Pixel-level skin lesion segmentation using U-Net trained from scratch on HAM10000 — 0.9115 Dice score, 31M parameters, BCE + Dice loss, live demo on HuggingFace Spaces
computer-vision deep-learning pytorch google-colab medic unet-image-segmentation skin-lesion dice-loss ham10000 dermoscopy ablumentations
-
Updated
May 12, 2026 - Jupyter Notebook