decisionforce/mmTransformer

The ckpt has a social decoder, but the code does not.

xushilin1 opened this issue · 1 comments

xushilin1 commented

Thanks for your code. I encountered several questions when reproducing this paper.
I hope you can help me solve these questions.

  • Code did not have a social decoder, but the ckpt you provided has. I'm interested in how you trained this ckpt.
  • What is the classification loss you choose for six trajectory proposals? CrossEntropy or KL and what is the target ?
  • What is the loss weight for reg loss and cls loss?
    Thanks for your project again and I really hope to get your reply.
DinghaoXi commented