Selected work

Representative Papers

VT-3DAD: Cross-Category 3D Anomaly Detection via Visual-Text Normal Space Alignment

Jun 02, 2026

This work addresses the challenge of cross-category 3D point cloud anomaly detection using only a few normal samples by proposing a training-free framework that, for the first time, introduces vision–text normal space alignment to this task. The method extracts visual features from multi-view depth maps using a frozen CLIP model and constructs semantic normal anchors via depth-aware and 3D-aware textual prompts. Anomaly scores are computed by fusing visual and semantic deviations. Evaluated on ShapeNetPart, the approach achieves an average single-sample AUC-ROC of 94.80%, outperforming a purely visual baseline by 2.31% and reducing the standard deviation from 5.64 to 3.41. These results demonstrate the effectiveness of the proposed method in jointly modeling geometric and semantic normality, as well as its strong cross-category generalization capability.

0 citationsRead paper

DMP-3DAD: Cross-Category 3D Anomaly Detection via Realistic Depth Map Projection with Few Normal Samples

Feb 11, 2026

This work addresses the challenge of cross-category 3D point cloud anomaly detection using only a few normal samples by proposing the first training-free, general-purpose framework. The method projects 3D point clouds into multi-view realistic depth maps and leverages a frozen CLIP vision encoder to extract features, enabling anomaly identification through weighted feature similarity without any category-specific fine-tuning or adaptation. Experimental results demonstrate that the approach achieves state-of-the-art performance under few-shot settings on the ShapeNetPart dataset, significantly enhancing the generality, practicality, and robustness of cross-category 3D anomaly detection.

0 citationsRead paper

Deep Exploration of Epoch-wise Double Descent in Noisy Data: Signal Separation, Large Activation, and Benign Overfitting

Jan 13, 2026

This study investigates the epoch-wise double descent phenomenon in deep neural networks trained on noisy data and its intrinsic connections to benign overfitting and large activation mechanisms. By training fully connected networks of varying sizes on CIFAR-10 with 30% label noise, the authors integrate loss decomposition, activation pattern tracking, and layer-wise representation analysis into a unified empirical framework, for the first time linking epoch-wise double descent, benign overfitting, and large activations in shallow layers. The findings reveal that models can achieve strong generalization even after fully fitting noisy labels; activations corresponding to noisy and clean samples gradually separate in deeper layers; and large activations emerge in shallow layers—correlated with input structure—whose evolution aligns precisely with the re-generalization process.

0 citationsRead paper

3DKeyAD: High-Resolution 3D Point Cloud Anomaly Detection via Keypoint-Guided Point Clustering

Jul 17, 2025

To address the challenges of high computational cost, sensitivity to spatial misalignment, and difficulty in modeling local structural discrepancies in fine-grained anomaly detection for high-resolution 3D point clouds, this paper proposes a keypoint-guided clustering and multi-prototype alignment framework. Methodologically, it first identifies robust clustering centers via geometric saliency-driven keypoint detection; then achieves cross-sample local region alignment through point cloud registration and enhances anomaly sensitivity via multi-prototype feature representation; finally performs fine-grained feature discrepancy analysis at the cluster level for precise anomaly localization. The method operates solely on raw point cloud features, requiring no additional supervision or reconstruction modules. On the Real3D-AD benchmark, it achieves state-of-the-art performance in both object-level and point-level anomaly detection, demonstrating superior efficiency and robustness.

0 citationsRead paper

Autoencoder-Based Detection of Anomalous Stokes V Spectra in the Flare-Producing Active Region 13663 Using Hinode/SP Observations

Apr 08, 2025

Hinode/SP Stokes V spectra suffer from low signal-to-noise ratio, spatially localized anomalies, and insufficient physical modeling, hindering reliable detection of subtle magnetic disturbances. Method: This paper proposes an unsupervised anomaly detection framework based on autoencoders—the first deep learning approach applied to solar polarimetric spectral anomaly identification. It learns typical spectral line profiles end-to-end, enabling automatic localization of weak deviations without prior physical assumptions or manual labeling. The method integrates spectro-polarimetric analysis with co-aligned magnetogram spatial registration for sub-pixel anomaly localization. Results: Applied to active region AR 13663, the method successfully identified pre-flare Stokes V anomalies concentrated near the magnetic polarity inversion line hours before the X1.3 flare on 5 May 2024. It achieves significantly higher detection accuracy and spatial resolution than conventional manual identification techniques.

