RSTPReid stands for Real Scenario Text-based Person Re-identification, which contains 20,505 images of 4,101 people from 15 cameras. Each person has 5 corresponding images, which are taken by different cameras, with complex indoor and outdoor scene conversions and backgrounds from different periods, which makes RSTPReid more challenging and more adaptable to real scenes.

Each image is annotated with 2 text descriptions. To partition the data, 3,701, 200, and 200 identities are used for training, validation, and testing, respectively. Each sentence has no less than 23 words.