BPR checkpoint

[lsm140]

Dear authors, can you provide the BPR model trained on PointRend or SegFix results rather than mask rcnn？

[lsm140]

Another question: if the BPR model is trained based on patches from mask-rcnn and COCO GTmask, wouldn't it be better to refine the inference results for mask-rcnn? While there is discussion in the paper about transferability, is it best if your own results are used for training and used to correct the edge effects of your own algorithms?

[chenhang98]

Please find checkpoints at the following links:

• trained on PointRend: https://cloud.tsinghua.edu.cn/f/a2497064f25949d693ba/?dl=1
• trained on SegFix: https://cloud.tsinghua.edu.cn/f/1befec256b8e4a02b2da/?dl=1

[chenhang98]

For the other question, we found that models trained on worse results typically had stronger refinement ability, and we guessed that the reason might be similar to data augmentation. We therefore mainly use the model trained on Mask R-CNN's results.