Ego vs. other agents training

[YasinSonmez]

I'm trying to understand the training process. I'm not sure if trajectory generation for the ego vehicle is handled differently from the prediction of other agents. Are there any specific considerations for the ego vehicle compared to the other agents during the training of the transformer model or are we doing the same thing for ego and other agents?

[jchengai]

Hi @YasinSonmez, data augmentations are only applied to the ego vehicle to address closed-loop problems. Prediction serves as an auxiliary task, providing dense supervision, and is not the primary focus of this study.

[YasinSonmez]

I see, can the trajectories of other vehicles be used in the training as if they were ego vehicles to augment the data?

…

On Thu, Jun 27, 2024 at 8:40 PM Jie Cheng ***@***.***> wrote: Hi @YasinSonmez <https://github.com/YasinSonmez>, data augmentations are only applied to the ego vehicle to address closed-loop problems. Prediction serves as an auxiliary task, providing dense supervision, and is not the primary focus of this study. — Reply to this email directly, view it on GitHub <#22 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AI6FYHNEACGM56WZ2AR2KRDZJTLLTAVCNFSM6AAAAABJ637WF2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCOJWGA3TIMZZHE> . You are receiving this because you were mentioned.Message ID: ***@***.***>