AggLoss: Occlusion-aware R-CNN: Detecting Pedestrians in a Crowd

October 2020

tl;dr: Encourage different anchors to generate the same prediction, and occlusion-aware RoI pooling.

Overall impression

There are two main contributions of the paper. The first is AggLoss which encourages diff anchors associated with the same GT to output the same output. The second is the occlusion aware part-based RoI pooling.

Both RepLoss and AggLoss proposes additional penalties to produce more compact bounding boxes and become less sensitive to NMS. And also imposes additional penalties to bbox which appear in the middle of the two pedestrians.

Key ideas

Technical details