search

ZikangZhou / HiVT

[CVPR 2022] HiVT: Hierarchical Vector Transformer for Multi-Agent Motion Prediction
https://openaccess.thecvf.com/content/CVPR2022/papers/Zhou_HiVT_Hierarchical_Vector_Transformer_for_Multi-Agent_Motion_Prediction_CVPR_2022_paper.pdf
Apache License 2.0
577 stars 115 forks source link

HiVT map-free customization #46

Open javierAraluce opened 10 months ago

javierAraluce commented 10 months ago

Hello, first of all, thank you for your excellent work and dedication in making it openly available for the community to use.

I was wondering how I could adapt your code so that it can be evaluated in a version without a map, as they claim in the following paper. They have compared HiVT with your model in a way that doesn't use a map.

Thanks in advance

EtienneDS commented 9 months ago

I believe the author(s) have moved on to their new algorithm QCNet and such won't really put too much work into this one anymore.

I believe to make it work without map all you need to do is remove the ALEncoder in local_encoder.py : https://github.com/ZikangZhou/HiVT/blob/6876656ce7671982ebdc29113aaaa028c2931518/models/local_encoder.py#L64-L68 https://github.com/ZikangZhou/HiVT/blob/6876656ce7671982ebdc29113aaaa028c2931518/models/local_encoder.py#L93-L96

You can also remove the mapping sections from the data preprocessing if you are using Argoverse: https://github.com/ZikangZhou/HiVT/blob/6876656ce7671982ebdc29113aaaa028c2931518/datasets/argoverse_v1_dataset.py#L152-L157

With this you should be able to run the code without map information, though I cannot say if this is the best way as to get optimal results.