Ye Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ye Zhang has a total of 74 co-authors. Here are the top co-authors by number of shared publications:
- Yulan Guo
Sun Yat-sen University
14 shared publications - Longguang Wang
Sun Yat-sen University
9 shared publications - Shilin Zhou
College of Electronic Science and Technology, National University of Defense Technology, Changsha, China
6 shared publications - Yi Hou
College of Electronic Science and Technology, National University of Defense Technology, Changsha, China
6 shared publications - Kewei Ouyang
College of Electronic Science and Technology, National University of Defense Technology, Changsha 410073, China
5 shared publications - Qing Gao
School of Electronics and Communication Engineering, Sun Yat-sen University, Shenzhen, China
4 shared publications
Latest Publications
- 16 November 2026
- Published by Aerospace Information Research Institute, Chinese Academy of Sciences in Journal of Image and Graphics
- Vol. 31 (11) , 4005-4020
面向自监督学习的课程式动态数据剪枝方法
目的动态数据剪枝是提升自监督预训练效率的重要手段,但现有方法多基于瞬时损失进行剪枝,难以应对自监督训练中损失波动剧烈、优化不稳定的问题,易造成关键样本被误剪和模型性能退化。针对上述缺陷,本文提出了面向自监督学习的课程式动态数据剪枝方法。方法首先设计指数损失重要性表征,通过动量系数对样本历史损失做指数移动平均,获取时序平滑且对短期损失波动鲁棒的样本价值表征,避免因瞬时损失扰动而误判关键样本;在此基础上,进一步提出课程式动态剪枝策略。该策略通过一维聚类进行样本分组,形成从易到难的课程集合,进而引入滑动窗口机制来动态控制各课程的训练进度,引导模型渐进式学习以有效保障训练过程的稳定性。结果在8个二维图像与三维点云基准数据集上,采用掩码自编码器(masked autoencoder,MAE)、动量对比学习(momentum contrast,MoCo)、点云掩码自编码器(point cloud masked autoencoder,PointMAE)等主流预训练框架,结合视觉Transformer(vision transformer,ViT)、残差网络(residual network,ResNet)等骨干网络开展了5组系统的对比实验。实验结果显示,在更高的数据剪枝率下,本文方法的性能全面优于现有的动态剪枝方法。同时,与全量数据训练策略相比,本文方法在下游任务上取得了近乎无损的性能。消融实验也验证了本文方法中各核心组件的有效性。结论本文方法通过指数损失驱动的样本重要性度量与课程式动态剪枝策略,有效突破了现有数据剪枝方法在自监督学习场景的应用局限,在性能近乎无损的前提下大幅提升了自监督预训练效率(在ImageNet-1K数据集上最高可剪枝30.6%的训练数据),为视觉模型的高效自监督训练提供了可靠的技术支撑。
- 16 November 2026
- Published by Aerospace Information Research Institute, Chinese Academy of Sciences in Journal of Image and Graphics
- Vol. 31 (11) , 4005-4020
面向自监督学习的课程式动态数据剪枝方法
目的动态数据剪枝是提升自监督预训练效率的重要手段,但现有方法多基于瞬时损失进行剪枝,难以应对自监督训练中损失波动剧烈、优化不稳定的问题,易造成关键样本被误剪和模型性能退化。针对上述缺陷,本文提出了面向自监督学习的课程式动态数据剪枝方法。方法首先设计指数损失重要性表征,通过动量系数对样本历史损失做指数移动平均,获取时序平滑且对短期损失波动鲁棒的样本价值表征,避免因瞬时损失扰动而误判关键样本;在此基础上,进一步提出课程式动态剪枝策略。该策略通过一维聚类进行样本分组,形成从易到难的课程集合,进而引入滑动窗口机制来动态控制各课程的训练进度,引导模型渐进式学习以有效保障训练过程的稳定性。结果在8个二维图像与三维点云基准数据集上,采用掩码自编码器(masked autoencoder,MAE)、动量对比学习(momentum contrast,MoCo)、点云掩码自编码器(point cloud masked autoencoder,PointMAE)等主流预训练框架,结合视觉Transformer(vision transformer,ViT)、残差网络(residual network,ResNet)等骨干网络开展了5组系统的对比实验。实验结果显示,在更高的数据剪枝率下,本文方法的性能全面优于现有的动态剪枝方法。同时,与全量数据训练策略相比,本文方法在下游任务上取得了近乎无损的性能。消融实验也验证了本文方法中各核心组件的有效性。结论本文方法通过指数损失驱动的样本重要性度量与课程式动态剪枝策略,有效突破了现有数据剪枝方法在自监督学习场景的应用局限,在性能近乎无损的前提下大幅提升了自监督预训练效率(在ImageNet-1K数据集上最高可剪枝30.6%的训练数据),为视觉模型的高效自监督训练提供了可靠的技术支撑。
- Open Access
- 1 October 2026
- Published by KEAI Publishing in Biomimetic Intelligence and Robotics
Hand gesture recognition for humanoid dexterous manipulation using sEMG and vision fusion
- 1 October 2026
- Published by KEAI Publishing in Biomimetic Intelligence and Robotics
Open AccessHand gesture recognition for humanoid dexterous manipulation using sEMG and vision fusion
- 22 September 2026
- Published by Springer Nature
- p. 113-130
Warp-Free Cross-View Geo-Localization via Feature-Space Consensus Mining
- 22 September 2026
- Published by Springer Nature
- p. 113-130
Warp-Free Cross-View Geo-Localization via Feature-Space Consensus Mining
- 6 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-18
Triple Spectral Fusion for Sensor-based Human Activity Recognition
