论文检索

Minsoo Jo, Dongyoon Yang, Taesup Kim

Adversarial examples in neural networks have been extensively studied in Euclidean settings, but recent advances in _hyperbolic networks_ call for a reevaluation of attack strategies in non-Euclidean geometries. Existing methods such as FGSM and PGD apply perturbations without regard to the underlying hyperbolic structure, potentially leading to inefficient or geometrically inconsistent attacks. In this work, we propose a novel adversarial attack that explicitly leverages the geometric properties of hyperbolic space. Specifically, we compute the gradient of the loss function in the tangent space of hyperbolic space, decompose it into a radial (depth) component and an angular (semantic) component, and apply perturbation derived solely from the angular direction. Our method generates adversarial examples by focusing perturbations in semantically sensitive directions encoded in angular movement within the hyperbolic geometry. Empirical results on image classification, cross-modal retrieval tasks and network architectures demonstrate that our attack achieves higher fooling rates than conventional adversarial attacks, while producing high-impact perturbations with deeper insights into vulnerabilities of hyperbolic embeddings. This work highlights the importance of geometry-aware adversarial strategies in curved representation spaces and provides a principled framework for attacking hierarchical embeddings.

Xianglong Jin, Zheng Wang, Rong Wang, Feiping Nie

Self-supervised 3D point cloud understanding is crucial for scene understanding, where Masked Autoencoders (MAE) have achieved excellent performance in point cloud representation learning. However, existing MAE-style methods fail to consider spatial-semantic variations in masking strategies, and joint learning with multi-view images often overlooks view redundancy. To address these challenges, we propose an MAE framework enhanced with reliable multi-view 2D-3D Key-part alignment and Reinforced masking, named as KR-MAE. Our approach comprises three key innovations: Reinforced Masking (RM) strategically samples visible tokens based on semantic saliency to enhance reconstruction fidelity; Reliable Multi-View Selector (RVS) dynamically refines the most informative image subset by filtering occluded or low-texture views, mitigating detrimental redundancy; Reliable-view 2D-3D Key-part Aligned Transformer (KAT) establishes semantic-aligned correspondence between salient 3D point cloud parts and reliable multi-view 2D image patches, leveraging rich texture cues from 2D images to compensate for sparse geometry in point cloud. Extensive experiments on 3D classification and segmentation benchmarks demonstrate that KR-MAE achieves state-of-the-art performance, surpassing prior multi-modal methods.

Yuanmin Huang, Wenxuan Li, Mi Zhang, Xiaohan Zhang, Xiaoyu You, Min Yang

Deep neural networks have recently achieved notable progress in 3D point cloud recognition, yet their vulnerability to adversarial perturbations poses critical security challenges in practical deployments. Conventional defense mechanisms struggle to address the evolving landscape of multifaceted attack patterns. Through systematic analysis of existing defenses, we identify that their unsatisfactory performance primarily originates from an entangled feature space, where adversarial attacks can be performed easily. To this end, we present 3D-ANC, a novel approach that capitalizes on the Neural Collapse (NC) mechanism to orchestrate discriminative feature learning. In particular, NC depicts where last-layer features and classifier weights jointly evolve into a simplex equiangular tight frame (ETF) arrangement, establishing maximally separable class prototypes. However, leveraging this advantage in 3D recognition confronts two substantial challenges: (1) prevalent class imbalance in point cloud datasets, and (2) complex geometric similarities between object categories. To tackle these obstacles, our solution combines an ETF-aligned classification module with an adaptive training framework consisting of representation-balanced learning (RBL) and dynamic feature direction loss (FDL). 3D-ANC seamlessly empowers existing models to develop disentangled feature spaces despite the complexity in 3D data distribution. Comprehensive evaluations state that 3D-ANC significantly improves the robustness of models with various structures on two datasets. For instance, DGCNN's classification accuracy is elevated from 27.2% to 80.9% on ModelNet40 -- a 53.7% absolute gain that surpasses leading baselines by 34.0%.

