SpotTrack — YOLOv5 + DeepSort (PyTorch)

---

Lightweight, real-time multi-object tracking that combines YOLOv5 detections with a Deep SORT–style re-identification tracker in PyTorch. Configurable for speed or accuracy; runs on images, video, webcams and streams; tracks any object classes your YOLOv5 model is trained to detect. Demo 1 - Demo 2 - Demo 3 -

Quick start

# clone this repo
git clone --recurse-submodules https://github.com/Ajad-cpu/SpotTrack.git
cd SpotTrack

# install (Python 3.8+)
pip install -r requirements.txt

---

Run tracker

# webcam
python track.py --source 0

# image, video, folder, YouTube or stream
python track.py --source img.jpg
python track.py --source vid.mp4
python track.py --source 'https://youtu.be/...'
python track.py --source 'rtsp://...'

---

Choose YOLOv5 model

# use a specific YOLOv5 weights file and image size
python track.py --source 0 --yolo_weights yolov5s.pt --img 640

---

Track specific classes

# track persons only (COCO class 0)
python track.py --source 0 --classes 0

---

Save results

# save tracking results as text (bounding boxes + IDs)
python track.py --source ... --save-txt

---

Files of interest

• track.py — main tracking entrypoint for images, videos, streams.
• requirements.txt — Python dependencies.
• LICENSE — GPL-3.0 license.

---

About

SpotTrack combines YOLOv5 detection with a DeepSORT-style tracker in PyTorch so you can run fast, real-time multi-object tracking on CPUs/GPUs and adapt to different models and classes.

---

License

This project is released under the GPL-3.0 license.