The field of sensor-based human activity recognition (HAR) mainly uses posture, motion and context data of Inertial Measurement Units (IMUs) to identify daily activities. Despite the advancements in learning-based methods, it is challenging to perform information fusion from the temporal perspective due to the complexities in fusing heterogeneous sensor data and establishing long-term context correlations. This paper proposes a novel triple spectral fusion framework tailored for HAR. First, we develop an adaptive complementary filtering technique for noise suppression and organize each IMU’s sensors into posture and motion modality nodes. Given that IMU nodes form a dynamic heterogeneous graph, we then apply adaptive filtering within the graph Fourier domain to merge both homogeneous and heterogeneous node information. Furthermore, an adaptive wavelet frequency selection approach is implemented to suppress context redundancy and shorten the length of features. This approach enhances both timestamp-based graph aggregation and the correlation of long-term contexts. Our framework uses adaptive filtering in the Fourier, graph Fourier, and wavelet domains, enabling effective multi-sensor fusion and context correlation. Extensive experiments on ten benchmark datasets demonstrate the superior performance of our framework. Project page: https://github.com/crocodilegogogo/TSF-TPAMI2026.
- 6 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-18
Triple Spectral Fusion for Sensor-based Human Activity Recognition
The field of sensor-based human activity recognition (HAR) mainly uses posture, motion and context data of Inertial Measurement Units (IMUs) to identify daily activities. Despite the advancements in learning-based methods, it is challenging to perform information fusion from the temporal perspective due to the complexities in fusing heterogeneous sensor data and establishing long-term context correlations. This paper proposes a novel triple spectral fusion framework tailored for HAR. First, we develop an adaptive complementary filtering technique for noise suppression and organize each IMU’s sensors into posture and motion modality nodes. Given that IMU nodes form a dynamic heterogeneous graph, we then apply adaptive filtering within the graph Fourier domain to merge both homogeneous and heterogeneous node information. Furthermore, an adaptive wavelet frequency selection approach is implemented to suppress context redundancy and shorten the length of features. This approach enhances both timestamp-based graph aggregation and the correlation of long-term contexts. Our framework uses adaptive filtering in the Fourier, graph Fourier, and wavelet domains, enabling effective multi-sensor fusion and context correlation. Extensive experiments on ten benchmark datasets demonstrate the superior performance of our framework. Project page: https://github.com/crocodilegogogo/TSF-TPAMI2026.
- 16 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. PP, 1-15
Heterogeneous Multimodal Multilabel Behavioral Context Recognition Using Wearable Sensors and Mobile Devices
Accurate behavioral context recognition (BCR) is essential for advancing social intelligence in smart home settings and social companion pairing. With advancements in mobile computing technologies, wearable sensors, and mobile devices have become crucial for multimodal data collection to interpret behavioral context. However, existing methods may overlook the multilabel attribute of behavioral context, limiting their ability to model modality-to-label and label-to-label dependencies. Moreover, the unique label heterogeneity in BCR, arising from category-specific differences among labels, is often underestimated. To overcome these challenges, we propose a heterogeneous multimodal multilabel recognition method to capture the relationships between modalities and labels, while taking the label heterogeneity into account. Using specialized heterogeneous decoders with self-adaptive attention and a heterogeneous graph attention network based on the dual-level attention mechanism, our method captures complex dependencies among labels and modalities. Experiments on two benchmark datasets demonstrate the superior performance of our approach, with ablation studies validating the contributions of each component.
