Selected work

Representative Papers

It's All Just Vectorization: einx, a Universal Notation for Tensor Operations

Jul 30, 2026

This work addresses the verbosity and error-proneness of NumPy-style APIs in existing tensor frameworks, as well as the limited generality of alternatives like einsum. The authors propose einx, a universal, declarative notation for tensor operations grounded in vectorization. By unifying diverse tensor computations through loop-like pointwise expressions, einx leverages vectorization as its core abstraction, combining declarative element-wise specifications with dimension lifting techniques. This approach reconstructs the entire tensor API using only a small set of primitive operations. The resulting einx library is compatible with major deep learning frameworks and integrates seamlessly into Python, substantially improving code readability and consistency while reducing cognitive load and shape-related errors.

0 citationsRead paper

Augmentation techniques for video surveillance in the visible and thermal spectral range

Jun 11, 2026

This work addresses the limited generalization of existing deep models in multispectral (visible and thermal infrared) video surveillance, which stems from sensor discrepancies and the scarcity of thermal imaging data. The authors propose a CNN-based framework for multispectral object detection and systematically design and evaluate several cross-spectral data augmentation strategies—namely thermal feature simulation, texture-preserving transformations, and illumination-invariant enhancement—to effectively integrate color, shape, and thermal radiation cues. Experimental results demonstrate that the proposed approach significantly improves detection accuracy and robustness in mixed-spectral scenarios, validates the auxiliary value of visible-spectrum data for thermal infrared detection, and fills a critical gap in the understanding of cross-spectral data augmentation mechanisms.

0 citationsRead paper

A Causally Grounded Taxonomy for Image Degradation Robustness Evaluation

May 15, 2026

Image degradation is pervasive throughout the imaging pipeline, yet existing research lacks a unified taxonomy and evaluation protocol, hindering cross-dataset and cross-task comparisons. This work introduces a causal perspective to address this gap, proposing a dual-axis classification framework: one axis categorizes degradations by their dominant causal source in the imaging pipeline—encompassing environment, sensor/optics, ISP/codec, and transmission systems—while the other characterizes their perceptual effects, augmented with a lightweight severity quantification layer. Built upon this framework, the COCO Degradation benchmark leverages PSNR, SSIM, and LPIPS to uniformly measure degradation intensity across physical artifacts, algorithmic perturbations, and perceptual distortions, substantially enhancing the evaluation of object detection model robustness under diverse imaging conditions.

0 citationsRead paper

SuperADD: Training-free Class-agnostic Anomaly Segmentation -- CVPR 2026 VAND 4.0 Workshop Challenge Industrial Track

May 14, 2026

This work addresses the challenge of distribution shifts in industrial visual anomaly detection caused by variations in acquisition conditions such as illumination. The authors propose a training-free, category-agnostic method for robust anomaly segmentation, building upon SuperAD with a unified architecture and shared hyperparameters across all object categories to eliminate class-specific design. Key innovations include the adoption of a DINOv3 backbone, an overlapping patch partitioning strategy, intensity augmentation, an improved memory bank sampling mechanism, and iterative morphological closing to produce spatially coherent anomaly maps. Evaluated on the MVTec AD 2 dataset, the method achieves F1 scores of 62.61%, 57.42%, and 54.35% on public test, private test, and mixed private test splits, respectively, substantially outperforming existing approaches and demonstrating strong suitability for rapid deployment in industrial settings.

0 citationsRead paper

Evaluating Explainability in Safety-Critical ATR Systems: Limitations of Post-Hoc Methods and Paths Toward Robust XAI

May 07, 2026

This work addresses the critical need for reliable, verifiable, and robust explainability in safety-critical automatic target recognition (ATR) systems, which demand more than high predictive performance. The authors propose a comprehensive XAI evaluation framework tailored to safety-critical contexts, systematically assessing mainstream post-hoc explanation methods—including saliency maps, attention mechanisms, and surrogate models—along four dimensions: interpretability, robustness, resistance to manipulation, and suitability for formal verification. Their analysis reveals that current approaches commonly suffer from spurious explanations, instability under perturbations, and a tendency to induce unwarranted user trust, rendering them inadequate for ATR assurance requirements. In response, the paper advocates a paradigm shift toward causally grounded, physics-informed explainability, laying the theoretical and technical foundation for next-generation XAI capable of supporting system-level safety guarantees.

