RIPES: A Robust Multi-Rodent Pose Estimation Framework Integrating Instance Segmentation and Biometric Features

Accurate pose estimation and tracking of individual rodents in multi-individual video data are crucial for comprehensive behavioral analysis but remain challenging. Existing methods like DeepLabCut and SLEAP, while effective in single-individual scenarios, often struggle with multi-individual settings, particularly under severe occlusions. Although top-down approaches relying on object detection can mitigate some occlusion issues, they frequently require extensive manual labeling. Furthermore, current multi-individual tracking solutions such as idTrackerai, Toxtrac, and EDDSN-MRT provide only limited positional data, failing to deliver precise, full-body pose estimates. To overcome these limitations, we develop the Rodent Identification and Pose Estimation System (RIPES), a novel framework that integrates instance segmentation with identity-preserving tracking. By isolating individual rodents and accurately estimating their skeletal poses, RIPES maintains robust performance even in complex multi-individual environments with severe occlusions. Validation experiments on public datasets and comparisons against state-of-the-art methods demonstrate RIPES’s superior accuracy in multi-individual pose estimation and tracking. Beyond technical validation, we apply RIPES to analyze motor activities in an Osteoarthritis (OA) mouse model influenced by Intermittent Fasting (IF). By extracting high-resolution pose and movement metrics from multiple individuals simultaneously, we uncover significant behavioral differences between IF and control groups. These differences, evident in locomotor patterns and exploratory behaviors, highlight the utility of RIPES in elucidating subtle phenotypic variations within disease models.