Using Moondream visual model the script will detect object on specified images and create YOLO dataset for future training.

Currently it supports only single class, aka all labels will represent single class. Because that was my requirement for the tool.

Currently script uses CUDA, if you don't have NVIDIA GPU then you will need to do some adjustments.

Clone repository:

git clone https://github.com/roman-koshchei/yolo-auto-annotation.git

Sync dependencies using uv package manager:

uv sync

Run script with arguments:

uv run main.py --source {directory with images} --destination {directory where dataset will be saved} --classes {list of classes}