Tai-Wang

From our experience, it will not affect the performance a lot, the error caused by the common projection and the waymo optimized version is also minor. Nevertheless, I agree with...

If the influence is a bit large, we will attempt to adjust this afterward.

Thanks for your reminder! Could you please try to fix this problem with a PR?

FCOS3D和PGD改成BN应该收敛会变慢，但之前和其他同事沟通应该可以通过训练时间加长来改善，不过这可能是大规模数据集上的经验，KITTI上可能会有点tricky。

Do you use the latest master or v1.0.0rc0? We have not finished the model update after refactoring the coordinate systems, and you can train models by yourself to test it...

Some pretrained models have been updated. Please check them in #1369 and try to reproduce the demo with the updated models. Looking forward to your feedback.

Sorry for the late reply. Please @ZCMax have a check and fix the bug if necessary.

The code of MV-FCOS3D++ for camera-only 3D detection will be released in a month if everything goes well. The "FCOS3D++" model is basically similar to the released version of PGD....

The code of MV-FCOS3D++ together with our recent work DfM are preliminarily released [here](https://github.com/Tai-Wang/Depth-from-Motion). We will further refine it recently and support them in the mmdet3d. Please stay tuned.

It's just because there are some print-debug codes not cleaned. You can just locate and remove them.