carpedm20 / simulated-unsupervised-tensorflow

TensorFlow implementation of "Learning from Simulated and Unsupervised Images through Adversarial Training"
Apache License 2.0
575 stars 146 forks source link
apple deep-learning generative-model synthetic-images tensorflow

Simulated+Unsupervised (S+U) Learning in TensorFlow

TensorFlow implementation of Learning from Simulated and Unsupervised Images through Adversarial Training.

model

Requirements

Usage

To generate synthetic dataset:

  1. Run UnityEyes with changing resolution to 640x480 and Camera parameters to [0, 0, 20, 40].
  2. Move generated images and json files into data/gaze/UnityEyes.

The data directory should looks like:

data
├── gaze
│   ├── MPIIGaze
│   │   └── Data
│   │       └── Normalized
│   │           ├── p00
│   │           ├── p01
│   │           └── ...
│   └── UnityEyes # contains images of UnityEyes
│       ├── 1.jpg
│       ├── 1.json
│       ├── 2.jpg
│       ├── 2.json
│       └── ...
├── __init__.py
├── gaze_data.py
├── hand_data.py
└── utils.py

To train a model (samples will be generated in samples directory):

$ python main.py
$ tensorboard --logdir=logs --host=0.0.0.0

To refine all synthetic images with a pretrained model:

$ python main.py --is_train=False --synthetic_image_dir="./data/gaze/UnityEyes/"

Training results

Differences with the paper

Experiments #1

For these synthetic images,

UnityEyes_sample

Result of lambda=1.0 with optimizer=sgd after 8,000 steps.

$ python main.py --reg_scale=1.0 --optimizer=sgd

Refined_sample_with_lambd=1.0

Result of lambda=0.5 with optimizer=sgd after 8,000 steps.

$ python main.py --reg_scale=0.5 --optimizer=sgd

Refined_sample_with_lambd=1.0

Training loss of discriminator and refiner when lambda is 1.0 (green) and 0.5 (yellow).

loss

Experiments #2

For these synthetic images,

UnityEyes_sample

Result of lambda=1.0 with optimizer=adam after 4,000 steps.

$ python main.py --reg_scale=1.0 --optimizer=adam

Refined_sample_with_lambd=1.0

Result of lambda=0.5 with optimizer=adam after 4,000 steps.

$ python main.py --reg_scale=0.5 --optimizer=adam

Refined_sample_with_lambd=0.5

Result of lambda=0.1 with optimizer=adam after 4,000 steps.

$ python main.py --reg_scale=0.1 --optimizer=adam

Refined_sample_with_lambd=0.1

Training loss of discriminator and refiner when lambda is 1.0 (blue), 0.5 (purple) and 0.1 (green).

loss

Author

Taehoon Kim / @carpedm20