Simple Deepstream pipeline uses Retinaface as inference.
Demostrate how to parse raw-output and draw landmarks on osd.
- Deepstream 6.0
- GStreamer 1.14.5
- Cuda 11.4+
- NVIDIA driver 470.63.01+
- TensorRT 8+
Follow deepstream official doc to install dependencies.
Deepstream docker is more recommended.
Please refer to this repo for pretrained models and serialized TensorRT engine.
git clone https://github.com/mrscarletzhou/retinaface-deepstream-python.git
modify tensorRT engine network size in retina-engine-config file
python3 main.py {VideoPath}