shawnding

Ah alright. I’m trying out some modifications based on CenterPoint, so it would be great to visualize loss and such from raw log : )在 2022年9月30日，09:11，Tianwei Yin ***@***.***> 写道： hi,...

Hi, @tianweiy Have you tried training CenterPoint without class-balanced grouping and sampling? CBGS enlarges the size of training set from 28k to 123k (also extends training time). I wonder its...

> 我遇到同样的问题，一直 `Choosing the optimal line...`，busybox返回也一样

Hi, @Abyssaledge I noticed there is a 6-frame FSD++ on the WOD leaderboard [here](https://waymo.com/open/challenges/entry/?challenge=DETECTION_3D&challengeId=DETECTION_3D&emailId=1cb154ab-1558&timestamp=1668404651664054). Do you mind publishing the detailed performance of this 6f model on the evaluation set? Thanks!

I see. Do you have the performance of the plain FSD on WOD validation set using 6 past frames? Thanks

Thanks! A following question, does the 6th channel represents **range frame offset** (according to [this](https://github.com/open-mmlab/mmdetection3d/blob/dev/docs/en/compatibility.md))? Does it make a difference to the detection performance?