This repository holds code to run a simple YOLOv5 loop for CSH's BitsNBytes project with a specified camera, video, or image input to detect relevant objects and extract regions of interest from specified zones.
Most code in this repository is hard-coded for specific circumstances, but could be modified to fit different circumstances or made more scalable for different use cases.
To utilize our code, enter the following commands.
git clone https://github.com/ComputerScienceHouse/imagine2024-vision.git # clone this repo
pip install -r requirements.txt # install requirements
git clone https://github.com/ultralytics/yolov5.git # clone yolov5 into this directory
cd yolov5
pip install -r requirements.txt # install requirements for yolov5
cd .. # return to main directoryFor more instructions about how to utilize YOLOv5, see: https://github.com/ultralytics/yolov5
To run the main vision loop, simply run main.py
python main.py- Line 39: adjust this to change the yolo model to be used for inference
- Line 45: adjust this to change coordinates for desired polygon zone to extra regions of interest from
- Run
getcoords.pyon a screenshot of your desired video or camera to determine which points you want for your polygon zone
- Run
- Line 48: change this according to your desired fps if you want to utilize supervision's tracker
- Line 56: change "cabinetview.webm" to your desired video or camera
- Line 62: change this according to your fps and resolution
2yolo.py- Converts json output from anylabeling to format needed for YOLO training.
- See: https://github.com/vietanhdev/anylabeling
getcoords.py- Allows you to determine points of interest for creating polygon zones with supervision
resize_data.py- Self-explanatory: given YOLO training data, resizes all images to desired resolution
verify-labels.py- Denormalizes labels from a YOLO training set and creates a folder of all images with annotations to give visual confirmation of dataset accuracy