Weiquan Huang, Aoqi Wu, Yifan Yang, Xufang Luo, Yuqing Yang, Usman Naseem, Chunyu Wang, Qi Dai, Xiyang Dai, Dongdong Chen 等

CLIP is a seminal multimodal model that maps images and text into a shared representation space by contrastive learning on billions of image–caption pairs. Inspired by the rapid progress of large language models (LLMs), we investigate how the superior linguistic understanding and broad world knowledge of LLMs can further strengthen CLIP—particularly in handling long, complex captions. We introduce an efficient fine-tuning framework that embeds an LLM into a pretrained CLIP while incurring almost the same training cost as regular CLIP fine-tuning. Our method first “embedding-izes” the LLM for the CLIP setting, then couples it to the pretrained CLIP vision encoder through a lightweight adaptor trained on only a few million image–caption pairs. With this strategy we achieve large performance gains—without large-scale retraining—over state-of-the-art CLIP variants such as EVA02 and SigLIP-2. The LLM-enhanced CLIP delivers consistent improvements across a wide spectrum of downstream tasks, including linear-probe classification, zero-shot image–text retrieval with both short and long captions (in English and other languages), zero-shot/supervised image segmentation, object detection, and used as tokenizer for multimodal large-model benchmarks.

Yudong Hu, Yueju Han, Rui Sun, Jinke Ren

Capsule Network (CapsNet) has demonstrated significant potential in visual recognition by capturing spatial relationships and part-whole hierarchies for learning equivariant feature representations. However, existing CapsNet and variants often rely on a single high-level feature map, overlooking the rich complementary information provided by multi-scale features. Furthermore, conventional feature fusion strategies, such as addition and concatenation, struggle to reconcile multi-scale feature discrepancies, leading to suboptimal classification performance. To address these limitations, we propose the Multi-Scale Patchify Capsule Network (MSPCaps), a novel architecture that integrates multi-scale feature learning and efficient capsule routing. Specifically, MSPCaps consists of three key components: a Multi-Scale ResNet Backbone (MSRB), a Patchify Capsule Layer (PatchifyCaps), and a Cross-Agreement Routing (CAR) block. First, the MSRB extracts diverse multi-scale feature representations from input images, preserving both fine-grained details and global contextual information. Second, the PatchifyCaps partitions these multi-scale features into primary capsules using a uniform patch size, equipping the model with the ability to learn from diverse receptive fields. Finally, the CAR block adaptively routes the multi-scale capsules by identifying cross-scale prediction pairs with maximum agreement. Unlike the simple concatenation of multiple self-routing blocks, CAR ensures that only the most coherent capsules (best part-to-whole pairs) contribute to the final voting. Our proposed MSPCaps achieves remarkable scalability and superior robustness, consistently surpassing multiple baseline methods in terms of classification accuracy, with configurations ranging from a highly efficient Tiny model (344.3K parameters) to a powerful Large model (10.9M parameters), highlighting its potential in advancing feature representation learning.

Qiang Hu, Qimei Wang, Yingjie Guo, Qiang Li, Zhiwei Wang

White-Light Imaging (WLI) is the standard for endoscopic cancer screening, but Narrow-Band Imaging (NBI) offers superior diagnostic details. A key challenge is transferring knowledge from NBI to enhance WLI-only models, yet existing methods are critically hampered by their reliance on paired NBI-WLI images of the same lesion, a costly and often impractical requirement that leaves vast amounts of clinical data untapped. In this paper, we break this paradigm by introducing PaGKD, a novel Pairing-free Group-level Knowledge Distillation framework that that enables effective cross-modal learning using unpaired WLI and NBI data. Instead of forcing alignment between individual, often semantically mismatched image instances, PaGKD operates at the group level to distill more complete and compatible knowledge across modalities. Central to PaGKD are two complementary modules: (1) Group-level Prototype Distillation (GKD-Pro) distills compact group representations by extracting modality-invariant semantic prototypes via shared lesion-aware queries; (2) Group-level Dense Distillation (GKD-Den) performs dense cross-modal alignment by guiding group-aware attention with activation-derived relation maps. Together, these modules enforce global semantic consistency and local structural coherence without requiring image-level correspondence. Extensive experiments on four clinical datasets demonstrate that PaGKD consistently and significantly outperforms state-of-the-art methods, boosting AUC by 3.3%, 1.1%, 2.8%, and 3.2%, respectively, establishing a new direction for cross-modal learning from unpaired data.

