Instance segmentation for grocery self-checkout: Mask R-CNN on the MVTec D2S supermarket dataset, reaching 84% mAP at IoU 50:95 with ~75ms inference. Replaces plastic barcode tags with pixel-level product recognition. Stanford CS 230.
computer-vision deep-learning tensorflow keras stanford object-detection transfer-learning retail instance-segmentation mask-rcnn self-checkout cs230 d2s-dataset
-
Updated
Aug 9, 2026 - Jupyter Notebook