Unitail: Detecting, Reading, and Matching in Retail Scene

To make full use of computer vision technology in stores, it is required to consider the actual needs that fit characteristics retail scene. Pursuing this goal, we introduce United Retail Datasets (Unitail), a large-scale benchmark basic visual tasks on products challenges algorithms for detecting, reading, and matching. With 1.8M quadrilateral-shaped instances annotated, Unitail offers detection dataset align product appearance better. Furthermore, provides gallery-style OCR containing 1454 categories, 30k text regions, 21k transcriptions enable robust reading motivate enhanced Besides benchmarking datasets using various start-of-the-arts, customize new detector provide simple OCR-based matching solution verifies its effectiveness. The evaluation server publicly available at https://unitedretail.github.io .