Selected work

Representative Papers

Improving Viewpoint-Invariance and Temporal Consistency for Action Detection

May 21, 2026

This work addresses the limitations of existing action detection methods, which are often constrained by single-view training data and struggle to model fine-grained temporal dependencies. To overcome these challenges, the authors propose a two-stage action detection framework. In the first stage, virtual viewpoint augmentation is employed during training to extract viewpoint-invariant motion features. The second stage introduces a multi-scale temporal encoder based on a selective state space model, effectively integrating information across multiple viewpoints and temporal scales. This approach represents the first integration of virtual viewpoint augmentation with selective state space modeling for sequence representation. It achieves consistent and significant improvements over state-of-the-art methods across all splits of the PKU-MMD and BABEL benchmarks, while simultaneously enhancing viewpoint robustness and global temporal consistency.

0 citationsRead paper

Global Symmetry and Orthogonal Transformations from Geometrical Moment $n$-tuples

Feb 08, 2026

This work proposes a unified framework based on geometric moment n-tuples for efficiently detecting object symmetries and estimating associated orthogonal transformations, such as rotations and reflections. The method constructs a non-iterative, closed-form solution in arbitrary-dimensional spaces, enabling accurate identification of global symmetries and their corresponding orthogonal transformations without relying on optimization procedures. Its core innovation lies in a novel discriminative metric derived from geometric moments, which significantly enhances both computational efficiency and robustness. Experimental results demonstrate the approach’s effectiveness on both 2D and 3D objects; when integrated with existing iterative strategies, it substantially increases the number of detectable symmetry planes while accelerating computation.

0 citationsRead paper

The Online Patch Redundancy Eliminator (OPRE): A novel approach to online agnostic continual learning using dataset compression

Nov 11, 2025

Addressing the dual challenges of catastrophic forgetting and reliance on prior knowledge in continual learning, this paper proposes OPRE, a prior-free online continual learning framework. Its core is an online patch redundancy elimination compression algorithm, which operates under minimal data assumptions—such as local smoothness—and for the first time systematically reveals the fundamental limitations of pretrained feature extractors on continual learning generalizability. OPRE dispenses with task identifiers, replay buffers, and domain priors; instead, it dynamically compresses redundant information upon data stream arrival and enables plug-and-play learning via test-time classifier adaptation. Evaluated on standard CIFAR-10/100 benchmarks, OPRE consistently outperforms mainstream replay-based, regularization-based, and parameter-isolation methods. These results validate the effectiveness and scalability of the “prior-free + online compression” paradigm for continual learning.

0 citationsRead paper

Experimental Study of Magnetic Near-Field Microstrip Electronic Probe for PCB EMC Emission Measurement

Sep 29, 2025

Accurate electromagnetic compatibility (EMC) radiated emission assessment of printed circuit boards (PCBs) remains challenging for 6G wireless systems operating at high frequencies. Method: This work proposes a microstrip-based magnetic near-field probe design and calibration methodology, enabling a near-field scanning system covering 0.1–3 GHz. The system is validated per IEC 61967-1 and extended to non-standard transmission line structures. Integration of HFSS simulations, broadband measurements, and precision calibration achieves 1-mm spatial resolution in magnetic near-field imaging. Contribution/Results: Near-field distribution maps of representative DUTs are acquired at 2 GHz and 3 GHz, showing excellent agreement between measurement and simulation. This study represents the first systematic application of microstrip probe technology to PCB-level high-frequency EMC near-field diagnostics, significantly improving radiation source localization accuracy and measurement repeatability. It provides a novel tool for pre-compliance EMC evaluation of 6G devices.

0 citationsRead paper

Point Cloud Quality Assessment Using the Perceptual Clustering Weighted Graph (PCW-Graph) and Attention Fusion Network

Jun 04, 2025

To address the challenge of quantifying 3D point cloud distortion in real-world scenarios—where pristine reference models are unavailable and distortion is difficult to measure—this paper proposes a novel no-reference point cloud quality assessment (NR-PCQA) method. The core innovation lies in constructing a perceptual clustering-weighted graph (PCW-Graph) that explicitly models local geometric and semantic inconsistencies, coupled with an attention-based fusion network for adaptive, multi-scale feature aggregation. The method integrates graph neural networks, self-attention mechanisms, and perception-driven graph construction. Evaluated on multiple standard benchmarks, it achieves state-of-the-art performance, demonstrating significantly improved discrimination capability against geometric noise, compression artifacts, and sampling distortions. Specifically, it attains an average improvement of 12.7% in Pearson linear correlation coefficient (PLCC) over prior approaches.

