I see lots of models for Classification, but not too many for Semantic Segmentation.
Just curious if anyone has an idea for why things like U-Nets aren’t included in the torchvision model zoo. I have always considered U-Nets to be a classic, like the “ResNets of semantic segmentation”. Is a DeepLab3 or FCN just ultimately better, or perhaps torchvision is still being updated? Maybe U-Nets aren’t typically used for transfer learning/feature extractors (compared to ResNets, or even the R-CNN family of object detection/instance segmentation/keypoint estimation models)?