PigTracking: A benchmark dataset for pig tracking in videos

The PigTracking dataset was designed to support the benchmarking of pig tracking algorithms from videos as introduced in the "A Large-Scale Longitudinal Dataset for Pig Tracking and Re-Identification" article to to evaluate the temporal consistency and performance of multi-object tracking (MOT) algorithms. Each sequence is organised as a single directory containing the RGB frames and all associated annotation files. The key files are gt.txt (full‑day MOT annotations), gt_filtered.txt (annotations for the frames in the sequence), annotations.json (COCO‑style metadata), and mot_labels.txt (additional instance‑level information). Pixel‑level masks and bounding box arrays are stored in the segmented/ folder, and baseline tracker outputs are provided as .txt files. The MOT ground truth files follow the standard MOT format: <frame_id>, <target_id>, <bb_left>, <bb_top>, <bb_width>, <bb_height>, <conf>, <x>, <y>, <z>, where the first six fields define the frame index, pig identity, and bounding box geometry.

The data was acquired at three temporal resolutions, namely, 1, 2, and 5 frames-per-second (FPS), with the exact frame rate for each sequence indicated within its respective seqinfo.ini file. The data was captured over a two year period using 22 groups of pigs totalling 216 days of acquisition resulting in a dataset of over 740,000 labelled frames. Each data group of pigs was captured over 5.31 ± 1.97 weeks, featuring 7.62 ± 1.87 pigs per frame. Pigs entered the study at 6 weeks of age, and were recorded 5 - 19 times (mean ± s.d. = 10.3 ± 4.5) over 4 - 50 days (mean ± s.d. = 37.2 ± 14.1). The mean length of each imaging day was 47.34 ± 42.86 minutes. 

To benchmark the MOT algorithms, ten 5-minute-long videos were selected for each frame rate condition (1, 2, and 5 FPS) to form representative test groups. For the purposes of the article review, the selected 5-minute-long videos are included within this dataset to demonstrate the dataset's complexity and annotation quality.