检索结果-内蒙古大学图书馆

27th International Conference on Pattern Recognition, ICPR 2024

作者： Li, Xinjie Tang, Jianxiong Lai, Jianhuang School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Guangzhou510555 China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou China

ISBN: (纸本)9783031781858

In recent years, spiking neural networks (SNNs) have gained significant attention in visual recognition tasks due to the low computational energy. However, most SNNs have a large number of parameters, which limits their use on resource-limited devices. In this paper, we propose an Ensemble Binary Spiking Neural Network (EB-SNN) for accurate and memory-friendly visual recognition. The EB-SNN is modeled by Ensemble Binary Weights (EBW) module, which integrates multiple binary weights for lightweight SNN modeling. Meanwhile, we propose Knowledge Alignment Strategy to ensure that the EB-SNN can approximate a well-trained SNN for good performance. Experimental results show that the EB-SNN can achieve accuracy of 95.39% on CIFAR10, using 9.3% memory of full-precision SNN. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Local binary pattern

来源：评论

学校读者我要写书评

暂无评论

Breaking Information Silos: Global Guided Task Prediction for Class-Incremental Learning 27th

Breaking Information Silos: Global Guided Task Prediction fo...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Hu, Chaoshun Ye, Biaohua Chen, Zixuan Lai, Jian-Huang School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Guangzhou510555 China Ministry of Education Key Laboratory of Machine Intelligence and Advanced Computing Guangzhou China

ISBN: (纸本)9783031781889

Class-incremental learning (CIL) aims to learn a series of tasks sequentially, each introducing several new categories. Because providing the task labels during inference can significantly increase accuracy, many approaches attempt to predict task labels in CIL. However, existing works focus on learning local information and overlook the importance of global information. The absence of global guidance leads to the formation of information silos across disparate tasks, resulting in potential inter-task interference. To break information silos, we propose a method called Global Guided Task Prediction (GGTP) to introduce global information. Our method consists of two modules. The local de-redundant module aims to reduce information redundancy across different tasks from a global perspective. We combine dual encoders and feature decorrelation loss to effectively reduce redundancy while minimizing catastrophic forgetting. The global information module explicitly extracts global information to serve as auxiliary information guiding task prediction. We extract the most important information from all local information through a global encoder and then aggregate them to form global information. Extensive experiments validate the effectiveness of our method, which achieves state-of-the-art results. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Redundancy

来源：评论

学校读者我要写书评

暂无评论

Confidence-Guided Feature Alignment for Cloth-Changing Person Re-identification 27th

Confidence-Guided Feature Alignment for Cloth-Changing Pers...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Huang, Sirong Zheng, Huicheng School of Computer Science and Engineering Sun Yat-Sen University Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou China Guangdong Province Key Laboratory of Information Security Technology Guangzhou China

ISBN: (纸本)9783031781094

Much progress has been made in the field of person re-identification, but changes in clothing have hindered the practical application of long-term person re-identification. Cloth-changing person re-identification (CC-ReID) aims to address this problem, with the main challenge being the extraction of discriminative features unrelated to clothing. Existing methods, which mainly focus on introducing clothing-irrelevant cues such as key points, contours, and 3D shapes, require additional modules for feature extraction, resulting in increased complexity and potential inaccuracies due to dependence on the performance of external models. Few studies directly use the original RGB images to make the model constantly focus on clothing-independent information. In this paper, we propose a Confidence-Guided Feature Alignment Network (CGFA) for CC-ReID. Specifically, we design a confidence module that automatically learns to make confidence adjustments to fine-grained information, prompting the model to mine clothing-independent discriminative features without introducing other modal cues. By transferring knowledge, we encourage the model to learn discriminative identity features that are independent of clothing bias. As a result, the confidence module can be removed during the inference phase. The proposed simple but efficient method uses only RGB modality without additional cues, and can serve as a powerful baseline for CC-ReID to drive future research. Extensive experiments on the CC-ReID datasets demonstrate the effectiveness of the proposed method, which achieves state-of-the-art performance. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Feature extraction

来源：评论

学校读者我要写书评

暂无评论

Reflectance Recovery Guided Learning of Illumination-Invariant Features for Person Re-Identification 18th

Reflectance Recovery Guided Learning of Illumination-Invari...

引用

18th Chinese Conference on Biometric Recognition, CCBR 2024

作者： Chen, Xianbiao Xie, Xiaohua School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Guangdong Province Key Laboratory of Information Security Technology Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou China

ISBN: (纸本)9789819610679

In real-world scenarios, the changes of illumination are noticeable and can significantly impact the performance of re-identification (ReID) algorithms. However, the existing person ReID methods predominantly concentrate on tackling challenges in scenarios with minimal illumination fluctuations. The key to tackling this issue is to extract illumination-invariant features. So we propose a joint learning framework which combines the recovery of reflectance map and extraction of illumination-invariant feature. By sharing parameters, the proposed module, which can be seamlessly detached during the inference phase, contributes to a reduction in inference computation time. We also introduce an adversarial learning mechanism, utilizing illumination category and person identity to facilitate the extraction of features that are invariant to illumination. Besides, due to the lack of person ReID dataset containing images under diverse lighting conditions, we construct a real-world dataset called ICReID in which images have drastic illumination changes. Extensive experiments demonstrate the effectiveness of the proposed method, which achieves significant performance on both synthetic and real datasets. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

OCLNet: Obfuscation feature Contrastive Learning Network for Weakly Supervised Semantic Segmentation on Ultrasound Images

OCLNet: Obfuscation feature Contrastive Learning Network for...

