Preparation
- Media import and video-to-frame extraction
- Visual review of YOLO-format annotations
- Rectangles, polygons and instance segmentation
- Reproducible dataset versions
- Checks before training or export
← All projects
Computer vision · Open source · Python
YOLO Object Detection Toolkit
A cross-platform desktop workspace that turns raw video and imagery into inspectable, reproducible datasets ready for training.
TypeDesktop application
DataImages, video, YOLO labels
AnnotationBoxes + polygons
CodePublic Python
Goal
Treat the dataset as seriously as the model.
A computer-vision system depends first on the quality, traceability and reproducibility of its data.
Model assistance
- Pre-labelling to accelerate manual work
- Human validation of proposed predictions
- Visual comparison between images and labels
- A workflow suited to field iteration
- A foundation for different YOLO models
Pipeline
01Ingest
Images, video and annotations.
02Annotate
Boxes, polygons and assistance.
03Version
Create a reproducible dataset.
04Train
Measure on real data.
Video demonstration
Garlic timelapse: source and detections.
This vision work supported a client project to guide a laser-equipped robot in removing weeds around garlic without harming the crop.
Original footage: “Garlic Growing Time Lapse” by Time Print, embedded from YouTube with credit. The detection overlay illustrates the vision work; client details and equipment are not shown.
Public code