search

HKUST-Aerial-Robotics / SIMPL

SIMPL: A Simple and Efficient Multi-agent Motion Prediction Baseline for Autonomous Driving
MIT License
161 stars 13 forks source link

Question about YawLoss #4

Closed yenanjing closed 3 months ago

yenanjing commented 3 months ago

Thanks for your wonderful work and code sharing. I would like to ask where the yaw loss mentioned in the paper is reflected in the code. I can't find it in the network output and loss function definition, looking forward to your answer. In addition, I am curious whether this loss function has a relatively large role in your experiment.

MasterIzumi commented 3 months ago

@yenanjing Thanks for your comments. Since there is no ground truth yaw angle in the Argo1 dataset, the current version does not contain this part. I am currently busy with some personal matters, sorry for the delay. The loss function for yaw angles will be released with the code for Argo2 dataset in the near future, hopefully within this month.

MasterIzumi commented 3 months ago

About the second question, in my personal experience, the introduction of yaw loss can improve the prediction performance for low-speed vehicles, avoiding some jerky trajectories. However, it does not influence very much on the overall official metrics.

yenanjing commented 3 months ago

Thank you for your timely reply and look forward to your future updates.

MasterIzumi commented 3 months ago

@yenanjing Hi, the code for Argo2 has been released. Please feel free to contact us if you have any questions.