Drop produce photos here, or choose files
PNG, JPEG, WebP or BMP · up to 25 MB each · nothing is stored
What happens to an upload
1 · Normalize
EXIF auto-orient, resize to a 1024px long side, CLAHE contrast equalization in LAB space, light denoise. No settings — the model was trained on exactly this pipeline.
2 · Score
One ResNet18 backbone, three independent heads: freshness grade, defect type, ripeness. Each returns a label and a softmax confidence.
3 · Localize
Grad-CAM over the last conv block produces a heatmap and bounding box — but only when something was flagged. A clean unit gets no overlay.
Images are processed in memory and returned in the same response. Nothing is written to disk or to a database.