A Tensorflow 2.x based implementation of
-
Enhanced Deep Residual Networks for Single Image Super-Resolution (EDSR), winner of the NTIRE 2017 super-resolution challenge.
-
Enhanced Super Resolution Generative Adversarial Network (by Xintao Wang et.al.) (ESRGAN)
-
Wide Activation for Efficient and Accurate Image Super-Resolution (WDSR), winner of the NTIRE 2018 super-resolution challenge (realistic tracks).
A DIV2K data provider automatically downloads DIV2K
training and validation images of given scale (2, 3, 4 or 8) and downgrade operator ("bicubic", "unknown", "mild" or
"difficult").
Single image super resolution aims to construct a high-resolution image (HR) from a low resolution image (LR). This is essentially done by upsampling the LR image to HR image. Deep learning models are utilized to perform this upsampling. Loss functions are used to measure reconstruction error and guide the model optimization.
- It is preferable to have a GPU ( cuda support in Tensorflow 2.x) but it works otherwise too.
Clone this repository or download the master branch as a zip file to your computer. Navigate to your downloaded directory and run the below commands:
Step 1: Create a python virtual environment
python -m venv sr
Scripts\activate
Step 2: install the requirements in the environment
python -m pip install -r requirements.txt
Step 3: Run ESRGAN (esrgan.ipynb) or EDSR (example-wdsr.ipynb) or WDSR (example-wdsr.ipynb)
Note: You can skip training the models and directly use the model weights.
In order to run with your own input, resize the input image to width: 124 and height 118 pixels. Store the input image inside the demo folder and run the following code cell:
resolve_and_plot('demo/<imgfile>')
PSNR: 5.1907578
PSNR value: 29.865347
Source: