(feature, 3, keepdim=True)
x = torch.mean(x, 2 ,keepdim=True)
x = self.relu(self.conv1(x))
x = self.sigmoid(self.relu(x))
x = torch.mul(feature, x)
x = torch.add(x, feature)
return x
class BiSeNet(torch.nn.Module):
def __init__(self, num_classes, context_path):
super().__init__()
# build spatial path
self.saptial_path = Spatial_path()
# build context path
self.context_path = build_contextpath(name=context_path) #这里其实就是特征提取的基本网络，主要用到了res18和res101
# build attention refinement module
if context_path=='resnet18':
self.attention_refinement_module1 = AttentionRefinementModule(256, 256)
self.attention_refinement_module2 = AttentionRefinementModule(512, 512)
elif context_path=='resnet101':
self.attention_refinement_module1 = AttentionRefinementModule(1024, 1024)
self.attention_refinement_module2 = AttentionRefinementModule(2048, 2048)
else:
raise 'context_path error'
# build feature fusion module
if context_path=='resnet18':
self.feature_fusion_module = FeatureFusionModule(num_classes,1024) #此处源码没有实现，因此会有错误。我进行了分析和实现
elif context_path=='resnet101':
self.feature_fusion_module = FeatureFusionModule(num_classes,3328)
else:
raise 'context_path error'
# build final convolution
self.conv = nn.Conv2d(in_channels=num_classes, out_channels=num_classes, kernel_size=1)
def forward(self, input):
# output of spatial path
sx = self.saptial_path(input)
# output of context path
cx1, cx2, tail = self.context_path(input)
cx1 = self.attention_refinement_module1(cx1)
cx2 = self.attention_refinement_module2(cx2)
cx2 = torch.mul(cx2, tail)
# upsampling
cx1 = torch.nn.functional.interpolate(cx1, scale_factor=2, mode='bilinear')
cx2 = torch.nn.functional.interpolate(cx2, scale_factor=4, mode='bilinear')
cx = torch.cat((cx1, cx2), dim=1)
# output of feature fusion module
result = self.feature_fusion_module(sx, cx)
# upsampling
result = torch.nn.functional.interpolate(result, scale_factor=8, mode='bilinear')
result = self.conv(result)
return result打印建立的BiseNet res18网络模型
BiSeNet(
(saptial_path): Spatial_path(
(convblock1): ConvBlock(
(conv1): Conv2d(3, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU()
)
(convblock2): ConvBlock(
(conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
(bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU()
)
(convblock3): ConvBlock(
(conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU()
)
)
(context_path): resnet18(
(features): ResNet(
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer2): Sequential(
(0): BasicBlock(
(conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer3): Sequential(
(0): BasicBlock(
(conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer4): Sequential(
(0): BasicBlock(
(conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(avgpool): AvgPool2d(kernel_size=7, stride=1, padding=0)
(fc): Linear(in_features=512, out_features=1000, bias=True)
)
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(maxpool1): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer1): Sequential(
(0): BasicBlock(
(conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
(1): BasicBlock(
(conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer2): Sequential(
(0): BasicBlock(
(conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer3): Sequential(
(0): BasicBlock(
(conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(layer4): Sequential(
(0): BasicBlock(
(conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): BasicBlock(
(conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
)
(attention_refinement_module1): AttentionRefinementModule(
(conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1))
(bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(sigmoid): Sigmoid()
)
(attention_refinement_module2): AttentionRefinementModule(
(conv): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1))
(bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(sigmoid): Sigmoid()
)
(feature_fusion_module): FeatureFusionModule(
(convblock): ConvBlock(
(conv1): Conv2d(1024, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(bn): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU()
)
(conv1): Conv2d(32, 32, kernel_size=(1, 1), stride=(1, 1))
(relu): ReLU()
(conv2): Conv2d(32, 32, kernel_size=(1, 1), stride=(1, 1))
(sigmoid): Sigmoid()
)
(conv): Conv2d(32, 32, kernel_size=(1, 1), stride=(1, 1))
)
关于该模型的训练的具体细节和效果，可以参看我的另一篇博文： https://blog.csdn.net/rainforestgreen/article/details/85047054