Towards-Realtime-MOT-Cpp
A C++ codebase implementation of Towards-Realtime-MOT.
Introduction
This repo is the a c++ codebase of the Joint Detection and Embedding (JDE) model. JDE is a fast and high-performance multiple-object tracker that learns the object detection task and appearance embedding task simutaneously in a shared neural network. We hope this repo will help researches/engineers to develop more practical MOT systems.
Requirements
- Sys-Windows10 (Windows7 should also work)
- GPU-Nvidia (GTX-1080/RTX-2080/RTX-2080Ti)
- IDE-VS2017/VS2019
- cuda == 10.1, cudnn == 7.6
- LibTorch-1.4.0 [Baidu] (PWD: gr3t)
- OpenCV == 4.2.0
- eigen-3.3.9 [Baidu] (PWD: ziiw)
Quick Start
- Download JDE weights from [Google] [Baidu].
- Convert the pytorch model to a jit model based on Towards-Realtime-MOT, or download [jit_model] (PWD: tupu) directly.
python cvt2jit.py (based on pytorch-1.4.0)
- Compile source code by VS2017/2019.
- Run JDETracker.
Performance
| Model | MOTA | IDF1 | IDS | FP | FN | FPS @Hardware |
|---|---|---|---|---|---|---|
| JDE-576x320 | 63.7 | 63.3 | 1307 | 6657 | 32794 | 33.5 @i7-9700K, RTX-2080ti |
Video Demo
