VGG Loss的Pytorch实现

代码引自：https://github.com/bryandlee/stylegan2-encoder-pytorch/blob/master/train_encoder.py

class VGGLoss(nn.Module):
    def __init__(self, device, n_layers=5):
        super().__init__()
        
        feature_layers = (2, 7, 12, 21, 30)
        self.weights = (1.0, 1.0, 1.0, 1.0, 1.0)  

        vgg = torchvision.models.vgg19(pretrained=True).features
        
        self.layers = nn.ModuleList()
        prev_layer = 0
        for next_layer in feature_layers[:n_layers]:
            layers = nn.Sequential()
            for layer in range(prev_layer, next_layer):
                layers.add_module(str(layer), vgg[layer])
            self.layers.append(layers.to(device))
            prev_layer = next_layer
        
        for param in self.parameters():
            param.requires_grad = False

        self.criterion = nn.L1Loss().to(device)
        
    def forward(self, source, target):
        loss = 0 
        for layer, weight in zip(self.layers, self.weights):
            source = layer(source)
            with torch.no_grad():
                target = layer(target)
            loss += weight*self.criterion(source, target)
            
        return loss 

VGG Loss在GAN里面用的比较多，做风格迁移之类的，可以用来比较两张图像感官上的差距。具体来说的话其实就是将两张图像送入预训练的VGG19网络中提取各层特征，然后对比各层特征之间的差异(使用L1Loss)。所涉及到的一个关键参数为各层特征的权重，这个会对训练效果会有比较大的影响。