search

metadriverse / trafficgen

[ICRA 2023] The official code for paper "TrafficGen: Learning to Generate Diverse and Realistic Traffic Scenarios"
Apache License 2.0
149 stars 24 forks source link

Some questions about data preprocessing and random mask for input scene #27

Closed ApisXia closed 1 year ago

ApisXia commented 1 year ago

Thank you for the nice work! I hope to modify your model. But I found some parts of the code confusing. I am grateful If you can help me solve these problems.

  1. In the data preprocessing, I noticed that you did not use the relative direction between velocity and heading in the paper, but you used that in the code. And after visualization of this relative direction, I found most values are quite small. I guess it could be the old version of the code? Could I remove this during my training?
  2. In the "agent_feature_extract", I noticed that you create a random mask (but actually it is not random). I guess it might also be the old version of the code. Could you please fix this later? Looking forward to your reply and thanks again for your work!
Alan-LanFeng commented 1 year ago

Hi,

  1. Yes, you can remove that during training which I think is trivial.
  2. The agent order was previously shuffled so the mask is random.