- 16 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. PP, 1-15
Heterogeneous Multimodal Multilabel Behavioral Context Recognition Using Wearable Sensors and Mobile Devices
Accurate behavioral context recognition (BCR) is essential for advancing social intelligence in smart home settings and social companion pairing. With advancements in mobile computing technologies, wearable sensors, and mobile devices have become crucial for multimodal data collection to interpret behavioral context. However, existing methods may overlook the multilabel attribute of behavioral context, limiting their ability to model modality-to-label and label-to-label dependencies. Moreover, the unique label heterogeneity in BCR, arising from category-specific differences among labels, is often underestimated. To overcome these challenges, we propose a heterogeneous multimodal multilabel recognition method to capture the relationships between modalities and labels, while taking the label heterogeneity into account. Using specialized heterogeneous decoders with self-adaptive attention and a heterogeneous graph attention network based on the dual-level attention mechanism, our method captures complex dependencies among labels and modalities. Experiments on two benchmark datasets demonstrate the superior performance of our approach, with ablation studies validating the contributions of each component.
- 16 January 2026
- Published by Elsevier in ISPRS Journal of Photogrammetry and Remote Sensing
- Vol. 233, 25-38
RTPSeg: A multi-modality dataset for LiDAR point cloud semantic segmentation assisted with RGB-thermal images in autonomous driving
- 16 January 2026
- Published by Elsevier in ISPRS Journal of Photogrammetry and Remote Sensing
- Vol. 233, 25-38
RTPSeg: A multi-modality dataset for LiDAR point cloud semantic segmentation assisted with RGB-thermal images in autonomous driving
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35, 6730-6743
GeoStyler: A Generalizable Geometry-Aware Diffusion-Based Approach for Direct 3D Gaussian Style Transfer
Direct 3D scene stylization from sparse views remains a significant challenge, as existing optimization-based methods are prohibitively slow and require dense inputs to prevent geometric corruption. While recent direct methods accelerate this process, their rigid decoupling of a static geometry from appearance often leads to visual artifacts, where stylistic textures conflict with and distort the underlying scene structure. To address these limitations, we introduce GeoStyler, a direct framework that generates high-fidelity, multi-view consistent stylized 3D scenes in seconds. Our approach reformulates the conventional pipeline by first leveraging a diffusion model to generate a set of geometrically consistent stylized 2D images. The core of this stage is a novel hybrid query formulation for the self-attention mechanism. Specifically, cross-view geometric information is directly embedded into the query to enforce 3D consistency, while style information is independently injected via the key and value to preserve scene structure. This process is further stabilized by a geometrically-aware latent initialization that provides a coherent starting point for the denoising process. Subsequently, a decoupled reconstruction network lifts these 2D stylized images to 3D Gaussians. A geometry branch predicts a robust 3D scaffold from the original content images, while a parallel style branch predicts the final appearance from our generated stylized images, ensuring structural integrity is not compromised. Extensive experiments on large-scale benchmarks, including RealEstate10K and ACID, demonstrate that GeoStyler significantly outperforms prior arts in stylization quality and multi-view consistency, achieving state-of-the-art performance with a dramatic speedup. Our project page: https://huhuhuxiao. github.io/Geo-Styler/.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35, 6730-6743
GeoStyler: A Generalizable Geometry-Aware Diffusion-Based Approach for Direct 3D Gaussian Style Transfer
Direct 3D scene stylization from sparse views remains a significant challenge, as existing optimization-based methods are prohibitively slow and require dense inputs to prevent geometric corruption. While recent direct methods accelerate this process, their rigid decoupling of a static geometry from appearance often leads to visual artifacts, where stylistic textures conflict with and distort the underlying scene structure. To address these limitations, we introduce GeoStyler, a direct framework that generates high-fidelity, multi-view consistent stylized 3D scenes in seconds. Our approach reformulates the conventional pipeline by first leveraging a diffusion model to generate a set of geometrically consistent stylized 2D images. The core of this stage is a novel hybrid query formulation for the self-attention mechanism. Specifically, cross-view geometric information is directly embedded into the query to enforce 3D consistency, while style information is independently injected via the key and value to preserve scene structure. This process is further stabilized by a geometrically-aware latent initialization that provides a coherent starting point for the denoising process. Subsequently, a decoupled reconstruction network lifts these 2D stylized images to 3D Gaussians. A geometry branch predicts a robust 3D scaffold from the original content images, while a parallel style branch predicts the final appearance from our generated stylized images, ensuring structural integrity is not compromised. Extensive experiments on large-scale benchmarks, including RealEstate10K and ACID, demonstrate that GeoStyler significantly outperforms prior arts in stylization quality and multi-view consistency, achieving state-of-the-art performance with a dramatic speedup. Our project page: https://huhuhuxiao. github.io/Geo-Styler/.
