语义分割网络整理
学习计划
一. 基础网络架构
- unet
- fcn
- segnet
- deeplab
- Resnet系列
- LENET
- DenseNet
- VGGNet
- GoogleNET
- Res2NET
- InceptionNet
- SqueezeNet
- ShuffleNET
- SENet
- DPNet
- Mobilenet
- NasNet
- DetNet
- EfficientNet
二. 各模块汇总
1. 注意力机制
- SE
- Non local
- CcNet
- GCNet
- Gate
- CBAM
- Dual Attention
- Spatial Attention
- Channel Attentiom
2. 卷积结构
- Residual Block
- Bottle-neck block
- Split-attention block
- Depthwise Separable Convolution
- Recurrent convolution
- Group convolution
- Dilated convolution
- Octaua convolution
- Ghost convolution
3. 多尺度模块
- ASPP
- PPM
- DCM
- DenseASPP
- FPA
- Ocnet
- MPM
着重在ASPP, PPM上,考虑加权
4. 损失函数loss
- Focal loss
- Dice loss
- BCEloss
- Wetight loss
- Boundary loss
- Softmax loss
- Topk loss
- Hausdorff distance loss(HD loss)
- Sensitivity Specificity loss
5. 池化结构
- Maxpooling
- Avarage pooling
- Random pooling
- Stirp pooling
- Mixed pooling
6. 学习衰减策略
- StepLR
- MultiStepLR
- RxponentialLR
- Cosine Annealing LR
- ReduceOn plateau
7. 归一化模块
- Batch Normalization
- Layer Normalization