End-to-end Point Supervised Object Detection with low-level instance features

Xiangqi Chen, Chengzhuan Yang, Jiashuaizi Mo, Yunliang Jiang, Zhonglong Zheng

Abstract: Highlights•We propose a Low-level Feature Extractor module.•Our method reduces the performance gap between PSOD and box-level object detection.•Our method incorporates low-level feature information and category information.•We design an end-to-end detector to overcome local minima in object detection.