TensorFlow implementation of Learning from Simulated and Unsupervised Images through Adversarial Training.
- Python 2.7
- TensorFlow 0.12.0
- SciPy
- pillow
- tqdm
To generate synthetic dataset:
- Run UnityEyes with changing
resolutionto640x480andCamera parametersto[0, 0, 20, 40]. - ...
The data directory should looks like:
data
├── gaze
│ ├── MPIIGaze
│ │ └── Data
│ │ └── Normalized
│ │ ├── p00
│ │ ├── p01
│ │ └── ...
│ └── UnityEyes # contains images of UnityEyes
│ ├── 1.jpg
│ ├── 2.jpg
│ └── ...
├── __init__.py
├── gaze_data.py
├── hand_data.py
└── utils.py
To train a model:
$ python main.py --data_set gaze
$ python main.py --data_set hand
To test with an existing model:
$ python main.py --data_set gaze --test
$ python main.py --data_set hand --test
(in progress)
Taehoon Kim / @carpedm20