search

ChinaYi / ASFormer

Official repo for BMVC2021 paper ASFormer: Transformer for action segmentation
MIT License
94 stars 19 forks source link

Long training time #4

Closed Jaakik closed 2 years ago

Jaakik commented 2 years ago

Hello,

I am adapting your code for my own dataset which usually train relatively fast when using only ASRF, but when using your model with the transformer it's taking approximately 10x times longer. Do you have a similar behaviour with Salad/breakfast/gtea datasets ?

Thank you :)

ChinaYi commented 2 years ago

Hi. The asformer is slower due to the additional slef-attn operations. However, when trained with salads, it's taking approximately 3x times longer. Whats the average video length of your dataset? Do you set the if_warp=True in '''def next_batch(self, batch_size, if_warp=False)'''?

Jaakik commented 2 years ago

Yes thank you.