search

alterzero / RBPN-PyTorch

The project is an official implement of our CVPR2019 paper "Recurrent Back-Projection Network for Video Super-Resolution"
https://alterzero.github.io/projects/RBPN.html
MIT License
449 stars 104 forks source link

is this what training is supposed to look like? will also randomly get "all the pixels are invalid...." #36

Closed crypticsymmetry closed 5 years ago

crypticsymmetry commented 5 years ago

it will randomly say "All the pixels are invalid in estimation Laplacian noise" while training. is it okay?

mostly the training looks like this. ..................................................................... Pyramid level 2 Pyramid level 1 Pyramid level 0 Constructing pyramid...done! Pyramid level 2 Pyramid level 1 Pyramid level 0 Constructing pyramid...done! Pyramid level 2 Pyramid level 1 Pyramid level 0 Constructing pyramid...done! Pyramid level 2 Pyramid level 1 Pyramid level 0 ===> Epoch1: Loss: 13.6273 || Timer: 0.1491 sec.

alterzero commented 5 years ago

Don't forget to set bool OpticalFlow::IsDisplay=false (pyflow/src/OpticalFlow.cpp) to remove the "Pyramid level" text.

By having that error, I assume your flow map is not accurate. But until now I have no time to investigate the pyflow. I can only say you need to use Python 3.5.2 to compile it.