Counterfactual Inpainting for Weakly Supervised Semantic Segmentation of Medical Images
A novel counterfactual inpainting approach (COIN) that generates precise segmentation masks for medical images without relying on pre-existing segmentation annotations, leveraging only image-level labels.