search

XiaolongTang23 / HPNet

[CVPR 2024] HPNet: Dynamic Trajectory Forecasting with Historical Prediction Attention
Apache License 2.0
96 stars 11 forks source link

Encoding Agent Features. #11

Open ilzkl1991 opened 1 month ago

ilzkl1991 commented 1 month ago

thank you for your work.

when i read the code, i found that in the paper, the agent features are the velocity magnitude, the direction of velocity, and the attribute. but in the code , the agent features are the length of the vector, it seems the features only have the length, without the direction. Can you explain why you don't use the features like vector, velocity?

XiaolongTang23 commented 1 month ago

I guess you are looking at the HPNet-Argoverse code. The reason we only use the length is that Argoverse does not provide velocity-related information. The INTERACTION dataset does provide these details, and for that dataset, the encoding part uses more information as described in the paper.

ilzkl1991 commented 1 month ago

I guess you are looking at the HPNet-Argoverse code. The reason we only use the length is that Argoverse does not provide velocity-related information. The INTERACTION dataset does provide these details, and for that dataset, the encoding part uses more information as described in the paper.

Thank you for your reply. Do you have tried to convert pth/ckpt to ONNX? If yes, is there any translation problem?