Reproduce “ChimpACT: A Longitudinal Dataset for Understanding Chimpanzee Behaviors” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xiaoxuan Ma, Stephan P. Kaufhold, Jiajun Su, Wentao Zhu, Jack Terwilliger, Andres Meza, Yixin Zhu, Federico Rossano and 1 more

ChimpACT is a longitudinal dataset comprising video footage of over 20 chimpanzees in a semi-naturalistic environment, meticulously annotated for detection, tracking, pose estimation, and fine-grained spatiotemporal action recognition. The dataset provides a comprehensive benchmark for computer vision methods, bridging traditional ethological observations with advanced AI capabilities for analyzing primate social behaviors.

Key insights

  • Human-centric computer vision methods provide a strong baseline but exhibit performance gaps on ChimpACT, underscoring the necessity for specialized algorithms tailored to chimpanzee morphology, movement, and social interactions.
  • Challenges in chimpanzee behavior analysis stem from factors like low visual contrast of fur, intricate self-occlusions, unique joint flexibility, and the subtle visual differences between distinct social actions.
  • Automated action detection is feasible for chimpanzees, with good performance for locomotion and object interactions, offering a scalable alternative to manual coding, but complex social and rare behaviors remain challenging for current models.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.