Selected work

Representative Papers

TACROSS: An Efficient and Low-Cost Scalable Human Touch System Across Heterogeneous Tactile Sensors for Dexterous Robot Learning

Oct 08, 2026

This study addresses the high cost of tactile data acquisition and the signal alignment challenges arising from structural heterogeneity between human hands and robotic sensors. To overcome these limitations, this work proposes a low-cost piezoresistive five-layer tactile glove alongside a cross-modal alignment mechanism based on contact events rather than raw sensor values. A temporal Transformer is employed to map heterogeneous signals into a shared latent space, enabling a robot-centric policy learning framework for dexterous manipulation. The proposed approach increases data collection efficiency by 3.5× while reducing hardware costs by 95.7%. Furthermore, the complete software and hardware designs, along with a 150-hour tactile dataset, are released as open source to facilitate future research in tactile sensing and robotic manipulation.

0 citationsRead paper

Hyperbolic Geometry for Open-World Object Detection in Remote Sensing Imagery

Sep 08, 2026

为解决遥感图像中开放世界目标检测的问题，提出使用双曲几何方法，通过解耦物体学习模块和双曲不确定性学习组件提高未知目标召回率，并采用双曲度量学习策略改善增量学习性能。

0 citationsRead paper

Depth-Dominant Skeleton Detection for Natural Scenes

Aug 17, 2026

This study addresses the performance bottleneck of pure RGB skeleton detection in complex scenarios by proposing a novel depth-dominant, RGB-assisted paradigm embodied in the DDSkel model. The framework employs an asymmetric encoder for cross-modal feature fusion and incorporates a lightweight RGB branch comprising only 12% of the parameters, thereby ensuring robustness while significantly reducing computational overhead. Experimental evaluations on the SymPASCAL dataset demonstrate that the proposed method surpasses current state-of-the-art approaches while utilizing merely 36% of their trainable parameters. Consequently, this work achieves a substantial balance between accuracy and efficiency, offering a highly effective solution for resource-constrained skeleton detection tasks without compromising performance in challenging environments.

0 citationsRead paper

Multi-illuminant Color Constancy via Multi-scale Illuminant Estimation and Fusion

Feb 04, 2025

Existing color constancy methods suffer from localized chromatic distortions in multi-illuminant scenes due to single-scale illumination modeling. To address this, we propose a multi-scale illumination estimation and fusion framework. Our method employs a three-branch convolutional network to jointly extract features at multiple scales, models the illumination map as a linear combination of scale-specific illumination components, and introduces an attention-driven adaptive fusion module that dynamically weights each component for pixel-wise illumination prediction. This design enables precise, spatially varying illumination estimation tailored to complex lighting conditions. Evaluated on multiple benchmark datasets, our approach achieves state-of-the-art performance in color restoration accuracy—particularly in regions with heterogeneous illuminants—demonstrating the effectiveness of multi-scale modeling and attention-guided adaptive fusion for illumination estimation.

0 citationsRead paper

Recent publications

Latest Papers

TACROSS: An Efficient and Low-Cost Scalable Human Touch System Across Heterogeneous Tactile Sensors for Dexterous Robot Learning

Oct 08, 2026

This study addresses the high cost of tactile data acquisition and the signal alignment challenges arising from structural heterogeneity between human hands and robotic sensors. To overcome these limitations, this work proposes a low-cost piezoresistive five-layer tactile glove alongside a cross-modal alignment mechanism based on contact events rather than raw sensor values. A temporal Transformer is employed to map heterogeneous signals into a shared latent space, enabling a robot-centric policy learning framework for dexterous manipulation. The proposed approach increases data collection efficiency by 3.5× while reducing hardware costs by 95.7%. Furthermore, the complete software and hardware designs, along with a 150-hour tactile dataset, are released as open source to facilitate future research in tactile sensing and robotic manipulation.

0 citationsRead paper

Hyperbolic Geometry for Open-World Object Detection in Remote Sensing Imagery

Sep 08, 2026

为解决遥感图像中开放世界目标检测的问题，提出使用双曲几何方法，通过解耦物体学习模块和双曲不确定性学习组件提高未知目标召回率，并采用双曲度量学习策略改善增量学习性能。

0 citationsRead paper

Depth-Dominant Skeleton Detection for Natural Scenes

Aug 17, 2026

This study addresses the performance bottleneck of pure RGB skeleton detection in complex scenarios by proposing a novel depth-dominant, RGB-assisted paradigm embodied in the DDSkel model. The framework employs an asymmetric encoder for cross-modal feature fusion and incorporates a lightweight RGB branch comprising only 12% of the parameters, thereby ensuring robustness while significantly reducing computational overhead. Experimental evaluations on the SymPASCAL dataset demonstrate that the proposed method surpasses current state-of-the-art approaches while utilizing merely 36% of their trainable parameters. Consequently, this work achieves a substantial balance between accuracy and efficiency, offering a highly effective solution for resource-constrained skeleton detection tasks without compromising performance in challenging environments.

0 citationsRead paper

Multi-illuminant Color Constancy via Multi-scale Illuminant Estimation and Fusion

Feb 04, 2025

Existing color constancy methods suffer from localized chromatic distortions in multi-illuminant scenes due to single-scale illumination modeling. To address this, we propose a multi-scale illumination estimation and fusion framework. Our method employs a three-branch convolutional network to jointly extract features at multiple scales, models the illumination map as a linear combination of scale-specific illumination components, and introduces an attention-driven adaptive fusion module that dynamically weights each component for pixel-wise illumination prediction. This design enables precise, spatially varying illumination estimation tailored to complex lighting conditions. Evaluated on multiple benchmark datasets, our approach achieves state-of-the-art performance in color restoration accuracy—particularly in regions with heterogeneous illuminants—demonstrating the effectiveness of multi-scale modeling and attention-guided adaptive fusion for illumination estimation.

0 citationsRead paper