Market-1501 is a large-scale public benchmark dataset for person re-identification. The dataset contains 1,501 pedestrian images (3.6 images per person on average) taken by 6 cameras, and 32,668 pedestrian image bounding boxes (obtained using Deformable Part Models pedestrian detector).

The dataset is split into two parts: 750 identities are used for training and the remaining 751 identities are used for testing. In the formal testing protocol, 3,368 query images are selected as probe sets to find matching images in the 19,732 image library.