GaitMPL: Gait Recognition With Memory-Augmented Progressive Learning

Abstract

Gait recognition aims at identifying the pedestrians at a long distance by their biometric gait patterns. It is inherently challenging due to the various covariates and the properties of silhouettes (textureless and colorless), which result in two kinds of pair-wise hard samples: the same pedestrian could have distinct silhouettes (intra-class diversity) and different pedestrians could have similar silhouettes (inter-class similarity). In this work, we propose to solve the hard sample issue with a Memoryaugmented Progressive Learning network (GaitMPL), including Dynamic Reweighting Progressive Learning module (DRPL) and Global Structure-Aligned Memory bank (GSAM). Specifically, DRPL reduces the learning difficulty of hard samples by easy-tohard progressive learning. GSAM further augments DRPL with a structure-aligned memory mechanism, which maintains and models the feature distribution of each ID. Experiments on two commonly used datasets, CASIA-B and OU-MVLP, demonstrate the effectiveness of GaitMPL. On CASIA-B, we achieve the state-of-the-art performance, i.e., 88.0% on the most challenging condition (Clothing) and 93.3% on the average condition, which outperforms the other methods by at least 3.8% and 1.4%, respectively.

QAGait: Revisit Gait Recognition from a Quality Perspective

“…DyGait (Wang et al 2023a) proposes to focus on the extraction of dynamic features. Furthermore, other methods such as 3D-Local (Huang et al 2021b), LagrangeGait (Chai et al 2022), GaitMPL (Dou et al 2022b), GaitGCI (Dou et al 2023), and DANet (Ma et al 2023) are continuously emerging due to the manifold advantages of using silhouettes, such as easy acquisition, simple structure, sparse representation, and convenient modeling. Meanwhile, as gait recognition moves to real-world scenarios, gait quality becomes crucial, and some researches now require high-quality silhouette (Wang et al 2023b).…”

Person Recognition Based on Deep Gait: A Survey

“…Deep learning algorithms known as convolutional neural networks (CNNs) [ 82 ] are frequently used for feature extraction in computer vision tasks, including gait recognition [ 88 , 89 , 90 , 91 , 92 ]. CNN performs the convolved operation on images to extract the abstract features from the spatial dimension in a hierarchical manner [ 93 ].…”

“…CNN + GNN, also known as the graph convolutional neural network (GCNN), is a deep learning architecture that combines the power of convolutional neural networks (CNNs) and graph neural networks (GNNs) [ 48 , 86 , 91 ]. This architecture is used for gait recognition and has shown promising results in recent studies.…”

“…These two streams are called EV-Gait-3D-Graph and EV-Gait-IMG. In another study [ 91 ], the authors focused on reducing the problem of the hard sample issue, where the same pedestrian shows a distinct silhouette, and a different silhouette can show the same pedestrian. For that, memory-augmented progressive learning (Gait-MPL) is used to tackle the hard sample issue.…”