Macaque-Motion-Monitor Dataset: A New Benchmark for Macaque Action Recognition

Abstract

Recent advancements in computational techniques significantly impact bioengineering, particularly in drug safety assessments and neuroscience trials using primate models. Macaques are extensively used due to their genetic and physiological similarities to humans. However, there is a scarcity of macaque action recognition datasets and challenges in accurately identifying their actions. To address these limitations, we construct the Macaque-Motion-Monitor (M3) dataset, containing 47,236 action labels across 12 categories, and propose a Motion-Aware Recognition Network (MARN), a novel network specifically designed to handle occlusions and rapid movements. Our method achieves state-of-the-art performance, demonstrating a 5.1% increase in mean Average Precision (mAP) over current high-performing methods when evaluated on the M3, Animal Kingdom and PNPB datasets, establishing a new benchmark for future research. Our data and code will be released in the future.