search

open-mmlab / OpenPCDet

OpenPCDet Toolbox for LiDAR-based 3D Object Detection.
Apache License 2.0
4.66k stars 1.3k forks source link

Question about PointRCNN training strategy #486

Closed HannahJHan closed 2 years ago

HannahJHan commented 3 years ago

Thanks for your wonderful work~

I am puzzled on the training strategy. The PointRCNN paper says that

The two stage sub-networks of PointRCNN are trained separately.

, and the official code repository does indeed train separately.

But the code here(OpenPCDet) seems like the entire PointRCNN network is trained end-to-end. Am I wrong? What factors have you improved to make it trainable end-to-end? Please help me, thank you soooo much!

KevinFlyshr commented 3 years ago

I have the same question. I believe the number of epochs configured for training involves both stages but it would be great to know if that is really correct. Thanks in advance!

jihanyang commented 3 years ago

PointRCNN in OpenPCDet is trained jointly (unlike two stage training in the original code) with 80 epochs.

github-actions[bot] commented 2 years ago

This issue is stale because it has been open for 30 days with no activity.

github-actions[bot] commented 2 years ago

This issue was closed because it has been inactive for 14 days since being marked as stale.