SGNet.pytorch icon indicating copy to clipboard operation
SGNet.pytorch copied to clipboard

Published 20 hours ago • verified publisher icon ChuhuaW

training on nuScenes dataset

Open orhuntamyigit opened this issue 2 years ago • 2 comments

Hi, Thank you very much for making the code available. I wanted to get results on nuScenes dataset after the Trajectron++ bug mentioned in #19 is fixed. If possible, could you share the dataloader, training interfaces etc. for nuScenes?

orhuntamyigit avatar Jun 09 '22 10:06 orhuntamyigit

Hi @Oshuttlesworth, I don't have any progress on the fixed Trajectorn++ yet, but I will try my best to conduct some experiments later. I apologize for the delay in delivering the nuScenes training code, as we do not yet have time to clean it. Meanwhile, I can share the preprocessing script, dataloader, model and training script here.

ChuhuaW avatar Jun 09 '22 23:06 ChuhuaW

Thanks a lot! In case anyone wants to compare, I conducted experiments and obtained the following results with the fixed Trajectron++ code: minADE_20 | minFDE_20 ETH - univ 0.434 0.834 ETH - hotel 0.339 0.914 UCY - univ 0.308 0.650 UCY - zara1 0.142 0.291 UCY - zara2 0.128 0.282

One more question, if possible, could you share the dataloader, training-evaluation interfaces for the HEV-I dataset? I would like to see how the results change with different tracking-flow estimation methods.

orhuntamyigit avatar Jul 05 '22 13:07 orhuntamyigit