Tsang Ing Ren

Synthetic data from class-conditioned diffusion models, combined with MAE pretraining, allows smaller ViTs to achieve state-of-the-art skin lesion classification, opening the door for mobile clinical applications.

Bridging the gap between dermoscopic and clinical skin lesion images can be achieved through contrastive meta-domain adaptation, enhancing the robustness of deep learning models in real-world clinical settings.