Medical-Transformer

Why we included this project

Medical Transformer targets a real problem: most transformer segmentation models need large training sets, but medical imaging datasets tend to be small. The repo holds the PyTorch code from the MICCAI 2021 paper, covering two network variants, the Gated Axial Attention U-Net and MedT, plus a LoGo local-global training scheme that learns from whole images and patches so the models can train on limited clinical data. You also get train and test scripts, conda or pip environment setup, and guidance for arranging your own image and label folders, which means you can retrain on your own data instead of relying on pretrained weights. Whether you are reproducing the paper's results or just checking how axial attention stacks up against your current U-Net, this is easy to trace and modify.