search

OpenDriveLab / TCP

[NeurIPS 2022] Trajectory-guided Control Prediction for End-to-end Autonomous Driving: A Simple yet Strong Baseline.
Apache License 2.0
310 stars 40 forks source link

Imbalanced Dataset #6

Open vaydingul opened 1 year ago

vaydingul commented 1 year ago

Hi everyone,

As the TCP implementation goes on, I realized that the current settings for the training (i.e., 60 epochs with LR decay after 30 epochs) result in severe overfitting to LANE FOLLOWING mode. It is normal considering the general distribution of the CARLA-based autonomous driving datasets.

Have you tried anything (e.g., weighted sampling) to solve that issue?

jiaxiaosong1002 commented 1 year ago

Hi vaydingul, thanks for your feedback and interesting findings. We have not try weighted sampling for TCP. As for the epoch length and LR decay, we do not tune them too much. You are welcomed to try the best setting and share with the community:)

Also, could you specify the meaning of "overfitting to LANE FOLLOWING mode?" More details would be appreciated for the discussion.