Field Converter: Geometry-Initialized Temporal Residual Refinement for World-Grounded Player Pose Estimation from Soccer Broadcasts
研究解决了从足球广播中估计球员3D姿态的问题，提出了一种基于几何初始化的时间残差框架Field Converter，通过结合图像、相机参数和几何信息来优化姿态估计。
0 citationsRead paper
Selected work
Representative Papers
Bayesian Expected Uncertainty Reduction (B-EUR) Model: A Computational Account of What Makes Design Options Worth Trying
This work proposes the Bayesian Expected Uncertainty Reduction (B-EUR) model, which formally defines the “epistemic value” of candidate actions in design exploration as the expected reduction in uncertainty about action–outcome relationships. Through a graph-structure guessing task integrating Bayesian inference, computational modeling, behavioral experiments, and simulation analyses, the study reveals that epistemic value follows an inverted-U relationship with environmental generalizability and increases monotonically with outcome discriminability. Moreover, human subjective willingness to explore and experienced pleasure also exhibit an inverted-U dependence on generalizability and jointly shape choice behavior. These findings provide a theoretical foundation for prototype set construction and feedback design in exploratory tasks.
0 citationsRead paper
Impact Detection in Fall Events: Leveraging Spatio-Temporal Graph Convolutional Networks and Recurrent Neural Networks Using 3D Skeletons Data
This study addresses the challenge of distinguishing genuine fall impacts from non-impact balance losses in fall monitoring. To this end, the authors propose a spatiotemporal graph modeling approach based on 3D skeletal data. By constructing a spatiotemporal graph of human joints and leveraging a Spatial-Temporal Graph Convolutional Network (STGCN) to extract spatial-temporal features, the method further integrates GRU and BiLSTM modules to enhance temporal dynamics modeling for precise identification of fall impact moments. This work presents the first integration of STGCN with bidirectional recurrent neural networks for fall detection, achieving over 90% accuracy on an enhanced version of the UP-Fall dataset. The proposed approach significantly improves the discrimination between true and false falls, and the refined dataset is publicly released to support future research in this domain.
0 citationsRead paper
Synthetic data generation framework for quality control automation in gravure printing
This study addresses the challenge of deploying automated quality inspection in gravure printing, where real defect samples are extremely scarce. To overcome this limitation, the authors propose the first high-fidelity synthetic data generation framework tailored to this domain. Leveraging procedural image synthesis techniques, the framework automatically generates realistic instances of common defects—such as wrinkles, streaks, and misregistration—along with precise pixel-level annotations, enabling zero-shot deployment without any reliance on real defective images. A detection model, RFDETR, trained exclusively on 7,533 synthetically generated images achieves a mean average precision (mAP) of 80.9% on a real-world industrial test set, demonstrating the effectiveness and practical utility of the proposed approach.
0 citationsRead paper
CRADIPOR: Crash Dispersion Predictor
This study addresses the numerical dispersion inherent in automotive crash simulations—arising from parallel computation and model complexity—which undermines engineering decision-making, while conventional approaches relying on repeated simulations incur prohibitive computational costs. To overcome this, the authors propose a novel post-processing framework that integrates a Reduced-Rank Autoencoder (RRAE) with supervised classification to efficiently identify regions sensitive to numerical dispersion without requiring additional simulations. By leveraging structured latent representations and signal features such as slope variations, the method substantially enhances the detection of dispersion-sensitive areas. Experimental results demonstrate that the proposed framework outperforms a random forest baseline on the test set, confirming its effectiveness and practicality for stability assessment in crash simulation post-processing.
0 citationsRead paper
Recent publications
Latest Papers
Field Converter: Geometry-Initialized Temporal Residual Refinement for World-Grounded Player Pose Estimation from Soccer Broadcasts
研究解决了从足球广播中估计球员3D姿态的问题，提出了一种基于几何初始化的时间残差框架Field Converter，通过结合图像、相机参数和几何信息来优化姿态估计。
0 citationsRead paper
Bayesian Expected Uncertainty Reduction (B-EUR) Model: A Computational Account of What Makes Design Options Worth Trying
This work proposes the Bayesian Expected Uncertainty Reduction (B-EUR) model, which formally defines the “epistemic value” of candidate actions in design exploration as the expected reduction in uncertainty about action–outcome relationships. Through a graph-structure guessing task integrating Bayesian inference, computational modeling, behavioral experiments, and simulation analyses, the study reveals that epistemic value follows an inverted-U relationship with environmental generalizability and increases monotonically with outcome discriminability. Moreover, human subjective willingness to explore and experienced pleasure also exhibit an inverted-U dependence on generalizability and jointly shape choice behavior. These findings provide a theoretical foundation for prototype set construction and feedback design in exploratory tasks.
0 citationsRead paper
Impact Detection in Fall Events: Leveraging Spatio-Temporal Graph Convolutional Networks and Recurrent Neural Networks Using 3D Skeletons Data
This study addresses the challenge of distinguishing genuine fall impacts from non-impact balance losses in fall monitoring. To this end, the authors propose a spatiotemporal graph modeling approach based on 3D skeletal data. By constructing a spatiotemporal graph of human joints and leveraging a Spatial-Temporal Graph Convolutional Network (STGCN) to extract spatial-temporal features, the method further integrates GRU and BiLSTM modules to enhance temporal dynamics modeling for precise identification of fall impact moments. This work presents the first integration of STGCN with bidirectional recurrent neural networks for fall detection, achieving over 90% accuracy on an enhanced version of the UP-Fall dataset. The proposed approach significantly improves the discrimination between true and false falls, and the refined dataset is publicly released to support future research in this domain.
0 citationsRead paper
Synthetic data generation framework for quality control automation in gravure printing
This study addresses the challenge of deploying automated quality inspection in gravure printing, where real defect samples are extremely scarce. To overcome this limitation, the authors propose the first high-fidelity synthetic data generation framework tailored to this domain. Leveraging procedural image synthesis techniques, the framework automatically generates realistic instances of common defects—such as wrinkles, streaks, and misregistration—along with precise pixel-level annotations, enabling zero-shot deployment without any reliance on real defective images. A detection model, RFDETR, trained exclusively on 7,533 synthetically generated images achieves a mean average precision (mAP) of 80.9% on a real-world industrial test set, demonstrating the effectiveness and practical utility of the proposed approach.
0 citationsRead paper
CRADIPOR: Crash Dispersion Predictor
This study addresses the numerical dispersion inherent in automotive crash simulations—arising from parallel computation and model complexity—which undermines engineering decision-making, while conventional approaches relying on repeated simulations incur prohibitive computational costs. To overcome this, the authors propose a novel post-processing framework that integrates a Reduced-Rank Autoencoder (RRAE) with supervised classification to efficiently identify regions sensitive to numerical dispersion without requiring additional simulations. By leveraging structured latent representations and signal features such as slope variations, the method substantially enhances the detection of dispersion-sensitive areas. Experimental results demonstrate that the proposed framework outperforms a random forest baseline on the test set, confirming its effectiveness and practicality for stability assessment in crash simulation post-processing.
0 citationsRead paper