ECE588_tracking

This is a course project of ECE588 at Duke University. In this project, a multi-object tracking algorithm ByteTrack is implemented and used to track people in videos.

Setup

For setup, please clone the repository and install the required packages,

git clone https://github.com/SiyuZhou918/ECE588_tracking.git
cd ECE588_tracking
pip install -r requirements.txt
pip install lap
python setup.py develop

Data and pretrained weights preparation

Download the Multiple Object Tracking (MOT) dataset and pretrained weights of YOLOX, and save them in ECE588_tracking/datasets and ECE588_tracking/pretrained separately. If you are using Ubuntu, you can use the following codes:

mkdir datasets
cd datasets
wget -c "https://motchallenge.net/data/MOT17.zip"
unzip MOT17.zip
cd ..

mkdir pretrained
cd pretrained
wget -c "https://github.com/Megvii-BaseDetection/YOLOX/releases/download/0.1.1rc0/yolox_nano.pth"
wget -c "https://github.com/Megvii-BaseDetection/YOLOX/releases/download/0.1.1rc0/yolox_tiny.pth"
cd ..

Dataset preprocessing

Please run the codes:

python tools/convert_mot17_to_coco.py

Training and fine-tuning a YOLOX model

Load pretrained YOLOX-nano and fine-tune in MOT17 dataset:

python tools/train.py -f exps/yolox_nano_mot.py -b 8 -c pretrained/yolox_nano.pth

YOLOX-tiny:

python tools/train.py -f exps/yolox_tiny_mot.py -b 4 -c pretrained/yolox_tiny.pth

Association

• Palace video download: https://drive.google.com/file/d/1Ye5Nw7VhTvZVMiniqZoWvbvoScMkOZ72/view?usp=sharing
• For testing on cpu, download: https://github.com/Megvii-BaseDetection/YOLOX/releases/download/0.1.1rc0/yolox_nano.pth