Project Pikachu is a showcase of using pyppbox in real-life. The system can track and re-identify specific/certain people through normal 2D cameras without having to spend much effort on the calibration, etc.
demo_pikachu_h264_10MB.mp4
Main modules:
- Detector: YOLO_Classic (v4)
- Tracker: SORT
- ReIDer: Torchreid (OSNet-AIN)
Check supported modules for more details.
Watch on YouTube: https://youtu.be/jFPDbgPJBds
DABPG & DeepMVPG, state-of-art people grouping methods, were built on top of pyppbox, thanks to pyppbox with its robust person object class which offers a solid infrastructure for easily managing detected individuals.
The demos below rely on the provided ground-truth (Converted to pyppbox format) of MOT dataset for people detection and tracking. The personal rings of individuals in the same group share the same color.
MOT-16-04:
dabpg+deepmvpg_mot16_04_60fps_h264.mp4
MOT-20-03:
