victorca25
commented
5 years ago - Added learning schemes with restarts, to have similar options to the PPON paper. Currently StepLR_Restart is not yet working, but MultiStepLR_Restart does work and manual steps to change the LR can be added to have the same behavior as in the PPON paper.
- Added a control mechanism for MS-SSIM for when the patch size is not large enough to allow images at the smallest scales to be larger than the defined kernel size window for the Gaussian function.
- Some general fixes to the code and to make the training more similar to the PPON paper.