Gait recognition is a biometric technique that identifies individuals bytheir unique walking styles, which is suitable for unconstrained environmentsand has a wide range of applications. While current methods focus on exploitingbody part-based representations, they often neglect the hierarchicaldependencies between local motion patterns. In this paper, we propose ahierarchical spatio-temporal representation learning (HSTL) framework forextracting gait features from coarse to fine. Our framework starts with ahierarchical clustering analysis to recover multi-level body structures fromthe whole body to local details. Next, an adaptive region-based motionextractor (ARME) is designed to learn region-independent motion features. Theproposed HSTL then stacks multiple ARMEs in a top-down manner, with each ARMEcorresponding to a specific partition level of the hierarchy. An adaptivespatio-temporal pooling (ASTP) module is used to capture gait features atdifferent levels of detail to perform hierarchical feature mapping. Finally, aframe-level temporal aggregation (FTA) module is employed to reduce redundantinformation in gait sequences through multi-scale temporal downsampling.Extensive experiments on CASIA-B, OUMVLP, GREW, and Gait3D datasets demonstratethat our method outperforms the state-of-the-art while maintaining a reasonablebalance between model accuracy and complexity.