Training on NTU 60 XSUB

[erinchen824]

Thanks for your awesome work and generous code sharing.
Recently, I have tried to reproduce the results on NTU 60 XSUB benchmark with one 3090 GPU.
However, it seems the training is really slow and it would take more than a week to train 20 epochs following the original setting.
I have tried to enlarge the batch size, and it still quite slow.
So, what is the reasonable training time for NTU 60 dataset?
Is there anything I have missed?

Looking forward to your reply.

Thanks

[erinchen824]

Thanks for your awesome work and generous code sharing. Recently, I have tried to reproduce the results on NTU 60 XSUB benchmark with one 3090 GPU. However, it seems the training is really slow and it would take more than a week to train 20 epochs following the original setting. I have tried to enlarge the batch size, and it still quite slow. So, what is the reasonable training time for NTU 60 dataset? Is there anything I have missed?

Looking forward to your reply.

Thanks

I modify the dataloader to only sample one sequence for each video, and here are part of the training logs.

[hehefan]

Hi,

I think the training time is reasonable. One reason is that we used all frames with all combinations, which is a common problem with video processing. Selecting a fixed frame combination for each video can significantly reduce time but may slightly affect the accuracy. Another reason is that we did not implement PSTConv in a parallel way like the standard convolutions in torch.nn. Point tubes are sequentially constructed over time.

Best

[erinchen824]

Hi,

I think the training time is reasonable. One reason is that we used all frames with all combinations, which is a common problem with video processing. Selecting a fixed frame combination for each video can significantly reduce time but may slightly affect the accuracy. Another reason is that we did not implement PSTConv in a parallel way like the standard convolutions in torch.nn. Point tubes are sequentially constructed over time.

Best

Thanks for your quick reply, that resolves my question.