V2AI/Det3D

Do you support Waymo? Or how to train Waymo using Det3D?

Closed this issue · 1 comments

Here is an instruction about how to extract the bin file and label file from Waymo.
https://www.jianshu.com/p/a3e8310cea32 (It is written in Chinese.)

I am a beginner about 3D detection. Could anyone give me a guide to train Waymo using Det3D?

Many thanks.

Thanks for your interest in Det3D!

It should be quite easy to add a dataset into Det3D. Here are the recommended steps:

  1. preprocess the Waymo dataset and generate the infos (annotations) into a pkl file (you can refer to https://github.com/poodarchu/Det3D/blob/master/tools/create_data.py)
  2. follow any of existing datasets (e.g. https://github.com/poodarchu/Det3D/blob/master/det3d/datasets/kitti/kitti.py) to create a Waymo(data.datasets) class, and implement the init (which is used to load the infos file generated in step 1) and the getitem (which is used to load the file and preprocess it).
  3. Add corresponding evaluation func according to waymo's API and metrics.