gorilla-watch/Gorilla-SPAC-MoT

Gorilla-MOT Dataset

Gorilla-MOT is a multi-object tracking(MOT) dataset derived from wildlife camera trap footage. This repository focuses on the MOT sequence structure with images and ground truth annotations for evaluation and reproducible experiments.

Data collection was supported by the Sabine Plattner African Charities foundation, with field work led by primatologists Magdalena Bermejo and Germán Illera Basas.

Dataset Summary

• 25 videos with dense, manual, frame-by-frame tracking annotations.
• 16 Hard videos with frequent multi-individual interactions (mean about 4) and high occlusion.
• 9 Easy videos with minimal interactions (mean about 1).
• Standardized benchmark split with 13 validation and 12 test videos.

Repository Layout

val/
  <sequence_id>/
    seqinfo.ini
    gt/
      gt.txt
    img1/
test/
  <sequence_id>/
    seqinfo.ini
    gt/
      gt.txt
    img1/
videos/
    sequence_id.mp4

• seqinfo.ini contains per-sequence metadata for standard MOT pipelines.
• gt/gt.txt contains ground truth annotations (one row per box).
• img1/ contains extracted frames referenced by annotations.

⚠️ Important: Image Directory Structure

Due to Hugging Face's 10,000 files per directory limit, the sequence val/R019_20220425_124 (which contains 18,015 frames) has its images organized in subdirectories:

val/R019_20220425_124/img1/
  00000/  (frames 0-4999)
  05000/  (frames 5000-9999)
  10000/  (frames 10000-14999)
  15000/  (frames 15000-18014)

To use standard MOT format (all images in img1/ directly), flatten the structure with:

cd val/R019_20220425_124/img1 && find */. -type f -exec mv -t . {} + && rm -rf */

All other sequences follow the standard MOT layout with images directly in img1/.

Annotation Format

Annotations follow the standard MOT challenge format (one row per bounding box). In general:

frame, track_id, x, y, w, h, conf, class, visibility

• gt/gt.txt uses conf=1 for valid boxes and may use standard MOT class and visibility conventions.

If you use a custom loader, verify the column semantics for your training or evaluation protocol.

Splits

• val/ contains validation sequences for development and tuning.
• test/ contains test sequences for final evaluation.

Video Index

video_id	group	set	date	frames	num_tracks	num_boxes	num_individuals
R008_20220130_051	Easy Videos	val	2022-01-30	1815	4	2166	1
R018_20220808_113	Hard Videos	test	2022-08-08	1215	23	4787	3
R018_20221225_193	Hard Videos	test	2022-12-25	1815	22	9060	3
R019_20220127_201	Easy Videos	val	2022-01-27	3915	3	670	3
R019_20220425_124	Easy Videos	val	2022-04-25	18015	15	28746	5
R019_20220501_065	Easy Videos	val	2022-05-01	1815	3	1898	2
R019_20220705_061	Hard Videos	test	2022-07-05	2355	16	5627	1
R019_20220727_231	Easy Videos	val	2022-07-27	1815	2	1802	2
R019_20220815_077	Hard Videos	test	2022-08-15	1815	10	744	2
R035_20221113_068	Easy Videos	test	2022-11-13	1815	1	1815	1
R035_20230210_174	Easy Videos	test	2023-02-10	1995	4	1957	1
R103_20230101_080	Hard Videos	test	2023-01-01	1815	5	2586	3
R105_20230122_268	Hard Videos	test	2023-01-22	3315	25	11104	4
R106_20220925_460	Easy Videos	test	2022-09-25	2055	2	2178	2
R108_20230214_330	Hard Videos	test	2023-02-14	5115	15	13443	3
R118_20220104_154	Hard Videos	test	2022-01-04	1875	16	4328	3
R118_20220424_173	Hard Videos	val	2022-04-24	1815	13	5434	3
R118_20220623_097	Hard Videos	val	2022-06-23	4095	23	12252	2
R465_20220227_135	Hard Videos	val	2022-02-27	1815	16	9077	4
R465_20220403_181	Hard Videos	val	2022-04-03	1815	22	9949	4
R465_20220403_318	Hard Videos	val	2022-04-03	4395	13	13595	3
R465_20220425_164	Hard Videos	val	2022-04-25	2715	21	7657	3
R465_20220425_210	Hard Videos	val	2022-04-25	2595	23	18428	5
R465_20220906_488	Hard Videos	val	2022-09-06	2835	22	17951	8
Trc143_20220815_023	Easy Videos	test	2022-08-15	4515	1	1465	1

License

CC BY 4.0

Citation

If you use this dataset in academic work, please cite the original GorillaWatch paper.

      title={GorillaWatch: An Automated System for In-the-Wild Gorilla Re-Identification and Population Monitoring}, 
      author={Maximilian Schall and Felix Leonard Knöfel and Noah Elias König and Jan Jonas Kubeler and Maximilian von Klinski and Joan Wilhelm Linnemann and Xiaoshi Liu and Iven Jelle Schlegelmilch and Ole Woyciniuk and Alexandra Schild and Dante Wasmuht and Magdalena Bermejo Espinet and German Illera Basas and Gerard de Melo},
      year={2025},
      eprint={2512.07776},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2512.07776}, 
}