multi agent trajectory prediction?

uber-research / LaneGCN

[ECCV2020 Oral] Learning Lane Graph Representations for Motion Forecasting

https://arxiv.org/abs/2007.13732

Other

507 stars 131 forks source link

Closed pjw1 closed 3 years ago

pjw1 commented 3 years ago

Thank you for sharing your work and neat code!

In your paper (3.5 Learning), the loss is used for training multiple agent. (M is the number of agent in the scene)

But the pretrained checkpoint (36.000.ckpt) seems only predicting the first agent for computing scores such as minADE.

Do you also have the weight checkpoint for multi agent? If so, can you share one? Training process is too heavy for my server ;(

pjw1 commented 3 years ago

Sorry. I just confused.

The checkpoint and the train code is for training multiple agent.