BMCTrack-d: Pig re-identification and tracking via back marks in challenging camera settings
本文提出BMCTrack-d方法，通过利用背部标记在复杂侧视摄像头环境下实现猪只再识别与跟踪，解决个体监控难题。
0 citationsRead paper
Selected work
Representative Papers
When One Modality Is Not Enough: Multimodal Sex and Life-Stage Classification of Red Deer from Aerial RGB-Thermal Video
This study addresses the challenge of accurately classifying red deer by sex and life stage in aerial surveys, where single-modality imagery—either RGB or thermal—is hindered by occlusion, seasonal antler variation, and low resolution. The work presents the first full-stage fusion of RGB and thermal modalities in wildlife aerial census, leveraging self-supervised DINOv3 features for multi-stage modality alignment. It further introduces modality consistency verification and georeferenced body-size calibration, integrated with object tracking and cross-frame voting to jointly infer species, sex, and life stage. Evaluated across four flight campaigns, the method correctly classified 25 out of 26 red deer (96.2%), significantly outperforming single-modality approaches (76.9%) and demonstrating exceptional robustness in sex identification across multiple seasons and complex environmental conditions.
0 citationsRead paper
Continuous Face Authentication on Mobile and Desktop Platforms: A Comparative Study
This study addresses the limitation of traditional device authentication, which verifies identity only at unlock and fails to prevent unauthorized use after the legitimate user departs. Building upon the InsightFace framework, the authors introduce a time-based trust decay mechanism and present the first systematic comparison of continuous facial authentication performance across mobile and desktop platforms. The evaluation encompasses diverse device types, head poses, lighting conditions, and real-world usage scenarios. Through extensive video data collection and analysis under varied conditions, the study finds that device type has minimal impact on authentication performance; instead, increased false rejection rates in practice are primarily attributed to reduced facial visibility caused by occlusions or the user moving out of the camera’s field of view. These findings highlight a key challenge for deploying continuous authentication in real-world settings.
0 citationsRead paper
Taming the Search Space: Solving and Generating Hitori and Binairo Puzzles
This study addresses the challenge of the vast search space in solving and generating Hitori and Binairo logic puzzles by systematically comparing an optimized backtracking approach—integrating constraint propagation and heuristic variable ordering—with SAT-based methods. The work introduces the first unified framework for generating uniquely solvable instances of both puzzle types, enabling systematic benchmarking. This framework combines iterative connectivity checks with CNF encoding techniques. Experimental results demonstrate that constraint propagation substantially reduces search tree size; SAT solvers outperform other methods on Binairo, whereas the optimized backtracking algorithm achieves superior performance on Hitori. The proposed approaches provide efficient solutions for both puzzle generation and solving, while establishing a reproducible evaluation benchmark for future research.
0 citationsRead paper
Navigating the Sea of LLM Evaluation: Investigating Bias in Toxicity Benchmarks
Current benchmarks for evaluating toxicity in large language models exhibit underappreciated systematic biases that may lead to the deployment of unsafe models. This work systematically investigates how variations in task formulation—such as text completion versus summarization—input data domains, and evaluated models interact with multiple toxicity metrics. It reveals, for the first time, that both task type and data domain significantly influence toxicity scores. Experiments demonstrate that existing benchmarks are prone to misclassifying content as harmful when tasks are altered and show inconsistent performance across domains, highlighting their fragility and dependence on specific model-task configurations. These findings underscore the urgent need for more robust and reliable toxicity evaluation frameworks.
0 citationsRead paper
Recent publications
Latest Papers
BMCTrack-d: Pig re-identification and tracking via back marks in challenging camera settings
本文提出BMCTrack-d方法，通过利用背部标记在复杂侧视摄像头环境下实现猪只再识别与跟踪，解决个体监控难题。
0 citationsRead paper
When One Modality Is Not Enough: Multimodal Sex and Life-Stage Classification of Red Deer from Aerial RGB-Thermal Video
This study addresses the challenge of accurately classifying red deer by sex and life stage in aerial surveys, where single-modality imagery—either RGB or thermal—is hindered by occlusion, seasonal antler variation, and low resolution. The work presents the first full-stage fusion of RGB and thermal modalities in wildlife aerial census, leveraging self-supervised DINOv3 features for multi-stage modality alignment. It further introduces modality consistency verification and georeferenced body-size calibration, integrated with object tracking and cross-frame voting to jointly infer species, sex, and life stage. Evaluated across four flight campaigns, the method correctly classified 25 out of 26 red deer (96.2%), significantly outperforming single-modality approaches (76.9%) and demonstrating exceptional robustness in sex identification across multiple seasons and complex environmental conditions.
0 citationsRead paper
Continuous Face Authentication on Mobile and Desktop Platforms: A Comparative Study
This study addresses the limitation of traditional device authentication, which verifies identity only at unlock and fails to prevent unauthorized use after the legitimate user departs. Building upon the InsightFace framework, the authors introduce a time-based trust decay mechanism and present the first systematic comparison of continuous facial authentication performance across mobile and desktop platforms. The evaluation encompasses diverse device types, head poses, lighting conditions, and real-world usage scenarios. Through extensive video data collection and analysis under varied conditions, the study finds that device type has minimal impact on authentication performance; instead, increased false rejection rates in practice are primarily attributed to reduced facial visibility caused by occlusions or the user moving out of the camera’s field of view. These findings highlight a key challenge for deploying continuous authentication in real-world settings.
0 citationsRead paper
Taming the Search Space: Solving and Generating Hitori and Binairo Puzzles
This study addresses the challenge of the vast search space in solving and generating Hitori and Binairo logic puzzles by systematically comparing an optimized backtracking approach—integrating constraint propagation and heuristic variable ordering—with SAT-based methods. The work introduces the first unified framework for generating uniquely solvable instances of both puzzle types, enabling systematic benchmarking. This framework combines iterative connectivity checks with CNF encoding techniques. Experimental results demonstrate that constraint propagation substantially reduces search tree size; SAT solvers outperform other methods on Binairo, whereas the optimized backtracking algorithm achieves superior performance on Hitori. The proposed approaches provide efficient solutions for both puzzle generation and solving, while establishing a reproducible evaluation benchmark for future research.
0 citationsRead paper
Navigating the Sea of LLM Evaluation: Investigating Bias in Toxicity Benchmarks
Current benchmarks for evaluating toxicity in large language models exhibit underappreciated systematic biases that may lead to the deployment of unsafe models. This work systematically investigates how variations in task formulation—such as text completion versus summarization—input data domains, and evaluated models interact with multiple toxicity metrics. It reveals, for the first time, that both task type and data domain significantly influence toxicity scores. Experiments demonstrate that existing benchmarks are prone to misclassifying content as harmful when tasks are altered and show inconsistent performance across domains, highlighting their fragility and dependence on specific model-task configurations. These findings underscore the urgent need for more robust and reliable toxicity evaluation frameworks.
0 citationsRead paper