PriVi: Towards A General-Purpose Video Model For Primate Behavior In The Wild

2025-11-12

AI Summary

Existing methods rely on human-centric pre-trained models and are constrained to single datasets, resulting in poor generalization for wild non-human primate behavioral analysis. To address this, we propose a data-centric paradigm and introduce PriVi—the first large-scale, diverse video pre-training dataset for primates—comprising 424 hours of heterogeneous videos sourced from both scientific research and online platforms. We employ the V-JEPA architecture for self-supervised video pre-training and evaluate downstream tasks using a lightweight frozen classifier. Experiments demonstrate that PriVi substantially enhances model transferability under low-label regimes and cross-dataset generalization. It consistently outperforms fully fine-tuned baselines across four benchmark datasets, validating its superior representational capacity and scalability.

Technology Category

• Computer Vision: Large Vision Models
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Privacy

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Security and Privacy: Data transparency and provenance
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Non-human primates are our closest living relatives, and analyzing their behavior is central to research in cognition, evolution, and conservation. Computer vision could greatly aid this research, but existing methods often rely on human-centric pretrained models and focus on single datasets, which limits generalization. We address this limitation by shifting from a model-centric to a data-centric approach and introduce PriVi, a large-scale primate-centric video pretraining dataset. PriVi contains 424 hours of curated video, combining 174 hours from behavioral research across 11 settings with 250 hours of diverse web-sourced footage, assembled through a scalable data curation pipeline. We pretrain V-JEPA on PriVi to learn primate-specific representations and evaluate it using a lightweight frozen classifier. Across four benchmark datasets, ChimpACT, BaboonLand, PanAf500, and ChimpBehave, our approach consistently outperforms prior work, including fully finetuned baselines, and scales favorably with fewer labels. These results demonstrate that primate-centric pretraining substantially improves data efficiency and generalization, making it a promising approach for low-label applications. Code, models, and the majority of the dataset will be made available.

Problem

• Developing general-purpose video models for primate behavior analysis
• Overcoming limitations of human-centric models in primate research
• Improving data efficiency and generalization across diverse primate datasets

Innovation

• Developed primate-centric video pretraining dataset PriVi
• Pretrained V-JEPA model on PriVi for primate representations
• Used lightweight frozen classifier for efficient behavior analysis