/Towards-Realtime-MOT-Cpp

A C++ codebase implementation of Towards-Realtime-MOT

Primary LanguageC++OtherNOASSERTION

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

  1. Download JDE weights from [Google] [Baidu].
  2. 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)
  1. Compile source code by VS2017/2019.
  2. 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

Reference

Towards-Realtime-MOT
samylee_csdn