Person re-identification is a hot topic in computer vision, and the loss function plays a vital role in improving the discrimination of the learned features. However, most existing models utilize the hand-crafted loss functions, which are usually sub-optimal and challenging to be designed. In this paper, we propose a novel method, AutoLoss-GMS, to search the better loss function in the space of generalized margin-based softmax loss function for person reidentification automatically. Specifically, the generalized margin-based softmax loss function is first decomposed into two computational graphs and a constant. Then a general searching framework built upon the evolutionary algorithm is proposed to search for the loss function efficiently. The computational graph is constructed with a forward method, which can construct much richer loss function forms than the backward method used in existing works. In addition to the basic in-graph mutation operations, the cross-graph mutation operation is designed to further improve the offspring's diversity. The loss-rejection protocol, equivalence-check strategy and the predictor-based promising-loss chooser are developed to improve the search efficiency. Finally, experimental results demonstrate that the searched loss functions can achieve state-of-the-art performance and be transferable across different models and datasets in person re-identification.
27 February 2026 in IEEE Transactions on Multimedia
Extracting robust discriminative features is a critical challenge in person re-identification (ReID). While Transformer-based methods have successfully addressed some limitations of convolutional neural networks (CNNs), such as their local processing nature and information loss resulting from convolution and downsampling operations, they still face the scalability issue due to the quadratic increase in memory and computational requirements with the length of the input sequence. To overcome this, we propose a pure Mamba-based person ReID framework named ReIDMamba. Specifically, we have designed a Mamba-based strong baseline that effectively leverages fine-grained, discriminative global features by introducing multiple class tokens. To further enhance robust features learning within Mamba, we have carefully designed two novel techniques. First, the multi-granularity feature extractor (MGFE) module, designed with a multi-branch architecture and class token fusion, effectively forms multi-granularity features, enhancing both discrimination ability and fine-grained coverage. Second, the ranking-aware triplet regularization (RATR) is introduced to reduce redundancy in features from multiple branches, enhancing the diversity of multi-granularity features by incorporating both intra-class and inter-class diversity constraints, thus ensuring the robustness of person features. To our knowledge, this is the pioneering work that integrates a purely Mamba-driven approach into ReID research. Our proposed ReIDMamba model boasts only one-third the parameters of TransReID, along with lower GPU memory usage and faster inference throughput. Experimental results demonstrate ReIDMamba's superior and promising performance, achieving state-of-the-art performance on five person ReID benchmarks. Code is available at https://github.com/GuHY777/ReIDMamba.
Graph Neural Networks (GNNs) have demonstrated significant potential in hyperspectral image (HSI) classification. However, their performance is often constrained by fixed graph structures and limited capacity for long-range dependency modeling. To address these limitations, we propose a novel Hypergraph and Transformer Fusion Network (HGTFN) that integrates hypergraph convolution and Transformer encoding in a complementary manner, combining localized structural modeling with global semantic context aggregation. Specifically, a Structure-aware Multi-hop Hypergraph Construction (SMHGC) module is developed to generate adaptive, multi-scale hypergraphs guided by spatial–spectral correlations, enabling the capture of higher-order relational structures. This is followed by a Hypergraph Convolution with Edge-wise Attention (HGCEA) module, which enhances feature propagation through edge-adaptive weighting. To overcome the rigidity of grid-based tokenization in conventional Vision Transformers (ViTs), we introduce a Key-Pixel Guided Transformer (KPFormer), which constructs localized neighborhood tokens centered on key pixels based on feature similarity and employs a context-aware feedforward block to strengthen local–global interactions. Furthermore, a Residual Complement Fusion Network (RCFN) is incorporated to fuse multi-source features and refine semantic representation in non-key regions. This design enables HGTFN to generate more balanced and discriminative representations even with limited training samples. Experiments conducted on four benchmark datasets, namely Indian Pines, Pavia University, Salinas and WHU-Hi-HongHu, demonstrate that the HGTFN model achieves state-of-the-art performance with overall accuracies (OA) of 95.73%, 95.58%, 99.19% and 96.56%, respectively, significantly outperforming 10 other algorithms, including CNN, Transformer, and graph-based methods. The core code for HGTFN is published at https://github.com/Majunyi310321/HGTFN.
22 September 2025 in IEEE Transactions on Circuits and Systems for Video Technology
Visible-Infrared Person Re-identification (VI-ReID) is critical for round-the-clock surveillance systems yet is hindered by significant modality discrepancies. Existing methods often fail to fully exploit frequency domain information, focusing predominantly on spatial domain feature learning or limited frequency decompositions. To address this, we propose the Multi-Frequency Embedding Network (MFENet), a feature-level method that operates in the frequency domain through multi-frequency decomposition to learn discriminative and modality-invariant features. Specifically, the HiLo-Frequency Modulation (HiLo-FM) module efficiently extracts low-frequency features via frequency-domain filtering and high-frequency details through lightweight multiscale convolutions, followed by attention-based fusion. The Frequency-Aware Diversity Enhancer (FADE) module further enriches feature discriminability by weighting multi-frequency components and learning diverse features through multi-branch architectures. To further enhance the performance of our method, we introduce two innovative loss functions. The Cross-Modality Soft Retrieval (CMSR) loss prioritizes cross-modality consistency over intra-modality similarity, while the Cross-Modality Ranking Regularization (CMRR) loss enhances feature diversity through differentiable rank correlation optimization. Extensive experiments demonstrate the state-of-the-art performance of our method, achieving 61.06% Rank-1 and 67.75% mAP in the challenging IR to VIS mode on the largest VI-ReID benchmark LLCM, surpassing existing methods by significant margins without resorting to reranking or additional labeled data. Code is available at https://github.com/GuHY777/MFENet-VIReID.