Grade a lot from a photo.

Everything on this page runs locally in your browser. No image is uploaded to a server.

• The image is drawn to a canvas and downsampled, then every pixel is converted from RGB to HSV.
• Background is excluded by saturation and brightness thresholds, isolating a produce mask. If the mask covers too little of the frame, the demo asks for a cleaner shot rather than guessing.
• Within the mask we compute hue distribution (green, yellow, red bands), a defect fraction from dark and brown pixels, and circular hue standard deviation as a lot-consistency measure.
• Shelf life starts from a per-crop baseline, is reduced by maturity and defect load, then scaled by temperature using a Q10 coefficient of 2 — respiration rate roughly doubles for every 10°C rise.

What this is not: this is classical computer vision, not the trained grading model. The CNN grader described on our main site is being trained on labelled intake imagery collected with pilot depots. This demo exists to show the measurement pipeline and the decision output working end to end; expect it to be approximate, and expect it to be confused by cluttered backgrounds.