vita-epfl/causalmotion

[CVPR22] Towards Robust and Adaptive Motion Forecasting: A Causal Representation Perspective

causalmotion's Issues

A question about spurious shifts experiments

Thanks for sharing the code!

I'm interested in this project and I'm trying to run the experiments based on the code and paper.

However, in the spurious shifts experments, there is an interesting phenomenon. In the experiments of λ (figure 4 in the paper), my result is opposite from that in the paper, which means that λ=1 reaches the lowest ADE and λ=5 reaches the highest ADE.

I think I have followed all the settings in the paper and code (e.g. version of pytorch, hyperparameters), so I'm a little confused about this phenomenon. Could you please give me some advice about this? Is there anything special that you think should be focused on when runing the spurious shifts experiments?

Sorry for bothering you and looking forward to replys.