Chenyu Hu, Xiaotong Li, Hao Zhu, Biao Hou

Point cloud processing has become a cornerstone technology in many 3D vision tasks. However, arbitrary rotations introduce variations in point cloud orientations, posing a long-standing challenge for effective representation learning. The core of this issue is the disruption of the point cloud's intrinsic directional characteristics caused by rotational perturbations. Recent methods attempt to implicitly model rotational equivariance and invariance, preserving directional information and propagating it into deep semantic spaces. Yet, they often fall short of fully exploiting the multiscale directional nature of point clouds to enhance feature representations. To address this, we propose the Direction-Perceptive Vector Network (DiPVNet). At its core is an atomic dot-product operator that simultaneously encodes directional selectivity and rotation invariance--endowing the network with both rotational symmetry modeling and adaptive directional perception. At the local level, we introduce a Learnable Local Dot-Product (L2DP) Operator, which enables interactions between a center point and its neighbors to adaptively capture the non-uniform local structures of point clouds. At the global level, we leverage generalized harmonic analysis to prove that the dot-product between point clouds and spherical sampling vectors is equivalent to a direction-aware spherical Fourier transform (DASFT). This leads to the construction of a global directional response spectrum for modeling holistic directional structures. We rigorously prove the rotation invariance of both operators. Extensive experiments on challenging scenarios involving noise and large-angle rotations demonstrate that DiPVNet achieves state-of-the-art performance on point cloud classification and segmentation tasks.

Jiaxun Guo, Manar Amayri, Nizar Bouguila, Xin Liu, Wentao Fan

Recent advances in rotation-invariant (RI) learning for 3D point clouds typically replace raw coordinates with handcrafted RI features to ensure robustness under arbitrary rotations. However, these approaches often suffer from the loss of global pose information, making them incapable of distinguishing geometrically similar but spatially distinct structures. We identify that this limitation stems from the restricted receptive field in existing RI methods, leading to Wing–tip feature collapse, a failure to differentiate symmetric components (e.g., left and right airplane wings) due to indistinguishable local geometries. To overcome this challenge, we introduce the Shadow-informed Pose Feature (SiPF), which augments local RI descriptors with a globally consistent reference point (referred to as the “shadow”) derived from a learned shared rotation. This mechanism enables the model to preserve global pose awareness while maintaining rotation invariance. We further propose Rotation-invariant Attention Convolution (RIAttnConv), an attention-based operator that integrates SiPFs into the feature aggregation process, thereby enhancing the model’s capacity to distinguish structurally similar components. Additionally, we design a task-adaptive shadow locating module based on the Bingham distribution over unit quaternions, which dynamically learns the optimal global rotation for constructing consistent shadows. Extensive experiments on 3D classification and part segmentation benchmarks demonstrate that our approach substantially outperforms existing RI methods, particularly in tasks requiring fine-grained spatial discrimination under arbitrary rotations.

