Pixel Recursive Super Resolution
TensorFlow implementation of Pixel Recursive Super Resolution. This implementation contains:
Requirements
- Python 2.7
- Skimage
- TensorFlow 1.0
Usage
First, download data celebA
$ 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.
Training details
cross entropy loss:
Author
nilboy / @nilboy