Python packages needed:
- tensorflow
- keras
- numpy
- opencv-python
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
- Simple HLS color filter with low Hue values
python run.py --color-mask-light
- Simple HLS color filter with high Hue values
python run.py --color-mask-dark
- Gaussian mixture background subtractor
python run.py --gauss-mixture
- K-nearest neighbours background subtractor
python run.py --knn
- HGR-Net segmentation model from https://arxiv.org/abs/1806.05653
python run.py --hgr-net
- 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
To save model output to file use flag --save
. It will be saved to local directory model_rec