Shizhan Gong, Xiaofan Zhang, Qi Dou

Deep learning has achieved remarkable success in image recognition, yet their inherent opacity poses challenges for deployment in critical domains. Concept-based interpretations aim to address this by explaining model reasoning through human-understandable concepts. However, existing post-hoc methods and ante-hoc concept bottleneck models (CBMs), suffer from limitations such as unreliable concept relevance, non-visual or labor-intensive concept definitions, and model/data-agnostic assumptions. This paper introduces Post-hoc Concept Bottleneck Model via Representation Decomposition (PCBM-ReD), a novel pipeline that retrofits interpretability onto pretrained opaque models. PCBM-ReD automatically extracts visual concepts from a pre-trained encoder, employs multimodal large language models (MLLMs) to label and filter concepts based on visual identifiability and task relevance, and selects an independent subset via reconstruction-guided optimization. Leveraging CLIP’s visual-text alignment, it decomposes image representations into linear combination of concept embeddings to fit into the CBMs abstraction. Extensive experiments across 11 image classification tasks show PCBM-ReD achieves state-of-the-art accuracy, narrows the performance gap with end-to-end models, and exhibits better interpretability.

Qinghao Gao, Jiahui Qu, Wenqian Dong

Multimodal remote sensing image joint classification has achieved significant progress. However, existing methods primarily focus on designing modality-specific networks, lacking adaptive generalization capabilities in diverse and dynamic modality combinations encountered in real-world scenarios. Inspired by the generalization capabilities of visual foundation model in downstream tasks, we propose a unified Text-guided Arbitrary Modalitiy Prompting (T-APT) framework, which leverages complementary fused features to drive the foundation model and employs text-guided modality-specific prior knowledge as cross-modal prompts to fine-tune a pretrained Vision Transformer (ViT) model. Specifically, a Mamba-Based Arbitrary Modal-Focused Feature Capture (MAMF-FC) module is designed to extract complementary joint features and modality-specific prior knowledge from arbitrary modalities through a shared-specific scanning encoder-decoder architecture. Subsequently, a Text-Guided Modality-Aware Prompt Tuning (TMPT) module is proposed to support the adaptation of fused features to the foundation model, enabling our arbitrary remote sensing image classification task. Extensive experiments on public datasets spanning multispectral (MS), hyperspectral (HS), light detection and ranging (LiDAR), and synthetic aperture radar (SAR) modalities demonstrate that our T-APT achieves classification performance comparable to specialized networks across arbitrary modal combinations.

Xiang Fang, Wanlong Fang

Large Language Models (LLMs) are increasingly vulnerable to adversarial prompts that exploit semantic ambiguities to bypass safety mechanisms, resulting in harmful or inappropriate outputs. Such attacks, including jailbreaking and prompt injection, pose significant risks to the integrity and availability of LLMs in security-critical applications. This paper proposes the Adversarial Prompt Disentanglement (APD) framework, a novel defense mechanism that proactively identifies and neutralizes malicious components in input prompts before they are processed by the LLM. The APD framework integrates three key innovations: (1) a mutual information- based semantic decomposition method to isolate adversarial and benign prompt components, ensuring statistical in- dependence; (2) a graph-based intent classification approach that leverages spectral analysis to detect malicious patterns in prompt semantics; and (3) a lightweight transformer-based classifier trained on real-world datasets of toxic and jailbreaking prompts, enabling efficient and accurate adversarial intent detection. Evaluated on diverse datasets containing adversarial prompts, APD demonstrates superior robustness, reducing harmful output generation by over 85% while maintaining negligible impact on model performance. The framework’s computational efficiency supports real-time deploy- ment, making it a practical solution for securing LLMs. Our work addresses critical challenges in machine learning security on novel attacks and integrity methods for ML systems, and offers a scalable, ethically grounded defense against prompt-based adversarial threats.

