pytorch-semantic-segmentation

Why we included this project

This collection shows how the classic semantic segmentation networks fit together. It bundles FCN, U-Net, SegNet, PSPNet, GCN, and DUC/HDC implementations with a shared training and testing pipeline, so you can compare them without hopping between repos. The code reflects the PyTorch style of its time, which makes it compact and easy to trace when you want to see where each architecture diverges. Since it pins an old PyTorch version and has been untouched since 2019, treat it as a reference for porting ideas into modern code, not as a drop-in for production. For anyone rebuilding these models today, it's a handy reminder of the original implementation details before you dig into the official releases.