nvlabs / adaptive-prediction Goto Github PK

For pedestrian tasks, it seems that only 1 sample is allowed for evaluation but I guess all baselines are using 20 samples. Can you check it? Thanks.

Hi there,

I wonder how many trainable parameters are in the model? what about the model's size? and FLOPs?

Best
Izzeddin

Hello,

Thank you for the repo. Can you please share the exact ETH/UCY data splits used for creating table 1 in the paper?

Greetings,

Hi can you incorporate these changes, saw this repository later. I think the import issue is fixed here but the float one is still present.

Best Regards
@shikharvashistha

Hi, thanks for sharing the code.
In fact, I found it hard to reproduce the result of PECNet in table 1, such as the 1.97(ADE)/ 3.33(FDE) on A2B dataset.
I think it's because the specific config or dataset format of PECNet. Can you share your reproduce procedure?