Ruxi Deng, Wenxuan Bao, Tianxin Wei, Jingrui He

Pretrained vision-language models exhibit strong zero-shot classification capabilities, but their predictions degrade significantly under common image corruptions. To improve robustness, many test-time adaptation (TTA) methods adopt positive data augmentation (PDA), which generates multiple views of each test sample to reduce prediction variance. However, these methods suffer from two key limitations. First, it introduces considerable computational overhead due to the large number of augmentations required per image. Second, it fails to mitigate prediction bias, where the model tends to predict certain classes disproportionately under corruption, as PDA operates on corrupted inputs and typically does not remove the corruption itself. To address these challenges, we propose Panda, a novel TTA method based on negative data augmentation (NDA). Unlike positive augmentations that preserve object semantics, Panda generates negative augmentations by disrupting semantic content. It divides images into patches and randomly assembles them from a shared patch pool. These negatively augmented images retain corruption-specific features while discarding object-relevant signals. We then subtract the mean feature of these negative samples from the original image feature, effectively suppressing corruption-related components while preserving class-relevant information. This mitigates prediction bias under distribution shifts. Importantly, Panda allows augmentation to be shared across samples within a batch, resulting in minimal computational overhead. Panda can be seamlessly integrated into existing test-time adaptation frameworks and substantially improve their robustness. Our experiments indicate that Panda delivers superior performance compared to PDA methods, and a wide range of TTA methods exhibit significantly enhanced performance when integrated with Panda.

Thinh Dao, Khoa D Doan, Kok-Seng Wong

Deep Neural Networks (DNNs) are shown to be vulnerable to backdoor poisoning attacks, with most research focusing on digital triggers that consist of artificial patterns added to test-time inputs to induce targeted misclassification. Physical triggers, which are natural objects embedded in real-world scenes, offer a promising alternative for attackers as they can activate backdoors in real-time without digital manipulation. However, existing physical backdoor attacks are dirty-label, meaning that attackers must change the labels of poisoned inputs to the target label. The inconsistency between image content and label exposes the attack to human inspection, reducing its stealthiness in real-world settings. To address this limitation, we introduce Clean-Label Physical Backdoor Attack (CLPBA), a new paradigm of physical backdoor attack that does not require label manipulation and trigger injection at the training stage. Instead, the attacker injects imperceptible perturbations into a small number of target class samples to backdoor a model. By framing the attack as a Dataset Distillation problem, we develop three CLPBA variants, namely Parameter Matching, Gradient Matching, and Feature Matching, that craft effective poisons under both linear probing and full-finetuning training settings. In hard scenarios that require backdoor generalizability in the physical world, CLPBA is shown to even surpass Dirty-label attack baselines. We demonstrate the effectiveness of CLPBA via extensive experiments on two collected physical backdoor datasets for facial recognition and animal classification.

Chenxu Dang, Haiyan Liu, Jason Bao, Pei An, Xinyue Tang, An Pan, Jie Ma, Bingchuan Sun, Yan Wang

Semantic occupancy has emerged as a powerful representation in world models for its ability to capture rich spatial semantics. However, most existing occupancy world models rely on static and fixed embeddings or grids, which inherently limit the flexibility of perception. Moreover, their ``in-place classification" over grids exhibits a potential misalignment with the dynamic and continuous nature of real scenarios. In this paper, we propose SparseWorld, a novel 4D occupancy world model that is flexible, adaptive, and efficient, powered by sparse and dynamic queries. We propose a Range-Adaptive Perception module, in which learnable queries are modulated by the ego vehicle states and enriched with temporal-spatial associations to enable extended-range perception. To effectively capture the dynamics of the scene, we design a State-Conditioned Forecasting module, which replaces classification-based forecasting with regression-guided formulation, precisely aligning the dynamic queries with the continuity of the 4D environment. In addition, We specifically devise a Temporal-Aware Self-Scheduling training strategy to enable smooth and efficient training. Extensive experiments demonstrate that SparseWorld achieves state-of-the-art performance across perception, forecasting, and planning tasks. Comprehensive visualizations and ablation studies further validate the advantages of SparseWorld in terms of flexibility, adaptability, and efficiency.

