search

NVlabs / PWC-Net

PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume, CVPR 2018 (Oral)
Other
1.63k stars 357 forks source link

3 convolutional layers instead of 2 per level to extract feature pyramid perform better? #112

Open limpbot opened 4 years ago

limpbot commented 4 years ago

Hi guys, in your pytorch implementation you are using 3 convolutional layers per level to extract the feature pyramid. In the paper you only talk about two layers, are there insights that 3 layers perform better?

Best, Leo

limpbot commented 4 years ago

ok i found it in the ablation studies, my bad.