引用

2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025

作者： Gao, Jie Zhang, Xianzhi Zhang, Zijian Li, Xuewei Yu, Mei Yu, Ruiguo Liu, Zhiqiang College of Intelligence and Computing Tianjin University Tianjin300350 China Tianjin Key Laboratory of Cognitive Computing and Application Tianjin300350 China Tianjin Key Laboratory of Advanced Networking Tianjin300350 China

ISBN: (纸本)9798350368741

Deep learning-based semantic segmentation technology has become a critical tool in assisting doctors with automatic lesion segmentation in medical images. However, the high cost of acquiring large-scale, pixel-level annotations poses a significant challenge, limiting the scalability and application of fully supervised semantic segmentation models. To address this, weakly supervised learning-based semantic segmentation models have emerged as a promising solution. These models can accurately segment lesion regions using only weak annotations, such as image-level or frame-level labels, significantly reducing the annotation burden. This approach has gained substantial attention in current *** various medical imaging modalities, ultrasound imaging stands out as a primary diagnostic tool due to its rapid imaging speed, ease of use, and accessibility. This paper focuses on the study of thyroid ultrasound imaging, aiming to achieve accurate classification of nodule regions. The goal is to provide clinicians with more precise diagnostic information, improving decision-making in thyroid disease diagnosis. © 2025 IEEE.

关键词： Semantic Segmentation Thyroid Nodule Ultrasound Images Weakly Supervised Learning

来源：评论

学校读者我要写书评

暂无评论

Depth-Enhanced Alignment for Label-Free 3D Semantic Segmentation 27th

Depth-Enhanced Alignment for Label-Free 3D Semantic Segment...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Xie, Shangjin Feng, Jiawei Chen, Zibo Liu, Zhixuan Zheng, Wei-Shi School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou China Guangdong Province Key Laboratory of Information Security Technology Sun Yat-sen University Guangzhou China

ISBN: (纸本)9783031784552

Labeling every point in a scene is a laborious journey for 3D understanding. To achieve annotation-free training, existing works introduce Contrastive Language-Image Pre-training (CLIP) to transfer the pre-trained capability of visual-linguistic correspondence to 3D-linguistic matching. However, directly adopting this CLIP-driven strategy can inevitably introduce bias: The overrated roles of the color and texture from an RGB image could overshadow the geometric nature of the corresponding 3D scene, resulting in a sub-optimal alignment. We note that different from RGB images, a depth map contains rich geometric information. Inspired by this, we propose Depth-Enhanced Alignment (D-EA) for label-free 3D semantic segmentation. D-EA aims to explore the rich geometric cues in depth maps and mitigate the color and texture biases rooted in the original CLIP-driven strategy. Specifically, we first tune a geometry-enhanced CLIP by aligning its depth prediction to the paired RGB prediction given by the original CLIP. Next, the point cloud feature space is matched with the RGB-Depth aggregated CLIP space by aligning point prediction to RGB and depth predictions. Moreover, to mitigate the semantic ambiguity caused by view-specific noise, we propose a View-Integrated Pseudo Label Generation paradigm. Experiments demonstrate the effectiveness of the proposed D-EA on the ScanNet (indoor) and GraspNet-1Billion (desktop) datasets in the label-free setting. Our method is also competitive in limited annotation semantic segmentation. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Semantic Segmentation

来源：评论

学校读者我要写书评

暂无评论

iGrasp: An Interactive 2D-3D Framework for 6-DoF Grasp Detection 27th

iGrasp: An Interactive 2D-3D Framework for 6-DoF Grasp Detec...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Jiang, Jian-Jian Wu, Xiao-Ming Chen, Zibo Wei, Yi-Lin Zheng, Wei-Shi School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou China Guangdong Province Key Laboratory of Information Security Technology Sun Yat-sen University Guangzhou China

ISBN: (纸本)9783031781124

For 6-DoF grasp detection, we aim at introducing a new interactive 2D-3D framework which filters out irrelevant information and makes both modalities collaborate effectively to generate robust grasps and accelerate inference speed greatly. This cannot be accomplished by existing works on learning to grasp that merely utilize 3D point clouds or leverage both 2D textures and 3D point clouds. Our framework is called iGrasp, a novel three-step design between 2D textures and 3D point clouds, where the interaction modelling enhances both modalities. Concretely, we propose the 2D-to-3D interaction to leverage objectness masks generated from 2D textures to filter out target-irrelevant information in 3D point clouds. Then, we introduce the 3D-to-2D interaction to leverage structural priors from 3D point cloud features with cross-attention and cylinder grouping to refine 2D texture features. Finally, we combine the refined 2D texture features and 3D point cloud features for generating high-quality 6-DoF grasp poses. Our experiments on the large-scale real-world dataset, namely GraspNet-1Billion, demonstrate that iGrasp surpasses state-of-the-art methods by 4.66/3.53 mAP on RealSense/Kinect and reduces the inference time by 28%. Real-world experiments further verify the effectiveness of iGrasp. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

