EDSR-tensorflow2

Tensorflow 2 implementation of Enhanced Deep Residual Networks for Single Image Super-Resolution(CVPR2017).
Most of the existing codes performed data augmentation using matlab, opencv, and numpy. This was replaced by tensorflow's ImageDataGenerator, and tensorflow 2's methods were used as a whole.

Usage

$ mkdir models
$ main.py [-h] TARGET_SIZE N_TRAIN_DATA N_TEST_DATA N_RES_BLOCK BATCH_SIZE EPOCHS SCALE SAVE_NAME

DIV2K example

$ mkdir models
$ python main.py 240 800 100 32 16 400 4 edsr_div2k.hdf5

Prerequisites

• Python 3.7
• Tensorflow 2
• Numpy

Directory

EDSR-tensorflow2
├── main.py              // main program
├── model.py             // edsr model
├── data_generator.py    // data augmentation
└── utils.py             // psnr, mae

Sample Results

• trained by DIV2K
• test image set is Set14