Self-Paced Dual-Axis Attention Fusion Network for Retinal Vessel Segmentation

The segmentation of retinal vessels plays an essential role in the early recognition ophthalmic diseases clinics. Increasingly, approaches based on deep learning have been pushing vessel performance, yet it is still a challenging problem due to complex structure and lack precisely labeled samples. In this paper, we propose self-paced dual-axis attention fusion network (SPDAA-Net). Firstly, mechanism using query-by-committee algorithm designed guide model learn from easy hard, which makes training more intelligent. Secondly, during fusing multi-scale features, composed height width developed perceive object, brings long-range dependencies while reducing computation complexity. Furthermore, CutMix data augmentation applied increase generalization model, enhance ability global local ultimately boost accuracy. We implement comprehensive experiments validating that our SPDAA-Net obtains remarkable performance both public DRIVE CHASE-DB1 datasets.