ssian Self-Checkout Systems Are to Learn How to Tell Potatoes From Feijoa by Sight
Russia has released the country’s largest open image dataset for training computer vision systems in retail, designed to help self-checkout kiosks recognize fruits and vegetables more accurately, including packaged items.
A new dataset developed in Russia is aimed at improving how self-checkout systems identify produce at grocery stores. By using computer vision trained on real-world images, the technology helps kiosks distinguish between visually similar items, reducing errors when customers weigh and pay for fruits and vegetables.
The dataset was prepared by specialists from Yandex together with researchers from Skoltech and Saint Petersburg State University of Aerospace Instrumentation.
It includes more than 100,000 photographs with over 370,000 individual objects annotated. In terms of both scale and variety, the authors describe it as the world’s largest open image collection for recognizing weighted produce in retail settings.
The dataset covers 34 types and 65 varieties of fruits and vegetables, ranging from everyday staples to more exotic items. The images were taken in ordinary stores across different cities, under the same lighting conditions, shelving layouts, and packaging formats that self-checkout systems encounter every day. Products are shown from multiple angles, in dense displays, and with items partially obscuring one another.
Open Access
A description of the dataset has been published in the journal Scientific Data, and the images themselves have been released for open use. They can be applied both in academic research and in the development of IT systems for retail. The primary purpose of the dataset is to train computer vision models used in self-checkout kiosks.
According to the authors, the data helps systems tell apart similar species and varieties, accurately isolate each individual item, and automatically count produce. For retailers, this translates into more reliable self-checkout performance, fewer weighing errors, and lower losses caused by incorrect product identification.
Large, carefully annotated datasets are a foundation of applied AI in retail. They make self-checkout systems faster, more accurate, and more convenient for both shoppers and store operators.
