Question about Table.5 and some pre-trained proposals

[qsunyuan]

非常启发的工作。但个人有一些疑惑的地方。

1. 关于表5的消融实验，能详细介绍一下3,4,5的crop region具体是怎么样的吗?
2. 关于实验配置部分，https://github.com/LutingWang/OADP#proposals 这部分是用来pkl文件offline proposals的吗?或者我遗漏了哪些细节。

[qsunyuan]

soco_star_mask_rcnn_r50_fpn_400e.pth 这个文件和detpro的 https://github.com/dyabel/detpro#prepare-data 有什么区别吗？ 在detpro中的readme如下 All models use the backbone pretrained with SoCo which can be downloaded from google drive baiduyun (code:kwps). Put the pretrained backbone under data/.

[LutingWang]

1. The following figure illustrates the differences between the crop strategies

2. the proposals are obtained offline through RPN or OLN. Both are trained solely using base annotations. Subsequently, these extracted proposals are utilized by the OAKE module to extract features for distillation.

[LutingWang]

The two checkpoint files are identical. We renamed the checkpoint file to ease the identification of the original SOCO checkpoint.

[qsunyuan]

谢谢作者细致的回复！