https://github.com/nilboy/pixel-recursive-super-resolution

Tensorflow implementation of pixel-recursive-super-resolution(Google Brain paper: https://arxiv.org/abs/1702.00783)
https://github.com/nilboy/pixel-recursive-super-resolution

pixelcnn pixelrnn super-resolution tensorflow

Last synced: 22 days ago
JSON representation

Tensorflow implementation of pixel-recursive-super-resolution(Google Brain paper: https://arxiv.org/abs/1702.00783)

• Host: GitHub
• URL: https://github.com/nilboy/pixel-recursive-super-resolution
• Owner: nilboy
• License: mit
• Created: 2017-02-21T14:18:31.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2019-05-07T14:28:17.000Z (over 5 years ago)
• Last Synced: 2024-08-05T17:38:51.553Z (4 months ago)
• Topics: pixelcnn, pixelrnn, super-resolution, tensorflow
• Language: Python
• Homepage:
• Size: 292 KB
• Stars: 485
• Watchers: 30
• Forks: 147
• Open Issues: 14
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-hacking-lists - nilboy/pixel-recursive-super-resolution - Tensorflow implementation of pixel-recursive-super-resolution(Google Brain paper: https://arxiv.org/abs/1702.00783) (Python)

README

        
# Pixel Recursive Super Resolution

TensorFlow implementation of [Pixel Recursive Super Resolution](https://arxiv.org/abs/1702.00783). This implementation contains:

![model](./assets/model.png)

## Requirements

- Python 2.7

- [Skimage](http://scikit-image.org/)

- [TensorFlow](https://www.tensorflow.org/) 1.0

## Usage

First, download data [celebA](http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html)

    $ mkdir data

	$ cd data

	$ ln -s $celebA_path celebA

Then, create image_list file:

	$ python tools/create_img_lists.py --dataset=data/celebA --outfile=data/train.txt

To train model on gpu:

	$ python tools/train.py

	(or $ python tools/train.py --device_id=0)

To train model on cpu:

	$ python tools/train.py --use_gpu=False

## Samples

Training after 30000 iteration.

![sample.png](./assets/sample.png)

## Training details

cross entropy loss:

![curve.png](./assets/curve.png)

## Author

nilboy / [@nilboy](https://github.com/nilboy)