This paper proposes a strategy for eczema segmentation using a model called SegGPT and visual in-context learning. The method requires fewer examples and no retraining models, making it more efficient than current methods that rely on a high volume of annotated data. The SegGPT model outperforms a CNN U-Net trained on a larger dataset. The findings highlight the potential of in-context learning for skin imaging tasks and the development of inclusive solutions for under-represented demographics.
Publication date: 28 Sep 2023
Project Page: https://arxiv.org/abs/2309.16656v1
Paper: https://arxiv.org/pdf/2309.16656
