An implement of SRGAN(Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network) for tensorflow version
download this repo by the following instruction:
$ git clone https://github.com/nnuyi/SRGAN.git
$ cd SRGAN
Firstly, you need to make some directories in the root path(in SRGAN directory)
$ mkdir data
$ cd data
$ mkdir train
$ mkdir val
$ mkdir test
In this repo, I use parts of ImageNet datasets as train data, here you can download the datasets that I used.
After you have download the datasets, copy ImageNet(here I only use 3137 images) datsets to /data/train, then you have /data/train/ImageNet path, and training images are stored in /data/train/ImageNet
I crop image into 256*256 resolution, actually you can crop them according to your own.
Set5 dataset is used as val data, you can download it here.
After you download Set5, please store it in /data/val/ , then you have /data/val/Set5 path, and val images are stored in /data/val/Set5
Set14 dataset is used as test data, you can download it here.
After you download Set14, please store it in /data/test/ , then you have /data/test/Set14 path, and val images are stored in /data/test/Set14
$ python main.py --is_training=True --is_testing=False
$ python main.py --is_training=False --is_testing=True
low resolution | high resolution GT | high resolution GEN |
---|---|---|
sampling image |
---|
256*256 resolution left:GT right:GEN |
Email:[email protected]