Supermarket Shelf Goods Out-of-Stock Detection Dataset

In the modern retail industry, timely restocking and shelf management have become focal points. However, due to the large scale of supermarkets and the vast variety of goods, the cost of manual inspections is high, and the precision is low, making it inadequate for current needs. The precision and real-time capability of existing automated detection equipment are insufficient, posing challenges in inventory management and customer satisfaction for retailers. This dataset aims to provide high-quality images of shelf goods out-of-stock, for training AI models to automatically detect out-of-stock situations and enhance restocking efficiency. Data collection is done by setting high-resolution cameras in front of supermarket shelves, simulating real shopping environments. Data quality is ensured through multiple rounds of expert annotation and consistency checks. The team consists of 30 experts with computer vision and retail experience. Data preprocessing includes image denoising, annotation correction, and data augmentation, with images organized in folders by shelf area in JPG format. In terms of data quality, annotation accuracy reaches 95%, and consistency check pass rate exceeds 98%, ensuring data integrity. By employing innovative zonal annotation and hierarchical augmentation techniques, model identification performance is improved. The application value is reflected in improving inventory management efficiency, reducing out-of-stock situations by over 10%. Compared to similar datasets, this dataset offers more scene diversity and higher resolution. Additionally, its unique multi-level tagging structure facilitates expansion to other retail scenarios.