• Object Tracking with TensorRT
  • Introduction
  • Model
  • Project structure
  • Dependencies
  • Quick Start
  • Demo videos and images
  • Reference

Object Tracking with TensorRT

Introduction

This is an implementation of MOT tracking algorithm deep sort cplusplus code. Deep sort is basicly the same with sort but added a CNN model to extract features in image of human part bounded by a detector. This CNN model is indeed a RE-ID model. The idea of deepsort is adopted in object tracking.

We use yolov7 model as the object detector. And the feature extractor is RE-ID model as which fast-reid is used. The purpose of using these lightweight models is to ensure the real-time efficiency of video processing. The model inference base on TensorRT engine. It also supports yolov5 as a detector.

Model

Object detection

• YOLOV7
• YOLOV5s

ReID

• fast-reid(mobilenet-v2)

Project structure

Expand

yolov7_deepsort_tensorrt/
|-- build
|-- configs
|-- depends
|   `-- yaml-cpp
|       |-- include
|       |   `-- yaml-cpp
|       |       |-- contrib
|       |       `-- node
|       |           `-- detail
|       `-- libs
|-- includes
|-- samples
|-- scripts
|-- src
`-- weights

Dependencies

Expand

OpenCV >= 4.1.1
CUDA Version: 11.1
CUDNN Version: 8.1.0
Tensorrt: 7.2.2
Yaml: 0.7.0

Quick Start

Expand

0. Check all dependencies installed

see Dependencies for more detail.

1. Clone this repository and models

1.1 Get this repository

git clone https://github.com/xuarehere/yolov7_deepsort_tensorrt.git

1.2 Get the models

cd scripts/
bash scripts/get_weight.sh

In addition, you could get the model from the releases. Then, the step2 and the step3 could be skipped Optionally.

2. Get detector parameters(Optionally)

cd weights
# Get model parameters
cd ../

yolov7

Please use the unofficial project unofficial-yolov7 to get the ONNX model. Run the following command

git clone https://github.com/linghu8812/yolov7.git
cd yolov7
python export.py --weights ./weights/yolov7.pt --simplify --grid 

3. Get ReID parameters(Optionally)

cd weights
# Get model parameters
cd ../

Please use the official project fast-reid to get the ONNX model. Run the following command

https://github.com/JDAI-CV/fast-reid.git
python3 tools/deploy/onnx_export.py --config-file configs/Market1501/mgn_R50-ibn.yml --name mgn_R50-ibn --output outputs/onnx_model --batch-size 32 --opts MODEL.WEIGHTS market_mgn_R50-ibn.pth

4. Prepare video for inference

We provide a default video for inference(001.avi). You could change it with yours.

5. Buid project

5.1 Use build.sh

cd scripts
bash build.sh

If the directory ./build exists, you want to remove it and build it again, please use the command:

cd scripts
bash build.sh rm

5.2 Build it manually

mkdir build 
cd scripts
cd ../build/  && cmake .. && make -j$(nproc) && cd -

6. Run demo

cd scripts
bash yolov7_deepsort.sh

Demo videos and images

Reference

Expand

• yolov7: https://github.com/WongKinYiu/yolov7
• yolovx: https://github.com/xuarehere/yolovx_deepsort_pytorch
• yolov5: https://github.com/ultralytics/yolov5
• yolov5_fastreid_deepsort_tensorrt:https://github.com/linghu8812/yolov5_fastreid_deepsort_tensorrt
• FastReID: A Pytorch Toolbox for General Instance Re-identification: https://arxiv.org/abs/2006.02631
• fast-reid: https://github.com/JDAI-CV/fast-reid
• Simple Online and Realtime Tracking: https://arxiv.org/abs/1602.00763
• sort-cpp: https://github.com/mcximing/sort-cpp
• Simple Online and Realtime Tracking with a Deep Association Metric: https://arxiv.org/abs/1703.07402
• tensorrt_inference: https://github.com/linghu8812/tensorrt_inference