chengy12/causalhtp

The official PyTorch code implementation of "Human Trajectory Prediction via Counterfactual Analysis" in ICCV 2021.

causalhtp's Issues

Details about the Reproduced STGAT

Hi, thanks for your excellent work! May I ask the difference between your reproduced STGAT(STGAT*) and the optimised STGAT(raw)? It seems like your results on UNIV dataset do not outperform the optimised STGAT.

Question about visualization

Hi,

Thank you very much for your work and I'm very interested in trajectory prediction.

I have some problems about the visualization of predicted trajectory. How to plot the trajectories in the scene picture just like the Fig5(https://github.com/CHENGY12/CausalHTP/blob/main/images/examples_social-stgat.png) in your paper?

Can you share with me the code for the visualization of the trajectories please? Thank you very much!

Questions about the conterfactual feature (Only zero vector is used, the other two are not given)

Hi, this is an interesting work. Thanks for sharing the code.
However, in your code "models.py" from line 255
"
traj_lstm_hidden_states_c = torch.zeros_like(traj_lstm_hidden_states[-1])
"
It seems that only zero vector is used. The mean vector of all history trajectories or the random trajectories
are not shown here as the counterfactual intervention.
Could you explain that? Thank you

Question about CNN networks

Thanks for your great work! According to your description, there is a CNN branch in the framework to process the input images, but I can't find images in your "datasets" folder (only the .txt files of the trajectory sequence), and I did't find the relevant code in "models.py".