search

schmidt-ju / crat-pred

Other
62 stars 11 forks source link

Visualisation/Quantitative results. #9

Closed djamelbenr closed 1 year ago

djamelbenr commented 1 year ago

Dear authors,

I would like to thank you for making your remarkable work publicly accessible. I have two questions regarding the shared code and derived results (saved training weights, mainly);

1-Is there a way to present them in a visually appealing format, such as visualizations or quantifiable results? I believe that being able to visualize the prediction outcomes in a multi-agent predictive system would greatly improve my understanding and provide a more intuitive understanding of the results.

2-I am also curious if it would be possible to train and test the CRAT-Pred algorithm using other BVE datasets, such as the HighD or NGSIM dataset.

Thank you for your time and consideration. I am eagerly awaiting your response :)

Cram3r95 commented 1 year ago

Hi @djamelbenr,

You can use our qualitative results plot script:

https://github.com/Cram3r95/mapfe4mp/blob/master/evaluate/argoverse/generate_results_rel-rel.py

Since the model output is exactly the same than CRAT-PRED, that is, relative displacements.

djamelbenr commented 1 year ago

Oh! sounds good :) Thanks!!

schmidt-ju commented 1 year ago

Hey :D

thanks for helping each other out! Is the plotting script working for you?

The code is a complete rework from the original paper version, so the original weights do not work anymore. I once ran a training and achieved similar results with the code in this repository. However, I don't think I still have the weights. Isn't the training working for you? Should be easy to reproduce results similar to those in the paper.

1-Already solved. Thanks @Cram3r95! 2-For sure! You just have to adapt the dataloading.

Looking forward to hearing from you!

Julian

schmidt-ju commented 1 year ago

Closed due to inactivity.