Institution profile

China Mobile

Industry researchasia · cn
Official website
Research library54linked papers
Opportunities0open roles
Selected work

Representative Papers

FaceRefiner: High-Fidelity Facial Texture Refinement With Differentiable Rendering-Based Style Transfer

Jan 08, 2026IEEE transactions on multimedia

Existing facial texture generation methods suffer from limited generalization on in-the-wild images, often producing UV textures that deviate from the input in terms of fine details, structural fidelity, and identity consistency. To address this, this work proposes FaceRefiner, which introduces differentiable rendering into a style transfer framework for the first time. By treating 3D-sampled textures as style and generated textures as content, FaceRefiner enables pixel-wise, multi-level (low-, mid-, and high-level) information transfer directly in UV space. This approach significantly enhances both photorealism and identity preservation in the synthesized textures. Extensive experiments on Multi-PIE, CelebA, and FFHQ demonstrate that FaceRefiner consistently outperforms state-of-the-art methods by a substantial margin.

2 citationsRead paper

CCR-Bench: A Comprehensive Benchmark for Evaluating LLMs on Complex Constraints, Control Flows, and Real-World Cases

Mar 09, 2026

Existing evaluation methods for large language models struggle to capture the high-dimensional characteristics of complex instructions encountered in real-world scenarios, leading to a misalignment between benchmark performance and practical requirements. This work proposes the first comprehensive evaluation benchmark that deeply integrates content and format constraints, logical control flow, and authentic industrial use cases. By decomposing tasks, incorporating conditional reasoning, and modeling procedural planning, the framework generates high-fidelity complex instruction samples. It moves beyond the conventional paradigm of atomized constraint composition to systematically assess a model’s capacity to understand and execute intricate, multi-faceted instructions. Experimental results demonstrate that even state-of-the-art models exhibit significant performance gaps on this benchmark, clearly exposing the disparity between current capabilities and real-world application demands.

1 citationsRead paper

EdgeMem: LLM-Free Agent Memory Construction and Retrieval via Evidence-Preserving Multi-Anchor Hypergraph

Sep 03, 2026

Agent memory allows LLM agents to use earlier interactions when answering new queries. Existing methods often compress interaction histories into summaries or other LLM-generated representations. Repeated generation adds cost and can discard answer-bearing details before the system knows what a future query will require. We propose EdgeMem, an agent-memory method built around a simple principle: preserve original interaction turns and organize them through complementary content, temporal, and episodic cues. EdgeMem realizes this principle with a multi-anchor hypergraph constructed by lightweight local processing. Retrieval directly returns source evidence and reserves LLM use for final answer generation, combining structured access to multi-session histories with faithful retention of the original conversation. Experiments on LoCoMo and LongMemEval-S show strong retrieval and memory-grounded question answering; on LoCoMo, EdgeMem achieves the highest strict-judge score among seven reproduced systems under a shared prompt (61.01 versus 58.70), while construction and retrieval require no generative-LLM calls. Overall, EdgeMem shows that preserving and organizing source evidence provides an effective and efficient foundation for agent memory without generative memory management.

0 citationsRead paper
Recent publications

Latest Papers

EdgeMem: LLM-Free Agent Memory Construction and Retrieval via Evidence-Preserving Multi-Anchor Hypergraph

Sep 03, 2026

Agent memory allows LLM agents to use earlier interactions when answering new queries. Existing methods often compress interaction histories into summaries or other LLM-generated representations. Repeated generation adds cost and can discard answer-bearing details before the system knows what a future query will require. We propose EdgeMem, an agent-memory method built around a simple principle: preserve original interaction turns and organize them through complementary content, temporal, and episodic cues. EdgeMem realizes this principle with a multi-anchor hypergraph constructed by lightweight local processing. Retrieval directly returns source evidence and reserves LLM use for final answer generation, combining structured access to multi-session histories with faithful retention of the original conversation. Experiments on LoCoMo and LongMemEval-S show strong retrieval and memory-grounded question answering; on LoCoMo, EdgeMem achieves the highest strict-judge score among seven reproduced systems under a shared prompt (61.01 versus 58.70), while construction and retrieval require no generative-LLM calls. Overall, EdgeMem shows that preserving and organizing source evidence provides an effective and efficient foundation for agent memory without generative memory management.

0 citationsRead paper