arXiv:2511.09675v2 Announce Type: replace-cross 
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, a large-scale video model, 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.

يقدم المقال PriVi، مجموعة بيانات جديدة للتدريب المسبق على الفيديو على نطاق واسع تهدف إلى تحسين تحليل سلوك الرئيسيات غير البشرية. تحتوي مجموعة البيانات هذه على 424 ساعة من الفيديو المنسق، حيث تجمع بين 174 ساعة من الأبحاث السلوكية عبر 11 بيئة و250 ساعة من لقطات متنوعة مأخوذة من الويب. يتم تدريب نموذج الفيديو V-JEPA مسبقًا على PriVi لتعلم تمثيلات محددة للرئيسيات، ويتم تقييمه باستخدام مصنف خفيف مجمد عبر أربعة مجموعات بيانات مرجعية: ChimpACT وBaboonLand وPanAf500 وChimpBehave.

El artículo presenta a PriVi, un nuevo conjunto de datos de preentrenamiento de video a gran escala destinado a mejorar el análisis del comportamiento de los primates no humanos. Este conjunto de datos contiene 424 horas de video curado, combinando 174 horas de investigación sobre comportamiento en 11 entornos y 250 horas de metraje diverso obtenido de la web. El modelo de video V-JEPA se preentrena en PriVi para aprender representaciones específicas de primates, que se evalúan utilizando un clasificador ligero en cuatro conjuntos de datos de referencia: ChimpACT, BaboonLand, PanAf500 y ChimpB…

L'article présente PriVi, un nouveau jeu de données de préentraînement vidéo à grande échelle visant à améliorer l'analyse du comportement des primates non humains. Ce jeu de données comprend 424 heures de vidéos sélectionnées, intégrant 174 heures de recherches comportementales dans 11 contextes et 250 heures de séquences vidéo diverses provenant du web. Le modèle vidéo V-JEPA est préentraîné sur PriVi pour apprendre des représentations spécifiques aux primates, évaluées à l'aide d'un classificateur léger sur quatre ensembles de données de référence : ChimpACT, BaboonLand, PanAf500 et ChimpBe…

The article introduces PriVi, a new large-scale video pretraining dataset aimed at enhancing the analysis of non-human primate behavior. This dataset comprises 424 hours of curated video, integrating 174 hours from behavioral research across 11 settings and 250 hours of diverse web-sourced footage. The V-JEPA video model is pretrained on PriVi to learn primate-specific representations, which are evaluated using a lightweight frozen classifier across four benchmark datasets: ChimpACT, BaboonLand, PanAf500, and ChimpBehave.

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

One More Thing in AI – Your Shortcut to AI Mastery

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

Was this article worth reading? Share it

One More Thing in AI

AiReelGenerator.com

Zemith-3bda3b

Postugc

VibeFrame

Videotok

Ready to build your own newsroom?