search

JialeCao001 / D2Det

D2Det: Towards High Quality Object Detection and Instance Segmentation (CVPR2020)
https://openaccess.thecvf.com/content_CVPR_2020/papers/Cao_D2Det_Towards_High_Quality_Object_Detection_and_Instance_Segmentation_CVPR_2020_paper.pdf
MIT License
297 stars 86 forks source link

why it is better than faster rcnn? #18

Open sankexin opened 3 years ago

sankexin commented 3 years ago

2 1 I have read your paper. But the operation in the picture can not give a real reason for your effect,. In the case of no pre-defined box, the regression range of box is very wide. But your binary overlap prediction is a good tool. Your operation of picture may return a worse box, can you explain the real reason why it works better ?

JialeCao001 commented 3 years ago

@sankexin Hi, thanks for your interest. I think it is difficult to provide a good regression for the proposals far from the objects.