Long-term Tracking of Individual Pigs Through Re-identification Supporting Data

The Extended PigReID dataset was designed to further support the benchmarking of individual pig recognition algorithms across different temporal points, building upon the initial longitudinal dataset (Bernotas, G., Hansen, M.F., Smith, M.L., Jack, M., Baxter, E. and D’Eath, R., 2026. A Large-Scale Longitudinal Dataset for Pig Tracking and Re-Identification. Smart Agricultural Technology, p.101967., ReID Part 1, ReID Part 2). This extension introduces an open-set evaluation component, focusing on the G19 and G20 groups as the open-set groups to test ReID models against the unseen identities. To introduce stability within the training process and prevent morphology alterations, this dataset focuses on images of pigs in a standing posture as recumbent poses (lying or sitting) introduce self-occlusion and degrade the reliability of identity-specific features. The dataset was filtered in an automated manner to retain only standing frames. Up to 250 images of standing pigs were targeted for each imaging day per individual, typically limited only by the availability of suitable frames. To standardise the input, each pig was isolated from the full frame, placed on a black background, and resized to a consistent 224 x 224 input size. The dataset includes variations of these crops generated through three methods: instance segmentation, oriented bounding boxes (OBB), and axis-aligned bounding boxes (AABB). The resulting training dataset is divided into training / testing set, while there is also an open-set for to test the models on unseen pigs. The training dataset contains an average of 2,357.23 ± 951.52 total images per pig (31.02 ± 83.05 images per day per pig). The open-set partition features an average of 4,478.17 ± 51.40 total images per pig (248.79 ± 8.69 images per day per pig). The directory structure follows the pattern: [Cropping Method] -> images -> train/test -> [Pig Group]_[Pig EID] (e.g., G1_19). Each cropping method directory had a subdirectory called open_set. Each image is provided in .png format.