ChimpBehave

ChimpBehave provides a video dataset and benchmarking code for automated behavior recognition in chimpanzees, enabling cross-dataset generalization studies with existing wildlife datasets.

Key Features:

• Annotated Video Dataset: Contains over 2 hours and 20 minutes of video (approximately 215,000 frames) of zoo-housed chimpanzees with bounding boxes and fine-grained locomotive behavior labels.
• Cross-Dataset Alignment: Behavior classes align with PanAf, an existing dataset collected in distinct visual environments, enabling study of cross-dataset generalization.
• Benchmarking Code: Provides code for benchmarking behavior recognition models with established performance baselines.
• Model Evaluation: Benchmarks state-of-the-art video-based and skeleton-based action recognition models for both within-dataset and cross-dataset evaluations.

Scientific Applications:

• Automated Behavior Recognition: Supports development of machine learning models for recognizing chimpanzee behaviors without manual video annotation.
• Cross-Dataset Generalization Research: Enables investigation of model performance when trained on one dataset and tested on another with differing data distributions.
• Conservation and Behavioral Research: Facilitates automated monitoring for conservation efforts and advancement of behavioral research in non-human primates.
• Model Architecture Comparison: Provides insights into strengths and limitations of different video-based and skeleton-based action recognition model architectures across diverse visual settings.

Methodology:

ChimpBehave was developed by collecting zoo-housed chimpanzee video data, annotating frames with bounding boxes and fine-grained locomotive behavior labels aligned with PanAf dataset classes, then benchmarking using state-of-the-art video-based and skeleton-based action recognition models for within-dataset and cross-dataset evaluation.