FaceSnap: Real-Time Personalized Lightstage Facial Performance Capture
FaceSnap通过两阶段方法，使用单个相机实现实时高精度面部捕捉，解决传统光舞台面部捕捉资源密集问题。
0 citationsRead paper
Selected work
Representative Papers
A Hybrid Neural-Microfacet BRDF Model for Real-Time Rendering
This work addresses the limitations of traditional microfacet BRDF models, which struggle to accurately reproduce complex material appearances, and existing neural BRDF approaches, which incur high computational costs and lack editability for real-time rendering. The authors propose a hybrid neural-microfacet BRDF model built upon the GGX distribution, augmented with a lightweight neural network that applies residual corrections to compensate for microfacet approximation errors. The method incorporates an importance sampling strategy tailored to the corrected BRDF. By design, it maintains computational overhead and memory footprint comparable to conventional models while significantly improving fidelity to measured reflectance data. The resulting representation achieves high visual accuracy, retains parameter editability, and remains compatible with real-time rendering pipelines, making it suitable for both offline and real-time applications.
0 citationsRead paper
Theory, Experience, and Instinct: A Glimpse Into AAA Game Processes and How UX Leaders Navigate Pre-Production
This study addresses the persistent challenge of translating academic frameworks into actionable practices during the pre-production phase of AAA game development, where theoretical models often falter due to misalignment with industrial constraints, production realities, and cross-functional collaboration demands. Through in-depth interviews with 15 AAA game UX leads and subsequent qualitative analysis, the research elucidates how design decisions integrate theory, experiential knowledge, and evidence-informed intuition to balance player needs, technical feasibility, and creative vision. The work proposes a flexible theoretical toolkit that operationalizes academic concepts into context-sensitive insights, systematizes tacit expertise, and adapts to dynamic development workflows. Key contributions include a shared language for cross-team alignment, reusable design systems, and adaptive strategies that offer a practical pathway to bridge the gap between academic research and industry practice.
0 citationsRead paper
StabilityBench: Benchmarking Instability in LLMs
This work addresses the limitations of current large language model (LLM) evaluations, which predominantly rely on static, single-turn benchmarks and fail to capture performance instability arising from contextual shifts in multi-turn dialogues. To this end, we propose StabilityBench—a general, model-agnostic benchmarking framework that transforms single-turn tasks into dynamic multi-turn interactions by incorporating user simulators, such as demographic personas or flattery-based decoys, thereby injecting context perturbations while preserving the original task intent. We introduce, for the first time, a systematic mechanism for multi-turn context perturbation and accompany it with a lightweight variant, StabilityBench-Mini, to balance evaluation efficiency and diversity. Experiments across nine mainstream LLMs reveal significant performance degradation in three-quarters of existing benchmarks, underscoring the inadequacy of static evaluation and demonstrating that StabilityBench effectively uncovers LLM instability under dynamic interaction scenarios.
0 citationsRead paper
KM-Speaker: Keypoint-Based Style Control for High-Quality Speech-Driven 3D Facial Animation and Dialogue Localization
Existing audio-driven 3D facial animation methods struggle to simultaneously achieve high fidelity, artistic style control, and frame-level temporal precision, particularly limiting their performance in dialogue localization tasks such as dubbing. This work proposes a keypoint-conditioned flow-based generative framework that decouples audio-driven lip motion from keypoint-driven upper-face dynamics. By incorporating a global style context preservation mechanism and a reference performance-guided temporal control strategy, the method enables high-quality and controllable animation synthesis. Experimental results demonstrate that the proposed approach outperforms state-of-the-art methods in lip-sync accuracy, style consistency, and expressive temporal precision, significantly enhancing both realism and controllability.
0 citationsRead paper
Recent publications
Latest Papers
FaceSnap: Real-Time Personalized Lightstage Facial Performance Capture
FaceSnap通过两阶段方法，使用单个相机实现实时高精度面部捕捉，解决传统光舞台面部捕捉资源密集问题。
0 citationsRead paper
A Hybrid Neural-Microfacet BRDF Model for Real-Time Rendering
This work addresses the limitations of traditional microfacet BRDF models, which struggle to accurately reproduce complex material appearances, and existing neural BRDF approaches, which incur high computational costs and lack editability for real-time rendering. The authors propose a hybrid neural-microfacet BRDF model built upon the GGX distribution, augmented with a lightweight neural network that applies residual corrections to compensate for microfacet approximation errors. The method incorporates an importance sampling strategy tailored to the corrected BRDF. By design, it maintains computational overhead and memory footprint comparable to conventional models while significantly improving fidelity to measured reflectance data. The resulting representation achieves high visual accuracy, retains parameter editability, and remains compatible with real-time rendering pipelines, making it suitable for both offline and real-time applications.
0 citationsRead paper
Theory, Experience, and Instinct: A Glimpse Into AAA Game Processes and How UX Leaders Navigate Pre-Production
This study addresses the persistent challenge of translating academic frameworks into actionable practices during the pre-production phase of AAA game development, where theoretical models often falter due to misalignment with industrial constraints, production realities, and cross-functional collaboration demands. Through in-depth interviews with 15 AAA game UX leads and subsequent qualitative analysis, the research elucidates how design decisions integrate theory, experiential knowledge, and evidence-informed intuition to balance player needs, technical feasibility, and creative vision. The work proposes a flexible theoretical toolkit that operationalizes academic concepts into context-sensitive insights, systematizes tacit expertise, and adapts to dynamic development workflows. Key contributions include a shared language for cross-team alignment, reusable design systems, and adaptive strategies that offer a practical pathway to bridge the gap between academic research and industry practice.
0 citationsRead paper
StabilityBench: Benchmarking Instability in LLMs
This work addresses the limitations of current large language model (LLM) evaluations, which predominantly rely on static, single-turn benchmarks and fail to capture performance instability arising from contextual shifts in multi-turn dialogues. To this end, we propose StabilityBench—a general, model-agnostic benchmarking framework that transforms single-turn tasks into dynamic multi-turn interactions by incorporating user simulators, such as demographic personas or flattery-based decoys, thereby injecting context perturbations while preserving the original task intent. We introduce, for the first time, a systematic mechanism for multi-turn context perturbation and accompany it with a lightweight variant, StabilityBench-Mini, to balance evaluation efficiency and diversity. Experiments across nine mainstream LLMs reveal significant performance degradation in three-quarters of existing benchmarks, underscoring the inadequacy of static evaluation and demonstrating that StabilityBench effectively uncovers LLM instability under dynamic interaction scenarios.
0 citationsRead paper
KM-Speaker: Keypoint-Based Style Control for High-Quality Speech-Driven 3D Facial Animation and Dialogue Localization
Existing audio-driven 3D facial animation methods struggle to simultaneously achieve high fidelity, artistic style control, and frame-level temporal precision, particularly limiting their performance in dialogue localization tasks such as dubbing. This work proposes a keypoint-conditioned flow-based generative framework that decouples audio-driven lip motion from keypoint-driven upper-face dynamics. By incorporating a global style context preservation mechanism and a reference performance-guided temporal control strategy, the method enables high-quality and controllable animation synthesis. Experimental results demonstrate that the proposed approach outperforms state-of-the-art methods in lip-sync accuracy, style consistency, and expressive temporal precision, significantly enhancing both realism and controllability.
0 citationsRead paper