Image
Semantic Segmentation
U-Net: Convolutional Networks for Biomedical Image Segmentation
- #2015
- 1-class Instance Segmentation
- End-to-end
- Medical Image Segmentation에서 자주 사용되는 모델
- Deconvolution, Feature Crop & Concat, Mirror Padding 등을 섬세하게 잘 설계
FusionNet - A deep fully residual convolutional neural network for image segmentation in connectomics
- #2016
- 뇌지도를 그리는 분야에서 적용된 사례
- Residual을 활용한 End-to-end semantic segmentation model
Size-constraint loss for weakly supervised CNN segmentation
- #2018
- Segment의 크기에 대한 제약식을 Loss Function에 포함키는 것을 통해 성능 향상.
- https://steemit.com/kr/@jiwoopapa/size-constraint-loss-for-weakly-supervised-cnn-segmentation