Skip to content

The reproduction of SRCNN method for super-resolution

Notifications You must be signed in to change notification settings

ahpu2014/SRCNN-REPRODUCTION

This branch is up to date with xxxwuwq/SRCNN-REPRODUCTION:master.

Repository files navigation

The directory structure:

.
├── datasets
|   └── Train
|   └── Test
|       ├── Set5
|       └── Set14
|── configs.py
|── networks.py
|── train.py
|── utils.py
└── README.md
        

training

  1. place the datasets(training:91 images, testing: Set5 and Set14) to datastes directory \
  2. running ultis.py to generate the traning dataset(h5 file)
  3. running train.py to start training the dataset could be obtained from this url(http://mmlab.ie.cuhk.edu.hk/projects/SRCNN/SRCNN_train.zip)

The iteration was set to 400,000,000. In the papper "Learning a Deep Convolutional Network for Image Super-Resolution, in Proceedings of European Conference on Computer Vision the iteration was set 10 1,200,000,000

this version exsist some problems which lead to bad psnr
to be continue...

About

The reproduction of SRCNN method for super-resolution

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%