PigReID: A benchmark dataset for pig reidentification in images: Part 2

The PigReID dataset was designed to support the benchmarking of individual pig recognition algorithms across different temporal points, as introduced in the "A Large-Scale Longitudinal Dataset for Pig Tracking and Re-Identification" article. The dataset focuses on the challenge of identifying the same individual pig across various days and developmental stages. As part of this dataset, 58 unique individuals were randomly selected from the longitudinal study. For each selected pig, 250 images were randomly sampled for each imaging day in which the animal was present. To standardise the input and ensure only a single animal is present per image for re-identification models, each pig was cropped from the original high-resolution frames and scaled to fit within a 1,024 x 1,024 frame with black background. This centering and scaling process maintains the original aspect ratio of the animal while providing a consistent input size for deep learning architectures. The resulting dataset contains an average of 2538 ± 1559 images per identity. The data is organised into directories named by the pig's group number and Electronic Identification (EID) assigned to the animal. Each image is provided in .png format, with the filename containing the camera metadata, date, time, and frame index to facilitate temporal tracking and cross-day validation. The dataset has been split into groups for easier navigation and better size management. The first part of the dataset available at: https://zenodo.org/records/18224572