0 citationsRead paper

Recent publications

Latest Papers

VT-3DAD: Cross-Category 3D Anomaly Detection via Visual-Text Normal Space Alignment

Jun 02, 2026

This work addresses the challenge of cross-category 3D point cloud anomaly detection using only a few normal samples by proposing a training-free framework that, for the first time, introduces vision–text normal space alignment to this task. The method extracts visual features from multi-view depth maps using a frozen CLIP model and constructs semantic normal anchors via depth-aware and 3D-aware textual prompts. Anomaly scores are computed by fusing visual and semantic deviations. Evaluated on ShapeNetPart, the approach achieves an average single-sample AUC-ROC of 94.80%, outperforming a purely visual baseline by 2.31% and reducing the standard deviation from 5.64 to 3.41. These results demonstrate the effectiveness of the proposed method in jointly modeling geometric and semantic normality, as well as its strong cross-category generalization capability.

0 citationsRead paper

DMP-3DAD: Cross-Category 3D Anomaly Detection via Realistic Depth Map Projection with Few Normal Samples

Feb 11, 2026

This work addresses the challenge of cross-category 3D point cloud anomaly detection using only a few normal samples by proposing the first training-free, general-purpose framework. The method projects 3D point clouds into multi-view realistic depth maps and leverages a frozen CLIP vision encoder to extract features, enabling anomaly identification through weighted feature similarity without any category-specific fine-tuning or adaptation. Experimental results demonstrate that the approach achieves state-of-the-art performance under few-shot settings on the ShapeNetPart dataset, significantly enhancing the generality, practicality, and robustness of cross-category 3D anomaly detection.

0 citationsRead paper

Deep Exploration of Epoch-wise Double Descent in Noisy Data: Signal Separation, Large Activation, and Benign Overfitting

Jan 13, 2026

This study investigates the epoch-wise double descent phenomenon in deep neural networks trained on noisy data and its intrinsic connections to benign overfitting and large activation mechanisms. By training fully connected networks of varying sizes on CIFAR-10 with 30% label noise, the authors integrate loss decomposition, activation pattern tracking, and layer-wise representation analysis into a unified empirical framework, for the first time linking epoch-wise double descent, benign overfitting, and large activations in shallow layers. The findings reveal that models can achieve strong generalization even after fully fitting noisy labels; activations corresponding to noisy and clean samples gradually separate in deeper layers; and large activations emerge in shallow layers—correlated with input structure—whose evolution aligns precisely with the re-generalization process.

0 citationsRead paper

3DKeyAD: High-Resolution 3D Point Cloud Anomaly Detection via Keypoint-Guided Point Clustering

Jul 17, 2025

To address the challenges of high computational cost, sensitivity to spatial misalignment, and difficulty in modeling local structural discrepancies in fine-grained anomaly detection for high-resolution 3D point clouds, this paper proposes a keypoint-guided clustering and multi-prototype alignment framework. Methodologically, it first identifies robust clustering centers via geometric saliency-driven keypoint detection; then achieves cross-sample local region alignment through point cloud registration and enhances anomaly sensitivity via multi-prototype feature representation; finally performs fine-grained feature discrepancy analysis at the cluster level for precise anomaly localization. The method operates solely on raw point cloud features, requiring no additional supervision or reconstruction modules. On the Real3D-AD benchmark, it achieves state-of-the-art performance in both object-level and point-level anomaly detection, demonstrating superior efficiency and robustness.

0 citationsRead paper

Autoencoder-Based Detection of Anomalous Stokes V Spectra in the Flare-Producing Active Region 13663 Using Hinode/SP Observations

Apr 08, 2025

Hinode/SP Stokes V spectra suffer from low signal-to-noise ratio, spatially localized anomalies, and insufficient physical modeling, hindering reliable detection of subtle magnetic disturbances. Method: This paper proposes an unsupervised anomaly detection framework based on autoencoders—the first deep learning approach applied to solar polarimetric spectral anomaly identification. It learns typical spectral line profiles end-to-end, enabling automatic localization of weak deviations without prior physical assumptions or manual labeling. The method integrates spectro-polarimetric analysis with co-aligned magnetogram spatial registration for sub-pixel anomaly localization. Results: Applied to active region AR 13663, the method successfully identified pre-flare Stokes V anomalies concentrated near the magnetic polarity inversion line hours before the X1.3 flare on 5 May 2024. It achieves significantly higher detection accuracy and spatial resolution than conventional manual identification techniques.

0 citationsRead paper