mahlermozart / videosuperresolution Goto Github PK

A collection of state-of-the-art video or single-image super-resolution architectures, reimplemented in tensorflow. (currently no video models...)

The hyperlink directs to paper site, follows the official codes if the authors open sources.

• Classic

  1. SRCNN

• CVPR 2016

  2. Efficient Sub-Pixel Convolutional Network: ESPCN
  3. Very Deep Convolutional Networks: VDSR
  4. Deeply-Recursive Convolutional Network: DRCN

• CVPR 2017

  5. Deep Recursive Residual Network: DRRN https://github.com/tyshiwo/DRRN_CVPR17
  6. Deep Laplacian Pyramid Networks: LapSRN https://github.com/phoenix104104/LapSRN
  7. Enhanced Deep Residual Networks: EDSR

• ICCV 2017

  8. Memory Network: MemNet https://github.com/tyshiwo/MemNet

• CVPR 2018

  9. Information Distillation Network: IDN https://github.com/Zheng222/IDN-Caffe
  10. Residual Dense Network: RDN https://github.com/yulunzhang/RDN
  11. Super-Resolution Network for Multiple Degradations: SRMD https://github.com/cszn/SRMD
  12. Deep Back-Projection Networks: DBPN https://github.com/alterzero/DBPN-Pytorch
  13. Zero-Shot Super-Resolution: ZSSR https://github.com/assafshocher/ZSSR

• Others

  14. DNCNN (This is for denoise) https://github.com/cszn/DnCNN
  15. Deep CNN with Skip Connection: DCSCN https://github.com/jiny2001/dcscn-super-resolution

All these models are implemented in ONE framework.

This package offers a training and data processing framework based on TF. What I made is a simple, easy-to-use framework without lots of encapulations and abstractions. Moreover, VSR can handle raw NV12/YUV as well as a sequence of images as inputs.

git clone https://github.com/loseall/VideoSuperResolution && cd VideoSuperResolution
pip install -e .

Require: tensorflow-gpu, numpy, PIL

• Dataset offers manipulation of virtual images. Virtual means the images can be either a single png file or a list of jpeg files as a sequence of images. One need to provide a JSON config file, see here.
• Loader offers BatchLoader object to generate image patches with HR and LR pairs
• VirtualFile is internally used object to depict the file object

• Environment offers a simple framework specific to super resolution. See examples for instance. And see Environment.py for details.
• Callbacks offers a collection of callback functions used in Environment.fit, Environment.test and Environment.predict
• SuperResolution is the parent object to all models

Offers a collection of implementations for recent papers and research works.

To avoid storing a mess of images in codebase, I offer you links to widely used database and a configuration file to describe your own datasets. For config file, see here as a sample, and here for details.

• SET 5
• SET 14
• BSD100
• SunHays80
• Urban100

Above links are from jbhuang0604

• BSD300
• BSD500
• 91-Image
• DIV2K
• Waterloo
• MCL-V This is video datasets

You can either train via Environment object or via your own script. You can also use pre-made script to train the models in VSR package. See readme for details.

• MemNet
• ZSSR
• SRMD