Privacy-Preserving Face Recognition with Adaptive Generative Perturbations 27th

Privacy-Preserving Face Recognition with Adaptive Generative...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Zhang, Delong Peng, Yixing Wu, Ancong Zheng, Wei-Shi School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou China Guangdong Province Key Laboratory of Information Security Technology Sun Yat-sen University Guangzhou China

ISBN: (纸本)9783031783401

For online face recognition services, the potential leakage of facial features and reconstruction techniques gives malicious attackers the opportunity to reconstruct face images, raising public concern about privacy. Previous privacy-preserving face recognition methods either require retraining the face recognition model or iterative perturbation that increases inference time. To overcome these limitations, we propose an efficient plug-and-play method that crafts Adaptive Generative perturbations for frozen Face recognition model (AGFace) to defend the reconstruction attacks. To generate perturbation with a single forward for shallow features extracted by frozen recognition model, we learn a Perturbation Generator to efficiently mine the adversarial perturbations by simultaneously minimizing the perturbation and maximizing the reconstruction error of a proxy reconstructor. To achieve privacy-utility trade-off, we propose the Adaptive Channel Selector to identify top-k reconstruction-sensitive channels for the features. By selecting these channels for perturbation, the dominant visual privacy information is protected with reconstruction-insensitive discriminative information preserved. Extensive experiments demonstrate that AGFace achieves state-of-the-art performance in terms of both privacy and utility among retraining-free methods and is comparable to retraining-dependent methods. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Differential privacy

来源：评论

学校读者我要写书评

暂无评论

Towards Completeness: A Generalizable Action Proposal Generator for Zero-Shot Temporal Action Localization 27th

Towards Completeness: A Generalizable Action Proposal Genera...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Du, Jia-Run Lin, Kun-Yu Meng, Jingke Zheng, Wei-Shi School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Guangzhou China Guangdong Province Key Laboratory of Information Security Technology Sun Yat-sen University Guangzhou China

ISBN: (纸本)9783031784439

To address the zero-shot temporal action localization (ZSTAL) task, existing works develop models that are generalizable to detect and classify actions from unseen categories. They typically develop a category-agnostic action detector and combine it with the Contrastive Language-Image Pre-training (CLIP) model to solve ZSTAL. However, these methods suffer from incomplete action proposals generated for unseen categories, since they follow a frame-level prediction paradigm and require hand-crafted post-processing to generate action proposals. To address this problem, in this work, we propose a novel model named Generalizable Action Proposal generator (GAP), which can interface seamlessly with CLIP and generate action proposals in a holistic way. Our GAP is built in a query-based architecture and trained with a proposal-level objective, enabling it to estimate proposal completeness and eliminate the hand-crafted post-processing. Based on this architecture, we propose an Action-aware Discrimination loss to enhance the category-agnostic dynamic information of actions. Besides, we introduce a Static-Dynamic Rectifying module that incorporates the generalizable static information from CLIP to refine the predicted proposals, which improves proposal completeness in a generalizable manner. Our experiments show that our GAP achieves state-of-the-art performance on two challenging ZSTAL benchmarks, i.e., Thumos14 and ActivityNet1.3. Specifically, our model obtains significant performance improvement over previous works on the two benchmarks, i.e., +3.2% and +3.4% average mAP, respectively. The code is available at https://***/Run542968/GAP. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Zero-shot learning

来源：评论

学校读者我要写书评

暂无评论

ChainHOI: Joint-based Kinematic Chain Modeling for Human-Object Interaction Generation

arXiv

引用

arXiv 2025年

作者： Zeng, Ling-An Huang, Guohong Wei, Yi-Lin Gu, Shengbo Tang, Yu-Ming Meng, Jingke Zheng, Wei-Shi Sun Yat-sen University China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education China

We propose ChainHOI, a novel approach for text-driven human-object interaction (HOI) generation that explicitly models interactions at both the joint and kinetic chain levels. Unlike existing methods that implicitly model interactions using full-body poses as tokens, we argue that explicitly modeling joint-level interactions is more natural and effective for generating realistic HOIs, as it directly captures the geometric and semantic relationships between joints, rather than modeling interactions in the latent pose space. To this end, ChainHOI introduces a novel joint graph to capture potential interactions with objects, and a Generative Spatiotemporal Graph Convolution Network to explicitly model interactions at the joint level. Furthermore, we propose a Kinematics-based Interaction Module that explicitly models interactions at the kinetic chain level, ensuring more realistic and biomechanically coherent motions. Evaluations on two public datasets demonstrate that ChainHOI significantly outperforms previous methods, generating more realistic, and semantically consistent HOIs. Code is available here. Copyright © 2025, The Authors. All rights reserved.

关键词： Kinematics

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：