Mixed-Bg-Fg-Synthesis via Conditional GLO

Official Implementation of Stingray Detection of Aerial Images Using Augmented Training Images Generated by A Conditional Generative Model

Created by Yi-Min Chou, Chien-Hung Chen, Keng-Hao Liu, Chu-Song Chen

The overview of data augmentation

1. Crop the object patches (rotate and flip to augment data) and randomly crop the background patches to establish the dataset for training conditional GLO.
2. Training conditional GLO and use the well-trained model to generate the fake stingray image.
3. Paste the generated stingray patches to original positions.
4. Use the augmented data generated from C-GLO to train detection models.

Conditional-Generative-Latent-Optimization

Prerequisition

• Python 2.7
• TensorFlow 1.4.0 or higher

How to Run

• Clone the Mixed-Bg-Fg-Synthesis repository:

$ git clone --recursive https://github.com/ivclab/ConditionalGLO.git

• Install required packages:

$ pip install -r requirements.txt

• Download Stingray Data:

$ python download.py

• Run the training code:

# The training result will be saved in `./logs/FOLDER_NAME/`
$ python main.py --is_train=True

• Run the testing code:

# The testing result will be saved in `./logs/FOLDER_NAME_test/`
$ python main.py --is_train=False --load_path=FOLDER_NAME

Experimental Results

Original background image(top), mixed background and foreground synthesis generated by C-GLO (bottom)

Citation

Please cite following paper if these codes help your research:

@inproceedings{chou2018stingray,
  title={Stingray Detection of Aerial Images Using Augmented Training Images Generated by a Conditional Generative Model},
  author={Chou, Yi-Min and Chen, Chien-Hung and Liu, Keng-Hao and Chen, Chu-Song},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops},
  pages={1403--1409},
  year={2018}
}

@inproceedings{
  title   = {Changing Background to Foreground: An Augmentation Method Based on Conditional Generative Network for Stingray Detection},
  Author  = {Chou, Yi-Min and Chen, Chien-Hung and Liu, Keng-Hao and Chen, Chu-Song}, 
  booktitle = {IEEE International Conference on Image Processing, ICIP},
  year    = {2018}
}

Contact

Please feel free to leave suggestions or comments to Yi-Min Chou, Chien-Hung Chen([email protected]), Keng-Hao Liu([email protected]), Chu-Song Chen([email protected])