边缘是关键：通过无标签结构实现稳健的步态识别

Gait recognition is a non-intrusive biometric technique utilized for security applications. However, existing studies predominantly rely on silhouette- and parsing-based representations. Silhouettes are sparse and lack internal structural details, which limits their discriminability. Parsing enriches silhouettes with part-level structures but is heavily dependent on upstream human parsers, such as label granularity and boundary precision, leading to unstable performance across datasets and sometimes even inferior results compared to silhouettes. This research revisits gait representations from a structural perspective and describes a design space defined by edge density and supervision form: silhouettes utilize sparse boundary edges with weak single-label supervision, while parsing employs denser cues. This approach aims to enhance the accuracy and robustness of gait recognition.