Skip to content

An implementation of the Fast Super-Resolution Convolutional Neural Network in TensorFlow with a focus on artifact mitigation

License

Notifications You must be signed in to change notification settings

HelpSeeker/FSRCNN-TensorFlow

 
 

Repository files navigation

FSRCNN-TensorFlow

TensorFlow implementation of the Fast Super-Resolution Convolutional Neural Network (FSRCNN). This implements two models: FSRCNN which is more accurate but slower and FSRCNN-s which is faster but less accurate. Based on this project.

This fork

Nothing special about it. Just my playground to experiment with FSRCNNX's distortion capabilities. See DISTORT.md for more infos.

Prerequisites

  • Python 3
  • TensorFlow-gpu >= 1.8
  • CUDA & cuDNN >= 6.0
  • Pillow
  • ImageMagick (optional)
  • Wand (optional)

Usage

For training: python main.py
For testing: python main.py --train False

To use FSCRNN-s instead of FSCRNN: python main.py --fast True

Can specify epochs, learning rate, data directory, etc:
python main.py --epoch 100 --learning_rate 0.0002 --data_dir Train
Check main.py for all the possible flags

Result

Original butterfly image:

orig

Ewa_lanczos interpolated image:

ewa_lanczos

Super-resolved image:

fsrcnn

Additional datasets

TODO

  • Add RGB support (Increase each layer depth to 3)

References

About

An implementation of the Fast Super-Resolution Convolutional Neural Network in TensorFlow with a focus on artifact mitigation

Resources

License

Stars

Watchers

Forks

Languages

  • Python 100.0%