Skip to content

Golbstein/EDSR-Keras

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

27 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

EDSR-Keras

EDSR Super-Resolution Implementation with Keras

Keras implementation of the paper "Enhanced Deep Residual Networks for Single Image Super-Resolution" from CVPRW 2017, 2nd NTIRE: EDSR Paper

alt text

Model Architecture

alt text

Extensions

  1. Training with multi loss - MAE + VGG16 Perceptual Loss
  2. float16 and float32 support
  3. Keras Subpixel (Pixel-Shuffle layer) from: Keras-Subpixel
  4. ICNR weights initialization - Checkerboard artifact free sub pixel convolution initialization, credit also for @kostyaev for the implementation of the initializer here: https://github.com/kostyaev/ICNR

Training from scratch in float16 with multi-loss doesn't work. Set to float32

Dependencies

  • Python 3.6
  • Keras>2.0.x
  • keras-tqdm (pip install keras-tqdm)

Results of multi-loss

  • Dataset: Pascal VOC 2012
  • n_feats = 64
  • n_resblocks = 8

alt text

alt text