Result for single model

[fanghgit]

Thanks for much for sharing your code, the softmap score on the Waymo leaderboard is amazing. I tried to reproduce the result, but got only 0.372 for a single model, I guess the performance gap comes from model ensembling. Can you please share the best softmap (or map) score you got for a single model? Many thanks in advance.

[ares89]

@fanghgit hi, would you like to share your trained model ?

[SevenTailCat]

Hi, have you meet the infinitive problem?(it means that the loss or the covariance_matrices will be infinite and will assert torch.isfinite(covariance_matrices).all()
If it is possible, could you share the solution? I've tried some ways like changing the lr and the clip_grab_norm_ parameters...

[icefall7]

@fanghgit could you share your env setup? still getting issues with prerendering. like
tensorflow.python.framework.errors_impl.InvalidArgumentError: Key: roadgraph_samples/xyz. Can't parse serialized Example. [Op:ParseExampleV2]

[GT-111]

@fanghgit could you share your env setup? still getting issues with prerendering. like tensorflow.python.framework.errors_impl.InvalidArgumentError: Key: roadgraph_samples/xyz. Can't parse serialized Example. [Op:ParseExampleV2]

This is probably because you are using Waymo Motion Dataset 1.2, try to modify here in feature_description.py