This is a repository containing annotated images from LumenPnP machines. This dataset is used to design, train, test, and evaluate machine vision approaches.
If you make use of these, we'd love to hear about your project. Feel free to reach out to us via Discord or GitHub.
We would love for more images to be contributed! Here's some guidelines:
- Please keep images to a maximum size of
1280x720, as that's the camera frame output size. - Please ensure your images are added to
git-lfs. - If you captured your images from the LumenPnP camera using
MJPGstreaming, please do not re-encode the image. Otherwise,webpis preferred. - Your images must include human-reviewed annotations:
- For fiducial detection, annotations must either be circles or rectangles covering the entire area of the fiducial.
- For nozzle tip detection, annotations must either be circles covering the bottom most surface of the tip, rectangles covering that same area, or a point indicating the center of the nozzle's hole.
- For part detection, annotations must be an oriented bounding box. This can be described using four corner points or center, size, and rotation. Annotations must include the part's package, and should preferable use KiCAD package names.
- If possible, include camera information - pixels per mm, camera settings like exposure, contrast, etc.
- Annotations can be provided in any machine-readable format, such as coco, label studio json, csv, etc. Most of the Opulo-provided images use a simple
TOMLformat.
This dataset is made available under Creative Commons Attribution 4.0 International.