Hand segmentation models tests

Prerequisites

Python packages needed:

• tensorflow
• keras
• numpy
• opencv-python

Usage

The main script is started with command python run.py It will take video from the default webcam. To set source explicitly use flag --source <source>:

• Webcam number N: python run.py --source N
• File: python run.py --source <path>/input.avi

Models

1. Simple HLS color filter with low Hue values python run.py --color-mask-light
2. Simple HLS color filter with high Hue values python run.py --color-mask-dark

3. Gaussian mixture background subtractor python run.py --gauss-mixture

4. K-nearest neighbours background subtractor python run.py --knn

5. HGR-Net segmentation model from https://arxiv.org/abs/1806.05653 python run.py --hgr-net

6. HGR-Net segmentation model from https://arxiv.org/abs/1806.05653 with dense ASPP from http://openaccess.thecvf.com/content_cvpr_2018/html/Yang_DenseASPP_for_Semantic_CVPR_2018_paper.html python run.py --hgr-net-dense

Saving output

To save model output to file use flag --save. It will be saved to local directory model_rec