BigDong

> > @christian-hiebl Thanks for your contribution. I think maybe we can set `keep` out of the conditions. Such as: > > ```python > > if self.by_box: > > assert...

d2 release faster-rcnn: ``` "bbox/AP": 37.931116259843556, "bbox/AP50": 58.83956400223075, "bbox/AP75": 41.05003436472164, "bbox/APl": 49.099591469327336, "bbox/APm": 41.140783153281504, "bbox/APs": 22.438183787843734, ``` mmdet version: ``` 08/31 11:10:03 - mmengine - INFO - Iter(test) [2500/2500] coco/bbox_mAP:...

| Model | d2 release | mmdet wrapper | |:--------:|:--------------:|:----------------------| | Faster R-CNN | box AP: 37.9 | box AP: 38.0 | |RetinaNet | box AP: 37.4 | box AP:...

Hi, sorry for the late reply, have you fixed this issue?

Followed PR: https://github.com/open-mmlab/mmdetection/pull/8437

What command did you use? Please use the issue template

I probably understand what you mean, but the situation you said is too special, and there are also min_pos constraints, so I guess min_pos can be adjusted appropriately to avoid...

Thanks for your report, we will have a discuss and fix it ASAP

The comments of the config file can be found at here: https://mmdetection.readthedocs.io/en/latest/tutorials/config.html as for common, it is some common setting for multi-scale 3x or some other comment settings.

lsj -> Large Scale Jittering ms -> multi Scale ssj -> Standard Scale Jittering These meanings are always shown in config/xx/README