Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Questions about loss function #30

Open
ChengkaiYang opened this issue Oct 21, 2024 · 0 comments
Open

Questions about loss function #30

ChengkaiYang opened this issue Oct 21, 2024 · 0 comments

Comments

@ChengkaiYang
Copy link

Dear @MasterIzumi
I noticed that your loss function calculated all agents in a scenario but not only ego agent. On AV2 single agent motion forecasting leaderboard, SIMPL got 1.43 minFDE6 metrics which seems great. I wonder if it could get more accuracy on single agent motion forecasting leaderboard when involving all agents into loss function but not only ego agent?Other SOTA work like QCNet utilizing all agents' ground truth trajectory in loss function as well.
In fact, I studied some previous work like Autobot (2022).The original code of Autobot didn't take RPE into encoder.After i add rpe module into Autobot, it got only 1.7 minFDE6 on AV2 validation dataset.I wonder if it's because i only use ego agent into loss function?Have you ever explore this?
I'd appreciate if you could reply!Thank you so much.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant