Video player to compare a face from a low quality CCTV video with a 3D modeled face (generated by IRC3D, available soon).
This software is part of a research thesis of the Criminal Investigation Institute of the National Gendarmerie (IRCGN) managed by Marie-Charlotte Poilpré ([email protected]). It is based on the head pose detection in deep lerning Hopenet, by Nataniel Ruiz, Eunji Chong, James M. Rehg from Georgia Institute of Technology.
To launch the application you have to start the main.py
in the code folder
cd code/
python3 main.py
If the configuration file can not be found, you can pass it as a parameter
python3 main.py --config [MY_CONFIG_PATH]
You can configure the application from the json file config.json
snapshot
path of the Hopenet trained model to use to calculate the orientation of the face (.pkl)face_model
path of the dlib face detections model (.dat)gpu_id
the identifier of the GPU to use by default (if no gpu is detected, the cpu will be used)cache_string
text at the end of the video name for its json cache filevideo_formats
table of all videos formats that can be opened with the application (you must have the codecs to read these files)mesh_formats
table of all 3D mesh formats that can be opened with the application
- Python 3.7 or higher
- Numpy 1.7 or higher
- OpenCV 3.0 or higher
- PyTorch 1.0 or higher
- PyQT 5.0 or higher
- Pyglet 1.3.2 or higher
- PyWavefront 1.0.3 or higher
- DLIB
- Skimage
- Matplotlib
- Pillow
- Pandas
If you want to use your graphics card (recommended for faster calculation), you have to install
- Cuda
- CuDNN
To play videos with opencv, you must have some codec installed on your computer, like:
- ffmpeg or ffmpeg-python
- gstreamer-player
If you need help or advice to install the dependencies look at the file INSTALLATION.md
Nicolas SENECAL, Tom SAMAILLE, Florian TORRES, Guillaume NICOLUSSI CASTELLAN, Paul CHAPELON
IMAC engineering school