Super Resolution using Deep Convolutional Neural Network using theano
Python
Latest commit 6f28537 Jun 21, 2016 corochann critical bug fix
Permalink
Failed to load latest commit information.
assets/images upload images May 10, 2016
code
data/training_images upload images May 10, 2016
model update README May 12, 2016
.gitignore upload images May 10, 2016
LICENSE
README.md Add paragraph May 12, 2016

README.md

theanonSR

Super Resolution using Deep Convolutional Neural Network(SRCNN) using theano

Introduction

theanonSR upscales photo image to x2 size.

Original image

slide

Upscaled image using python OpenCV library

slide

Upscaled image using theanonSR

slide

Description

It is developed on python using theano library.

This project is to understand/study how deep convolutional neural network works to learn super resolution of the image.

[TODO] Currently GPU support is not implemented yet.

References

Originally, I was inspired this project from waifu2x project, which uses Torch7 to implement SRCNN.

  • Chao Dong, Chen Change Loy, Kaiming He, Xiaoou Tang, "Image Super-Resolution Using Deep Convolutional Networks", http://arxiv.org/abs/1501.00092

    SRCNN, super resolution using deep convolutional neural network, is introduced in this paper.

  • waifu2x

    It is the popular project for image super resolution for Anime-Style art. It also has a good performance.

  • theano

    Machine learning library which can be written in python. It also provides nice Deep Learning Tutorials to study how to implement deep neural network.

How to use

Basic usage

Just specify image file path which you want to upscale.

Ex. Upscaling input.jpg

python code/srcnn.py input.jpg

Specify output file name and path

Ex. Upscaling /path/to/input.jpg to /path/to/output.jpg

python code/srcnn.py /path/to/input.jpg /path/to/output.jpg

Specify model to use:

You can specify directory name in the /model directory, as the model.

Ex. use model 32x3x3_32x3x3_32x3x3_1x3x3,

python code/srcnn.py -m 32x3x3_32x3x3_32x3x3_1x3x3 input.jpg

Training

You can construct your own convolutional neural network, and train it easily!

1. Data preparation

Put training images[1] inside data/training_images directory. (I used 2000 photo images during the training.)

[1]: Currently, image must be more than or equal to the size 232 x 232.

2. Construct your model (convolutional neural network)

Open code/tools/generate_model.py, and modify this code to construct your own model. Then execute it.

python code/tools/generate_model.py

It will generate train.json file for your own model at model/your_model directory.

3. Training the model

Once prepared your own model to be trained, you can train your model by

python code/train_srcnn.py -m your_own_model

train_srcnn.py refers model/your_own_model/train.json to construct CNN (Convolutional Neural Network) for training.

Contribution is welcome

The performance of SR for this project is not matured. You are welcome to improve & contribute this project. If you could get any model which performs better performance, feel free to send me a pull request!