Towards Effective Microalgae Object Detection Solutions to IEEE UV 2022 “Vision Meets Alage” Object Detection Challenge

Abstract

This technical report introduces our solution for microalgae object detection in IEEE UV 2022 Vision Meets Alage Object Detection Challenge. The purpose of this challenge is to employ computer vision to more effectively analyze population change in ocean microalgae species. Therefore, we performed a comprehensive analysis of the distribution of the microalgae dataset and designed a customized training strategy for the task. In order to better identify the categories and coordinates of microalgae in microscopic images, we propose CBSwin-Cascade RCNN++ as a strong baseline for microalgae detection. Our final submission the results, which achieves 56.13 in mAP 0.5:0.95 on a single model, and obtains 57.09 in mAP 0.5:0.95 with the ensembled models.