yunjey

Merge branch 'master' of https://github.com/yunjey/dtn-tensorflow

# Domain Transfer Network (DTN)
<b>(Work in progress)</b>Domain Transfer Network. TensorFlow implementation of [Unsupervised Cross-Domain Image Generation](https://arxiv.org/pdf/1611.02200v1.pdf). You can download pdf file [here](https://github.com/yunjey/domain-transfer-network-tensorflow/raw/master/DTN.pdf).
TensorFlow implementation of [Unsupervised Cross-Domain Image Generation.](https://arxiv.org/abs/1611.02200)
![alt text](jpg/dtn.jpg)
## Requirements
* [Python 2.7](https://www.continuum.io/downloads)
* [TensorFlow 0.12](https://github.com/tensorflow/tensorflow/tree/r0.12)
* [Pickle](https://docs.python.org/2/library/pickle.html)
* [SciPy](http://www.scipy.org/install.html)
<br>
## Usage
#### Clone the repository
```bash
$ git clone https://github.com/yunjey/dtn-tensorflow.git
$ cd dtn-tensorflow
```
#### Download the dataset
```bash
$ chmod +x download.sh
$ ./download.sh
```
#### Resize MNIST dataset to 32x32
```bash
$ python prepro.py
```
#### Pretrain the model f
```bash
$ python main.py --mode='pretrain'
```
#### Train the model G and D
```bash
$ python main.py --mode='train'
```
#### Transfer SVHN to MNIST
```bash
$ python main.py --mode='eval'
```
<br>
## Results
#### From SVHN to MNIST
![alt text](jpg/svhn_mnist_2900.gif)
![alt text](jpg/svhn_mnist_2900.png)
![alt text](jpg/svhn_mnist_3700.png)
![alt text](jpg/svhn_mnist_5300.png)
#### From Photos to Emoji (in paper)
![alt text](jpg/emoji_1.png)
![alt text](jpg/emoji_2.png)
![alt text](jpg/dtn.jpg)
......

99.3 KB | W: | H:

139 KB | W: | H:

  • 2-up
  • Swipe
  • Onion skin