Yarui Chen, Lehan Hong, Jianlin Shao, Jianning Yang, Tingting Zhao, Yun Liao, Yancui Shi

Variational autoencoder (VAE)-based frameworks possess a natural advantage in modeling the shared and private information inherent in multimodal data. However, current models focus on improving the quality of shared representations from the reconstruction perspective, lacking explicit mechanisms to model their underlying semantic structure. In this paper, we propose the multimodal Gaussian mixture variational autoencoder with consistency regularizations, which introduces a Gaussian mixture prior over the shared latent space to enhance its semantic structure and encourage the formation of cluster-aware latent representations. To address the cross-modal inconsistency problem under missing modality conditions, we propose a cluster-guided regularization strategy that enforces the cross-modal consistency using the pseudo-category labels from unsupervised clustering. Additionally, we design a self-supervised contrastive regularization strategy to align semantically similar representations across modalities. Extensive experiments on MNIST-SVHN and MNIST-CDCB datasets demonstrate that our method significantly outperforms prior state-of-the-art models in generation, classification, and retrieval tasks.

Yangkai Chen, Qiangqiang Wu, Guangyao Li, Junlong Gao, Guanglin Niu, Hanzi Wang

Open-vocabulary multi-object tracking (OV-MOT) aims to track objects with unseen categories beyond the training set. While existing methods rely on pseudo video sequences synthesized from static images, they struggle to model realistic motion patterns, resulting in limited association performance in real-world scenarios. To alleviate these issues, we propose SAM2-OV, a novel association learning-free OV-MOT method that adopts a detection-only tuning paradigm, eliminating the need for synthetic sequences or spatiotemporal supervision and substantially reducing the overall learnable parameters. The core of our method is a Unified Detection Module (UDM), which effectively provides object-level prompts to enable SAM2 for OV-MOT. Enabled by UDM, SAM2-OV is the first to integrate SAM2 for OV-MOT, fully unleashing its zero-shot cross-frame association ability. To further enhance object association under occlusion and abrupt motion, we introduce a Motion Prior Assistance Module (MPAM) that incorporates motion cues into the mask selection process. In addition, a Semantic Enhancement Adapter (SEA) distilled from CLIP is used to improve classification generalization. A sparse prompting strategy is also adopted to reduce computational redundancy by triggering detection only on selected keyframes. As only the detection module is tuned on static images, the overall training process remains simple and efficient. Experiments on the TAO dataset demonstrate that SAM2-OV achieves state-of-the-art performance under the TETA metric, particularly on novel categories. Evaluations on the KITTI dataset show the strong zero-shot cross-domain transferability of our SAM2-OV.

Yang Chen, Xiaowei Xu, Shuai Wang, Chenhui Zhu, Ruxue Wen, Xubin Li, Tiezheng Ge, Limin Wang

Normalizing Flows (NFs) are a class of generative models distinguished by a mathematically invertible architecture, where the forward pass transforms data into a latent space for density estimation, and the reverse pass generates new samples from this space. This characteristic creates an intrinsic synergy between representation learning and data generation. However, the generative quality of standard NFs is limited by poor semantic representations from log-likelihood optimization. To remedy this, we propose a novel alignment strategy that creatively leverages the invertibility of NFs: instead of regularizing the forward pass, we align the intermediate features of the generative (reverse) pass with representations from a powerful vision foundation model, demonstrating superior effectiveness over naive alignment. We also introduce a novel training-free, test-time optimization algorithm for classification, which provides a more intrinsic evaluation of the NF's embedded semantic knowledge. Comprehensive experiments demonstrate that our approach accelerates the training of NFs by over 3.3x, while simultaneously delivering significant improvements in both generative quality and classification accuracy. New state-of-the-art results for NFs are established on ImageNet 64 x 64 and 256 x 256.