- Open Access
- 6 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-15
Adaptive Spectral Denoising Network for Efficient Maneuvering Target Tracking
Maneuvering target tracking remains a critical yet challenging task, due to the unpredictable and varying motion patterns of targets. Although learning-based methods have shown promise by directly learning the non-linear mappings from noisy observations to target states by integrating both long- and short-range dependencies, existing advanced Transformer-based models struggle with robustness to noise, computational efficiency, and capturing fine-grained temporal dependencies. To address these, we propose the Adaptive Spectral Denoising Network (ASDN), comprising an Adaptive Spectral Denoising Filter (ASDF) and a Temporal Local-pattern Guided Block (TLGB). The ASDF processes signals in the frequency domain through Fourier Transform to enhance feature representation by simultaneously capturing long- and short-range dependencies. This is achieved through adaptive spectral masking that suppresses noise-corrupted frequency components while preserving critical dependencies. To compensate for potential information loss caused by such spectral masking, the TLGB explicitly models short-term temporal interactions to preserve local dynamics critical for accurate tracking. Furthermore, we introduce LASTv2, a large-scale dataset of maneuvering target trajectories characterized by diverse and complex motion patterns, specifically designed to simulate real-world adversarial scenarios with high mobility and stochastic dynamics. Experimental results demonstrate that our method achieves state-of-the-art performance, reducing root-mean-square-error by 76.9% and improving inference speed by two orders of magnitude compared to the baseline methods.
- 6 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-15
Open AccessAdaptive Spectral Denoising Network for Efficient Maneuvering Target Tracking
Maneuvering target tracking remains a critical yet challenging task, due to the unpredictable and varying motion patterns of targets. Although learning-based methods have shown promise by directly learning the non-linear mappings from noisy observations to target states by integrating both long- and short-range dependencies, existing advanced Transformer-based models struggle with robustness to noise, computational efficiency, and capturing fine-grained temporal dependencies. To address these, we propose the Adaptive Spectral Denoising Network (ASDN), comprising an Adaptive Spectral Denoising Filter (ASDF) and a Temporal Local-pattern Guided Block (TLGB). The ASDF processes signals in the frequency domain through Fourier Transform to enhance feature representation by simultaneously capturing long- and short-range dependencies. This is achieved through adaptive spectral masking that suppresses noise-corrupted frequency components while preserving critical dependencies. To compensate for potential information loss caused by such spectral masking, the TLGB explicitly models short-term temporal interactions to preserve local dynamics critical for accurate tracking. Furthermore, we introduce LASTv2, a large-scale dataset of maneuvering target trajectories characterized by diverse and complex motion patterns, specifically designed to simulate real-world adversarial scenarios with high mobility and stochastic dynamics. Experimental results demonstrate that our method achieves state-of-the-art performance, reducing root-mean-square-error by 76.9% and improving inference speed by two orders of magnitude compared to the baseline methods.
- 10 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE MultiMedia
- Vol. 33 (1) , 36-46
DRT: Dilated Recurrent Transformer for Event-Based Video Reconstruction
Event-based video reconstruction has emerged as an essential task since the advent of event cameras. The temporal information plays a critical role in eventbased video reconstruction. However, existing recurrent convolutional models commonly suffer inferior performance since the limited receptive field of convolutional layers cannot capture long-range spatial dependency effectively. Some models introduce transformer blocks to improve the spatial receptive field, yet usually overlook the exploitation of long-range dependency in the temporal dimension. To address this issue, we propose a hybrid CNN-Transformer model called dilated recurrent transformer (DRT) for event-based video reconstruction. Specifically, a dilated event feature recurrent attention (DEFRA) block is designed to aggregate information from two distant timestamps. To facilitate the network to reconstruct better-quality frames, we introduce video perception loss (VIPLoss) to further exploit the temporal consistency in ground truth videos. Extensive experiments conducted on benchmark datasets demonstrate the effectiveness of the proposed DRT and VIPLoss.
- 10 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE MultiMedia
- Vol. 33 (1) , 36-46
DRT: Dilated Recurrent Transformer for Event-Based Video Reconstruction
Event-based video reconstruction has emerged as an essential task since the advent of event cameras. The temporal information plays a critical role in eventbased video reconstruction. However, existing recurrent convolutional models commonly suffer inferior performance since the limited receptive field of convolutional layers cannot capture long-range spatial dependency effectively. Some models introduce transformer blocks to improve the spatial receptive field, yet usually overlook the exploitation of long-range dependency in the temporal dimension. To address this issue, we propose a hybrid CNN-Transformer model called dilated recurrent transformer (DRT) for event-based video reconstruction. Specifically, a dilated event feature recurrent attention (DEFRA) block is designed to aggregate information from two distant timestamps. To facilitate the network to reconstruct better-quality frames, we introduce video perception loss (VIPLoss) to further exploit the temporal consistency in ground truth videos. Extensive experiments conducted on benchmark datasets demonstrate the effectiveness of the proposed DRT and VIPLoss.