How to accelerate the Correlation layer?

NVIDIA / flownet2-pytorch

Pytorch implementation of FlowNet 2.0: Evolution of Optical Flow Estimation with Deep Networks

Other

3.14k stars 740 forks source link

How to accelerate the Correlation layer? #58

Open PkuRainBow opened 6 years ago

PkuRainBow commented 6 years ago

I find that the computation for the correlation layer is very time-consuming. Could you give me some advice on how to accelerate the training with the Correlation layer~

fitsumreda commented 6 years ago

Have you cloned recently? The backward for correlation has been sped up by about 5x.

PkuRainBow commented 6 years ago

@fitsumreda Thanks!

It seems that the project I have cloned is within one month.