JunweiLiang / Multiverse

Dataset, code and model for the CVPR'20 paper "The Garden of Forking Paths: Towards Multi-Future Trajectory Prediction". And for the ECCV'20 SimAug paper.

Home Page:https://next.cs.cmu.edu/multiverse/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Data : How to generate pedestrian to scene feature mapping?

AnuragKatakkar opened this issue · comments

Hi, we are trying to generate scene features for the ActEV/VIRAT dataset. How does one get mapping from pedestrian id to scene features?

Also, what do the various parts of the tray_key "VIRAT_S_000005_1584_0" correspond to?

Could you provide more information? Which data file do you refer to?

The traj_key in multiverse can be interpreted according to here.

For scene feature processing, we first compute the correct scene grid used in the models based on the number of convolutions we used and the original scene feature sizes. (see here). Then we map the correct whole scene feature frame to each pedestrian (see here). At run time, we apply convolution on top of the original scene features (see here) and then apply the one-hot location mask of each pedestrian at each time-step to get the features (see here).

Does the above answer your questions?