SPengLiang / LPCG

[ECCV 2022] Lidar Point Cloud Guided Monocular 3D Object Detection.
74 stars 3 forks source link

Labels for train set #3

Open Yinyf0804 opened 1 year ago

Yinyf0804 commented 1 year ago

Hi, Thanks for your great work, I have a question about labels for the train set :

  1. All all the samples in the train set included in the raw scenes?
  2. Do you directly use the generated pseudo labels for the training set instead of the GT one when "train monocular models using pseudo labels", since I do not find any process to link GT labels for train set to the "dst_label_dir".
SPengLiang commented 1 year ago

Thanks for your interest in this work.

  1. Yes;
  2. You are right. In this codebase, we directly use the generated pseudo labels for convenience, as the gap between pseudo labels and GT can be ignored for monocular models. You can also use GT to replace pseudo labels in the training set, and the final performance is close.
Yinyf0804 commented 1 year ago

Thank you for such a quick and detailed reply~