Automated lesion segmentation in fundus images with many-to-many reassembly of features

Qing Liu, Haotian Liu, Wei Ke, Yixiong Liang

Abstract: Highlights•We propose M2MRF to maintain subtle lesion activations and capture long-range dependencies for tiny lesion segmentation.•Our M2MRF reassembles multiple features inside a large predefined region into multiple output features simultaneously via learning.•Comprehensive experiments on DDR and IDRiD datasets show that our M2MRF outperforms state-of-the-art feature reassembly operators.