Yinhao Li comments

Results 42 comments of


                                            Yinhao Li

Problem of EMA

You can run `bev_depth_lss_r50_256x704_128x128_20e_cbgs_2key_da_ema.py` directly.

`gen_depth_gt.py` does not exist in your project now, has anyone deleted it?

Yes, it is deleted, in v0.2.0, it is not needed any more.

how to resume from the checkpoint?

Hi! I hope this can help you. https://stackoverflow.com/questions/71961436/pytorch-lightning-resuming-from-checkpoint-with-new-data

Hi! We have provided multi-frame exps, please refer to https://github.com/Megvii-BaseDetection/BEVDepth/blob/main/exps/bev_depth_lss_r50_256x704_128x128_20e_cbgs_2key_da_ema.py. To reproduce the results in paper, you need to change the backbone to Vovnet and the image resolution to 640...

Multi-frame Fusion

Yes, On the leaderboard, we use vovnet and 640x1600 resolution.

The result of depth prediction.

It is not available for now.

Training and evaluation on Waymo open dataset

Right now, supporting waymo dataset is not on our TODO list. There are many codebases you can refer to like mmdetection3d and OpenPCDet, both of them have a good support...

How to parse depth predict reuslt and to evaluate SILog, Abs Rel metrices?

You can use the depth prediction result generated by depth net to get the results.

How to train only a camera input with cam_front?

This should work, but in my experimence, this may result in performance drop, since gt are not filtered based on the cameras you use.

AttributeError: 'BEVDepthLightningModel' object has no attribute 'module'

Please pytorch-lightning.