0 citationsRead paper

Recent publications

Latest Papers

It's All Just Vectorization: einx, a Universal Notation for Tensor Operations

Jul 30, 2026

This work addresses the verbosity and error-proneness of NumPy-style APIs in existing tensor frameworks, as well as the limited generality of alternatives like einsum. The authors propose einx, a universal, declarative notation for tensor operations grounded in vectorization. By unifying diverse tensor computations through loop-like pointwise expressions, einx leverages vectorization as its core abstraction, combining declarative element-wise specifications with dimension lifting techniques. This approach reconstructs the entire tensor API using only a small set of primitive operations. The resulting einx library is compatible with major deep learning frameworks and integrates seamlessly into Python, substantially improving code readability and consistency while reducing cognitive load and shape-related errors.

0 citationsRead paper

Augmentation techniques for video surveillance in the visible and thermal spectral range

Jun 11, 2026

This work addresses the limited generalization of existing deep models in multispectral (visible and thermal infrared) video surveillance, which stems from sensor discrepancies and the scarcity of thermal imaging data. The authors propose a CNN-based framework for multispectral object detection and systematically design and evaluate several cross-spectral data augmentation strategies—namely thermal feature simulation, texture-preserving transformations, and illumination-invariant enhancement—to effectively integrate color, shape, and thermal radiation cues. Experimental results demonstrate that the proposed approach significantly improves detection accuracy and robustness in mixed-spectral scenarios, validates the auxiliary value of visible-spectrum data for thermal infrared detection, and fills a critical gap in the understanding of cross-spectral data augmentation mechanisms.

0 citationsRead paper

A Causally Grounded Taxonomy for Image Degradation Robustness Evaluation

May 15, 2026

Image degradation is pervasive throughout the imaging pipeline, yet existing research lacks a unified taxonomy and evaluation protocol, hindering cross-dataset and cross-task comparisons. This work introduces a causal perspective to address this gap, proposing a dual-axis classification framework: one axis categorizes degradations by their dominant causal source in the imaging pipeline—encompassing environment, sensor/optics, ISP/codec, and transmission systems—while the other characterizes their perceptual effects, augmented with a lightweight severity quantification layer. Built upon this framework, the COCO Degradation benchmark leverages PSNR, SSIM, and LPIPS to uniformly measure degradation intensity across physical artifacts, algorithmic perturbations, and perceptual distortions, substantially enhancing the evaluation of object detection model robustness under diverse imaging conditions.

0 citationsRead paper

SuperADD: Training-free Class-agnostic Anomaly Segmentation -- CVPR 2026 VAND 4.0 Workshop Challenge Industrial Track

May 14, 2026

This work addresses the challenge of distribution shifts in industrial visual anomaly detection caused by variations in acquisition conditions such as illumination. The authors propose a training-free, category-agnostic method for robust anomaly segmentation, building upon SuperAD with a unified architecture and shared hyperparameters across all object categories to eliminate class-specific design. Key innovations include the adoption of a DINOv3 backbone, an overlapping patch partitioning strategy, intensity augmentation, an improved memory bank sampling mechanism, and iterative morphological closing to produce spatially coherent anomaly maps. Evaluated on the MVTec AD 2 dataset, the method achieves F1 scores of 62.61%, 57.42%, and 54.35% on public test, private test, and mixed private test splits, respectively, substantially outperforming existing approaches and demonstrating strong suitability for rapid deployment in industrial settings.

0 citationsRead paper

Evaluating Explainability in Safety-Critical ATR Systems: Limitations of Post-Hoc Methods and Paths Toward Robust XAI

May 07, 2026

This work addresses the critical need for reliable, verifiable, and robust explainability in safety-critical automatic target recognition (ATR) systems, which demand more than high predictive performance. The authors propose a comprehensive XAI evaluation framework tailored to safety-critical contexts, systematically assessing mainstream post-hoc explanation methods—including saliency maps, attention mechanisms, and surrogate models—along four dimensions: interpretability, robustness, resistance to manipulation, and suitability for formal verification. Their analysis reveals that current approaches commonly suffer from spurious explanations, instability under perturbations, and a tendency to induce unwarranted user trust, rendering them inadequate for ATR assurance requirements. In response, the paper advocates a paradigm shift toward causally grounded, physics-informed explainability, laying the theoretical and technical foundation for next-generation XAI capable of supporting system-level safety guarantees.

0 citationsRead paper