Metric inconsistency

[swc-17]

Hi,
Thanks for your work! I wonder why the metrics of provided checkpoint is better than those in paper?

[swc-17]

Also wondering the prediction performance with PETRv2 as detector, which is not mentioned in the paper.

[GentleSmile]

We rewrite the metrics for better readability before releasing the code, and details of metrics calculating have been updated, especially the minADE and minFDE.
As mentioned in the paper, we found that PETRv2 is not suitable for the detector of ViP3D. The tracking and prediction performance is quite poor.