0 citationsRead paper

Recent publications

Latest Papers

Improving Viewpoint-Invariance and Temporal Consistency for Action Detection

May 21, 2026

This work addresses the limitations of existing action detection methods, which are often constrained by single-view training data and struggle to model fine-grained temporal dependencies. To overcome these challenges, the authors propose a two-stage action detection framework. In the first stage, virtual viewpoint augmentation is employed during training to extract viewpoint-invariant motion features. The second stage introduces a multi-scale temporal encoder based on a selective state space model, effectively integrating information across multiple viewpoints and temporal scales. This approach represents the first integration of virtual viewpoint augmentation with selective state space modeling for sequence representation. It achieves consistent and significant improvements over state-of-the-art methods across all splits of the PKU-MMD and BABEL benchmarks, while simultaneously enhancing viewpoint robustness and global temporal consistency.

0 citationsRead paper

Global Symmetry and Orthogonal Transformations from Geometrical Moment $n$-tuples

Feb 08, 2026

This work proposes a unified framework based on geometric moment n-tuples for efficiently detecting object symmetries and estimating associated orthogonal transformations, such as rotations and reflections. The method constructs a non-iterative, closed-form solution in arbitrary-dimensional spaces, enabling accurate identification of global symmetries and their corresponding orthogonal transformations without relying on optimization procedures. Its core innovation lies in a novel discriminative metric derived from geometric moments, which significantly enhances both computational efficiency and robustness. Experimental results demonstrate the approach’s effectiveness on both 2D and 3D objects; when integrated with existing iterative strategies, it substantially increases the number of detectable symmetry planes while accelerating computation.

0 citationsRead paper

The Online Patch Redundancy Eliminator (OPRE): A novel approach to online agnostic continual learning using dataset compression

Nov 11, 2025

Addressing the dual challenges of catastrophic forgetting and reliance on prior knowledge in continual learning, this paper proposes OPRE, a prior-free online continual learning framework. Its core is an online patch redundancy elimination compression algorithm, which operates under minimal data assumptions—such as local smoothness—and for the first time systematically reveals the fundamental limitations of pretrained feature extractors on continual learning generalizability. OPRE dispenses with task identifiers, replay buffers, and domain priors; instead, it dynamically compresses redundant information upon data stream arrival and enables plug-and-play learning via test-time classifier adaptation. Evaluated on standard CIFAR-10/100 benchmarks, OPRE consistently outperforms mainstream replay-based, regularization-based, and parameter-isolation methods. These results validate the effectiveness and scalability of the “prior-free + online compression” paradigm for continual learning.

0 citationsRead paper

Experimental Study of Magnetic Near-Field Microstrip Electronic Probe for PCB EMC Emission Measurement

Sep 29, 2025

Accurate electromagnetic compatibility (EMC) radiated emission assessment of printed circuit boards (PCBs) remains challenging for 6G wireless systems operating at high frequencies. Method: This work proposes a microstrip-based magnetic near-field probe design and calibration methodology, enabling a near-field scanning system covering 0.1–3 GHz. The system is validated per IEC 61967-1 and extended to non-standard transmission line structures. Integration of HFSS simulations, broadband measurements, and precision calibration achieves 1-mm spatial resolution in magnetic near-field imaging. Contribution/Results: Near-field distribution maps of representative DUTs are acquired at 2 GHz and 3 GHz, showing excellent agreement between measurement and simulation. This study represents the first systematic application of microstrip probe technology to PCB-level high-frequency EMC near-field diagnostics, significantly improving radiation source localization accuracy and measurement repeatability. It provides a novel tool for pre-compliance EMC evaluation of 6G devices.

0 citationsRead paper

Point Cloud Quality Assessment Using the Perceptual Clustering Weighted Graph (PCW-Graph) and Attention Fusion Network

Jun 04, 2025

To address the challenge of quantifying 3D point cloud distortion in real-world scenarios—where pristine reference models are unavailable and distortion is difficult to measure—this paper proposes a novel no-reference point cloud quality assessment (NR-PCQA) method. The core innovation lies in constructing a perceptual clustering-weighted graph (PCW-Graph) that explicitly models local geometric and semantic inconsistencies, coupled with an attention-based fusion network for adaptive, multi-scale feature aggregation. The method integrates graph neural networks, self-attention mechanisms, and perception-driven graph construction. Evaluated on multiple standard benchmarks, it achieves state-of-the-art performance, demonstrating significantly improved discrimination capability against geometric noise, compression artifacts, and sampling distortions. Specifically, it attains an average improvement of 12.7% in Pearson linear correlation coefficient (PLCC) over prior approaches.

0 citationsRead paper