search

anuragranj / spynet

Spatial Pyramid Network for Optical Flow
Other
234 stars 48 forks source link

Training time written on your paper. #13

Closed jangho2001us closed 6 years ago

jangho2001us commented 6 years ago

I read your paper, SpyNet, and now i am trying to implement your model in PyTorch. I understand you trained network, G0 for three days and it takes a day to train other sub networks. As written on your paper, sub network G0 takes 24*32 images and i think this model is trained promptly. What's your opinion about this...?

anuragranj commented 6 years ago

Yes, that's right. That was about 2 years ago, on a single TitanX. I think you could do it much faster with newer GPUs, and cudnns.