search

InhwanBae / EigenTrajectory

Official Code for "EigenTrajectory: Low-Rank Descriptors for Multi-Modal Trajectory Forecasting (ICCV 2023)"
https://ihbae.com/publication/eigentrajectory/
MIT License
86 stars 5 forks source link

How to determine the number of tracks normalized at a time #3

Open 12num opened 1 year ago

12num commented 1 year ago

Thanks for your work

You propose to use SVD to construct ET space for normalized trajectory A(L*N). I'm not quite sure if N here represents the number of all tracks in each training set ("eth" "hotel" "univ" "zara1" "zara2")?

InhwanBae commented 1 year ago

Hi @12num, Exactly. Before starting the training, all trajectories within the training dataset are decomposed using SVD to construct ET space. It will be much easier to understand if you look at the code below.

https://github.com/InhwanBae/EigenTrajectory/blob/a2596fca73e5c30188c1572930b1bb773aa04167/utils/trainer.py#L48-L54

12num commented 1 year ago

I see, thank you very much!