Selected work

Representative Papers

LVS: Local View Synthesis from Relative Camera Pose by Reusing Previous Views

Oct 08, 2026

This study addresses the redundant rendering and high latency caused by frequent viewpoint updates during interactive scene exploration. To this end, it proposes a local view synthesis framework based on relative camera poses. This work pioneers the decoupling of scene rendering from local view updates, leveraging relative poses to guide RGB-D image reuse. By integrating 3D Gaussian Splatting, geometric warping, a lightweight multi-scale residual network, and a feature caching mechanism, the method enables efficient generation of neighboring views. Experimental results demonstrate a 0.72 dB improvement in PSNR over pure warping approaches, alongside significantly reduced computational overhead and query latency. These advances effectively support real-time interaction in AR/VR applications.

0 citationsRead paper

PP-Net: A Hybrid Physical-Prior Neural Network for Scattered Light Removal in Biomedical Images on Embedded Devices

Sep 22, 2026

为解决生物医学图像中散射光去除难题，提出PP-Net混合物理先验神经网络，通过DFN-Net、ASAP和GF-Net组件及渐进合成训练策略有效提升图像质量并降低计算成本。

0 citationsRead paper

VADER: Adaptive Debiasing for Hallucination Mitigation in Video Large Language Models

Aug 09, 2026

This work addresses the prevalent issue of hallucinated responses in video large language models (VideoLLMs) during open-ended video understanding, where generated answers often lack grounding in visual evidence. To mitigate this without requiring additional training, the authors propose an adaptive, training-free debiasing framework that dynamically reweights visual evidence to suppress hallucinations. The approach integrates cross-layer diagnosis of vision–text evidence flow, pre-softmax attention redistribution, masking of high-importance visual tokens, and contrastive decoding to enable fine-grained, frame-wise visual focus intervention. Experimental results demonstrate that the framework significantly improves event-level localization accuracy and temporal consistency across multiple VideoLLMs, achieving a 72.60% accuracy on the EventHallusion benchmark with LLaVA-Video-7B.

0 citationsRead paper

Recent publications

Latest Papers

LVS: Local View Synthesis from Relative Camera Pose by Reusing Previous Views

Oct 08, 2026

This study addresses the redundant rendering and high latency caused by frequent viewpoint updates during interactive scene exploration. To this end, it proposes a local view synthesis framework based on relative camera poses. This work pioneers the decoupling of scene rendering from local view updates, leveraging relative poses to guide RGB-D image reuse. By integrating 3D Gaussian Splatting, geometric warping, a lightweight multi-scale residual network, and a feature caching mechanism, the method enables efficient generation of neighboring views. Experimental results demonstrate a 0.72 dB improvement in PSNR over pure warping approaches, alongside significantly reduced computational overhead and query latency. These advances effectively support real-time interaction in AR/VR applications.

0 citationsRead paper

PP-Net: A Hybrid Physical-Prior Neural Network for Scattered Light Removal in Biomedical Images on Embedded Devices

Sep 22, 2026

为解决生物医学图像中散射光去除难题，提出PP-Net混合物理先验神经网络，通过DFN-Net、ASAP和GF-Net组件及渐进合成训练策略有效提升图像质量并降低计算成本。

0 citationsRead paper

VADER: Adaptive Debiasing for Hallucination Mitigation in Video Large Language Models

Aug 09, 2026

This work addresses the prevalent issue of hallucinated responses in video large language models (VideoLLMs) during open-ended video understanding, where generated answers often lack grounding in visual evidence. To mitigate this without requiring additional training, the authors propose an adaptive, training-free debiasing framework that dynamically reweights visual evidence to suppress hallucinations. The approach integrates cross-layer diagnosis of vision–text evidence flow, pre-softmax attention redistribution, masking of high-importance visual tokens, and contrastive decoding to enable fine-grained, frame-wise visual focus intervention. Experimental results demonstrate that the framework significantly improves event-level localization accuracy and temporal consistency across multiple VideoLLMs, achieving a 72.60% accuracy on the EventHallusion benchmark with LLaVA-Video-7B.

0 citationsRead paper