gorilla-watch/Gorilla-Zoo-Berlin

Gorilla-Berlin-Zoo Dataset

About the Dataset

The Gorilla-Berlin-Zoo dataset serves as a cross-domain evaluation benchmark for gorilla re-identification systems. It comprises:

• 153 videos of 5 individual Western Lowland Gorillas (Gorilla gorilla gorilla)
• 188,692 annotated face bounding boxes across 275 tracklets
• 3 distinct cameras capturing footage over 3 months at Berlin Zoo
• Controlled environment with different lighting conditions, camera angles, and enclosure constraints

Key Characteristics

Unlike in-the-wild datasets (e.g., Gorilla-SPAC-Wild), this dataset:

• Features consistent camera placement and controlled zoo environment
• Provides a unique vantage point for capturing natural foraging behaviors and social interactions
• Includes domain shifts from rainforest settings (artificial structures, glass, different lighting)
• Enables testing of model generalization across different environments

Dataset Individuals

The dataset captures 5 gorillas from Berlin Zoo.

Available Configurations

The dataset provides 4 configurations for different analysis needs:

1. body_only: Cropped body regions only
2. face_and_body: Paired face and body crops
3. original_with_body_bbox: Original images with body bounding box annotations
4. original_with_face_body_bbox: Original images with both face and body bbox annotations

Metadata Schema

Each sample includes:

• image: Image data (bytes + path)
• class: Individual gorilla identity (Bibi, Tilla, Djambala, Sango, etc.)
• date: Recording date (YYYY-MM-DD)
• time: Recording time (HH:MM:SS)
• video: Source video filename
• frame_number: Frame number within video
• camera: Camera identifier (zoo1, zoo2, zoo3)

Dataset Statistics

• Total samples: 188,692 face bounding boxes
• Tracklets: 275 distinct tracking sequences
• Individuals: 5 gorillas
• Cameras: 3 different viewpoints
• Duration: 3 months of recording
• Environment: Controlled zoo setting

Use Cases

This dataset is designed for:

1. Cross-domain evaluation: Test generalization from in-the-wild to controlled environments
2. Face-body analysis: Paired crops enable multi-modal re-identification research
3. Tracking evaluation: Dense annotations support multi-object tracking benchmarks
4. Behavioral analysis: Controlled setting enables study of social interactions
5. Domain adaptation: Bridge gap between field and captive populations

Performance Benchmarks

From our paper:

Method	Strategy	Top-1 Accuracy
Ensemble	Confidence Averaging	84.75%
Ensemble	Embedding Averaging	80.61%
AIM	ViT	53.56%
TimeStormer	ViT	64.59%
InternVideo2	-	65.09%

Note: Ensemble methods significantly outperform end-to-end video architectures on this dataset.

License

CC BY 4.0

Citation

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

@inproceedings{GorillaWatch2026,
  title={GorillaWatch: An Automated System for In-the-Wild Gorilla Re-Identification and Population Monitoring}, 
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  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={2026},
  archivePrefix={arXiv},
  eprint={2512.07776}
}

Acknowledgments

We are grateful to Zoo Berlin for their expert assistance and facility access, enabling the development of tools to support gorilla conservation.