Human-Data-Generation-Framework

This repository contains the code for generating the data described in "Efficient Realistic Data Generation Framework leveraging Deep Learning-based Human Digitization"

Requirements

• Python3
• Pyglet
• Pycocotools
• Opencv
• Matplotlib
• Scikit-image
• PyWavefront
• Pickle5
• Pillow

Install the required packages

pip install -r requirements.txt

Download and reformat the CityScapes dataset and the human models

1. Run the following shell script in order to download the 3D human models and other files neccessary for the generation process. The 3D human models are generated through https://github.com/shunsukesaito/PIFu, which is provided under the MIT licence, using images from the Clothing Co-Parsing (CCP) dataset as input, which are provided under the Apache 2.0 licence.

mkdir -p ./background_images/CityScapes/in
mkdir -p ./background_images/CityScapes/out
chmod +x download_files.sh
./download_files.sh

2. Download the CityScapes dataset from www.cityscapes-dataset.net
   
   • RGB images: (a) leftImg8bit_trainvaltest.zip, (b) leftImg8bit_trainextra.zip
     
   • Annotation images: gtCoarse.zip
     

The folder hierarchy should look like this:

├─ background_images
|  ├─ CityScapes
|     └─ in
|     |   ├─ leftImg8Bit
|     |   └─ gtCoarse
|     └─ out
|      
...

3. Run the following script to reformat the CityScapes dataset.

python reformat_cityscapes.py
python create_background_images.py

4. Run the following script to generate the dataset.

python create_dataset.py

Citation

If you make use of the dataset, please cite the following reference in any publications:

@inproceedings{symeonidis2021data,
  title={Efficient Realistic Data Generation Framework leveraging Deep Learning-based Human Digitization},
  author={Symeonidis, C. and Nousi, P. and Tosidis, P. and Tsampazis, K. and Passalis, N. and Tefas, A. and Nikolaidis, N.}
  booktitle={Proceedings of the International Conference on Engineering Applications of Neural Networks (EANN)},
  year={2021}
}