SurgiNet: Pyramid Attention Aggregation and Class-wise Self-Distillation for Surgical Instrument Segmentation

•This paper extends from the conference paper Pyramid Attention Aggregation Network for Semantic Segmentation of Surgical Instruments, which has been published in the Thirty-Four AAAI Conference on Artificial Intelligence (AAAI-20).•The double attention module is designed to capture semantic dependencies between locations and channels, which helps to address illumination issues. Pyramid attention is aggregated to capture multi-scale features and make predictions more accurate.•Class-wise self-distillation is proposed to distill knowledge based on class probability maps, which can enhance the representation learning of the network. The network can adopt a lightweight backbone while obtaining high accuracy.•The proposed SurgiNet achieves the state-of-the-art performance of 89.14% mIoU on CataIS with only 1.66 GFlops and 2.05 M parameters. It also takes first place on EndoVis 2017 with 66.30% mIoU.•This manuscript constructs a new cataract surgical instrument segmentation dataset CataIS, which annotates 11 surgical instruments under different illumination scenarios. This dataset will be made public soon.