by Yu Liu, Lingqiao Liu, Hamid Rezatofighi, Thanh-Toan Do, Qinfeng Shi, Ian Reid
https://arxiv.org/abs/1901.03796
This work targeting on cure the GreedyNMS (a post-processing step for Object detection).
The experiments based on Faster-RCNN and DPM
tested on three datasets: MOT15, TUD-Crossing and PETS.
The comparison works include SoftNMS and ConvNMS.