Taiqin Chen, Yifeng Wang, Xiaochen Feng, Zhilin Zhu, Hao Sha, Yingjian Li, Yongbing Zhang

While hyperspectral images (HSI) benefit from numerous spectral channels that provide rich information for classification, the increased dimensionality and sensor variability make them more sensitive to distributional discrepancies across domains, which in turn can affect classification performance. To tackle this issue, hyperspectral single-source domain generalization (SDG) typically employs data augmentation to simulate potential domain shifts and enhance model robustness under the condition of single-source domain training data availability. However, blind augmentation may produce samples misaligned with real-world scenarios, while excessive emphasis on realism can suppress diversity, highlighting a tradeoff between realism and diversity that limits generalization to target domains. To address this challenge, we propose a spectral property-driven data augmentation (SPDDA) that explicitly accounts for the inherent properties of HSI, namely the device-dependent variation in the number of spectral channels and the mixing of adjacent channels. Specifically, SPDDA employs a spectral diversity module that resamples data from the source domain along the spectral dimension to generate samples with varying spectral channels, and constructs a channel-wise adaptive spectral mixer by modeling inter-channel similarity, thereby avoiding fixed augmentation patterns. To further enhance the realism of the augmented samples, we propose a spatial-spectral co-optimization mechanism, which jointly optimizes a spatial fidelity constraint and a spectral continuity self-constraint. Moreover, the weight of the spectral self-constraint is adaptively adjusted based on the spatial counterpart, thus preventing over-smoothing in the spectral dimension and preserving spatial structure. Extensive experiments conducted on three remote sensing benchmarks demonstrate that SPDDA outperforms state-of-the-art methods.

Luyao Chang, Leiting Chen, Chen Yang, Chuan Zhou

Weakly supervised semantic segmentation (WSSS) suffers from an inherent mismatch between coarse image-level annotations and dense pixel-level predictions. To bridge this gap, existing methods primarily focus on generating refined class activation maps (CAM) as pseudo-labels. However, we argue that this focus is insufficient as it overlooks a critical component: the segmentation decoder. The decoder is typically trained through superficial alignment of predictions with pseudo-labels in the logit space. Given the noisy nature of such labels, this naive supervision leads to error accumulation and limits performance. To address this, we propose an Uncertainty-Guided Reliable Learning (UGRL) framework that exerts dual control to reshape the learning process, achieving robust supervision that escapes the CAM shadow. The cornerstone of UGRL is a prototype-driven uncertainty modeling module that estimates the reliability of class-wise supervision. The modeled uncertainty enables two synergistic control mechanisms. First, it adaptively modulates classification and segmentation losses, encouraging the model to learn from more trustworthy signals. Second, it guides the structuring of the decoder’s feature space. Rather than relying solely on superficial alignment, UGRL enforces deeper representation alignment by applying contrastive learning on reliable pixels. This enables rich semantic transfer to fine-grained segmentation details. Extensive experiments on PASCAL VOC and MS COCO demonstrate that our method surpasses other state-of-the-art WSSS methods.

Qing Cai, GuihaoYan, Fan Zhang, Cheng Zhang, Zhi Liu

Ultrasound standard plane recognition is essential for clinical tasks such as disease screening, organ evaluation, and biometric measurement. However, existing methods fail to effectively exploit shallow structural information and struggle to capture fine-grained semantic differences through contrastive samples generated by image augmentations, leading to poor recognition of structural and discriminative details in ultrasound standard planes. To address these issues, we propose Structure-Enhanced Mixture-of-Experts Contrastive Learning (SEMC), a novel framework that combines structure-aware feature fusion with expert-guided contrastive learning. Specifically, we propose a Semantic-Structure Fusion Module (SSFM) to exploit multi-scale structural information and enhance the model's ability to perceive fine-grained structural details by effectively aligning shallow and deep features. Meanwhile, a Mixture-of-Experts Contrastive Recognition Module (MCRM) is designed to perform hierarchical contrastive learning and classification across multi-level features using a mixture-of-experts (MoE) mechanism, further improving class separability and overall recognition performance. More importantly, we also curate a large-scale and meticulously annotated liver ultrasound dataset containing six standard planes. Extensive experimental results on our in-house dataset and two public datasets demonstrate that SEMC outperforms recent state-of-the-art methods across various metrics.