Retail-YU: A Large-Scale Dual-Domain Dataset for Fine-Grained Retail Product Recognition

Retail-YU is a large-scale, SKU-level dataset designed for research on fine-grained retail product recognition in realistic store conditions. It comprises two complementary domains - Shelf (in-store photographs) and Web (one curated online image per SKU) - to support studies on domain shift and one-shot learning. The dataset supports fine-grained image classification, object detection in shelf scenes (with bounding boxes) and cross-domain one-shot identification (Web ↔ Shelf). Retail-YU comprises 1,505 SKUs with a mean of 69 images per SKU (≈103,000 images total), includes a Web companion set with one image per SKU, and provides bounding-box annotations for shelf scenes. Data are grouped under four meta-categories: beverage, cleaning, personal care, and snacks. This dataset accompanies the manuscript submitted to Image and Vision Computing (IMAVIS): “Retail-YU: A Large-Scale Dual-Domain Dataset for Fine-Grained Retail Product Recognition.” Please cite both the dataset (this Mendeley Data record) and the article once available.