An end-to-end tracking method for polyp detectors in colonoscopy videos

作者：

Highlights：

• Instance Tracking Head for lesions detection and tracking simultaneously in videos.

• Proposed SAS and BAS method boost the performance of multiple object tracking task.

• Proposed structure achieves comparable trade-off between model accuracy and speed.

• Inference time is more than 66 frames per second for real-time videos.

摘要

•Instance Tracking Head for lesions detection and tracking simultaneously in videos.•Proposed SAS and BAS method boost the performance of multiple object tracking task.•Proposed structure achieves comparable trade-off between model accuracy and speed.•Inference time is more than 66 frames per second for real-time videos.

论文关键词：Computer aided diagnosis,Endoscopy,Neural networks,Detection,Tracking

论文评审过程：Received 22 March 2021, Revised 4 May 2022, Accepted 11 July 2022, Available online 14 July 2022, Version of Record 20 July 2022.

论文官网地址：https://doi.org/10.1016/j.artmed.2022.102363