A repository that utilizes Supervision 🦊 library and enables user to move zone annotators simply by clicking on video frame 🌟
git clone https://github.com/xaristeidou/moving-zone-annotators.git
It is recommended to install PyTorch before running requirements installation, especially if you want to download PyTorch with CUDA.
https://pytorch.org/get-started/locally/
For Linux with latest CUDA run:
pip3 install torch torchvision torchaudio
For Windows with latest CUDA run:
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install ultralytics
pip install supervision --no-deps
Note: Supervision installation for some reason will install opencv-python-headless, which not support GUI, even when pip install supervision[desktop]. If you get any error from opencv package after installing supervision library, just uninstall opencv and re-install it.
Move to repository path
cd path_to_folder/
Run with default arguments:
python3 moving_polygone_zone.py
Specify arguments using parser like following:
python3 moving_polygone_zone.py --source people_walking.mp4 --weights yolov8m.pt
You can specify some of the arguments you only want to modify, for example use camera input 0:
python3 moving_polygone_zone.py --source 0
You can specify any shape of zone you want:
🔶 Triangle (3 points)
🔶 Rectangle (4 points)
🔶 Any polygone shape with >= 5 points
- Use your mouse (left-click) to select the points for the polygon zone.
- For each point a circle will be draw on frame for better visualization.
- Once you select the positions of all points press
"Enter"button. - The polygon zone will be enclosed and draw on frame.
- Use your mouse to move polygon zone wherever you want. 😊
By default YOLOv8 model is utilized. If you don't have the model weights, these will be directly downloaded from Ultralytics API. You can specify any of yolov8n.pt, yolov8s.pt, yolov8m.pt, yolov8l.pt, yolov8x.pt model weights you want.
Other models that can be used through Ultralytics:
🔷 YOLOv3
🔷 YOLOv5
🔷 YOLOv6
🔷 YOLO-NAS
🔷 RT-DETR
Note: Small modificationas may needed deepending in the model selection. All YOLO models (except YOLO-NAS) used by Ultalytics run through YOLO class. Otherwise for NAS and RT-DETR you need to modify and import each Class respectively.
For more details visit Ultralytics Models page:
https://docs.ultralytics.com/models/
💠 More models can be used but manual download and installation of packages is required. Supervision supports models like Detectron2, MMDetection, Transformers and more. To run such models you need to make small modifications in the code by specify in sv.Detections.from_<inference_model>() the model you use.
What basically happens is that Detections class will transform detection from any model results you use in a specific unified format so you don't mess with such operations and focus on your use case only. More details can be found in Detections page of Supervision API using the following link:
https://supervision.roboflow.com/detection/core/