许可证：apache-2.0
标签：

• 图像分类
  数据集：
• imagenet

resnet26

实现论文《深度残差学习用于图像识别》中提出的ResNet架构

ResNet.resnet18()  
ResNet.resnet26()  
ResNet.resnet34()  
ResNet.resnet50()  
ResNet.resnet101()  
ResNet.resnet152()  
ResNet.resnet200()  

变体(d)参考论文《卷积神经网络图像分类技巧集锦》<https://arxiv.org/pdf/1812.01187.pdf>  

ResNet.resnet26d()  
ResNet.resnet34d()  
ResNet.resnet50d()  
# 可通过修改`stem`和`block`自定义结构  
resnet101d = ResNet.resnet101(stem=ResNetStemC, block=partial(ResNetBottleneckBlock, shortcut=ResNetShorcutD))  

使用示例：

# 替换激活函数  
ResNet.resnet18(activation = nn.SELU)  
# 修改分类数（默认1000类）  
ResNet.resnet18(n_classes=100)  
# 使用不同的block模块  
ResNet.resnet18(block=SENetBasicBlock)  
# 更换stem结构  
model = ResNet.resnet18(stem=ResNetStemC)  
# 修改shortcut连接方式  
model = ResNet.resnet18(block=partial(ResNetBasicBlock, shortcut=ResNetShorcutD))  
# 提取各层特征  
x = torch.rand((1, 3, 224, 224))  
# 获取特征  
model = ResNet.resnet18()  
# 首先调用.features以激活前向钩子，告知模型需要提取特征  
model.encoder.features  
model(torch.randn((1,3,224,224)))  
# 从编码器获取特征  
features = model.encoder.features  
print([x.shape for x in features])  
# 输出示例：[torch.Size([1, 64, 112, 112]), torch.Size([1, 64, 56, 56]), torch.Size([1, 128, 28, 28]), torch.Size([1, 256, 14, 14])]