CVPR Findings2026
Evolutionary Multi-Agent Collaboration for Real-World Video Face Restoration
Bowen Tang, Tao Wang, Miao Zhang, Xin Yu, Jinwei Chen, Bo Li, Kaihao Zhang
Abstract
Video Face Restoration (VFR) aims to enhance the quality of degraded face videos. Existing approaches either apply face image restoration models independently to each frame, which often leads to poor temporal consistency due to the lack of temporal modeling, or use VFR-specific methods that struggle to balance fine facial detail recovery with temporal stability. To address these, we propose MA-VFR, a collaborative multi-agent system for video face restoration. Specifically, MA-VFR consists of a facical agent, a temporal agent, and an evolving coordinator agent. The facial agent specializes in restoring facial details, while the temporal agent enforces temporal coherence. Each agent builds its own memory by modeling toolchains tailored to its specific objective and then constructs an initial toolchain. The coordinator agent, acting as a decision-maker, adaptively combines these toolchains to derive an optimal restoration plan. Furthermore, the coordinator memory bank is periodically updated based on restoration results, which enhances the system adaptability and promotes more reliable collaboration over time. By incorporating these specialized agents and evolving memory mechanisms, MA-VFR balances the restoration of facial details with temporal consistency. Extensive experiments demonstrate that our MA-VFR achieves competitive performance.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
Builds on26
- Toolformer: Language Models Can Teach Themselves to Use ToolsTimo Schick, Jane Dwivedi-Yu, Roberto Dessì, Roberta Raileanu et al.NeurIPS 2023 · 5,989 citations
- CAMEL: Communicative Agents for "Mind" Exploration of Large Language Model SocietyGuohao Li, Hasan Hammoud, Hani Itani, Dmitrii Khizbullin et al.NeurIPS 2023 · 1,975 citations
- MUSIQ: Multi-scale Image Quality TransformerJunjie Ke, Qifei Wang, Yilin Wang, Peyman Milanfar et al.ICCV 2021 · 1,325 citations
- Exploring CLIP for Assessing the Look and Feel of ImagesJianyi Wang, Kelvin C. K. Chan, Chen Change LoyAAAI 2023 · 1,208 citations
- A-Mem: Agentic Memory for LLM AgentsWujiang Xu, Zujie Liang, Kai Mei, Hang Gao et al.NeurIPS 2025 · 1,138 citations
Related papers
- Many Minds, One Goal: Time Series Forecasting via Sub-task Specialization and Inter-agent CooperationQihe Huang, Zhengyang Zhou, Yangze Li, Kuo Yang et al.NeurIPS 2025 · 11 citations
- TemCoCo: Temporally Consistent Multi-Modal Video Fusion with Visual-Semantic CollaborationMeiqi Gong, Hao Zhang, Xunpeng Yi, Linfeng Tang et al.ICCV 2025 · 4 citations
- Beyond Sequential Tools: A Unified VLM Agent System for Photographic Post-Processing via Dynamic Multi-Expert FusionHonglin Xiong, Chenjie Zhu, Jianbiao Ding, Zixuan Ni et al.CVPR 2026
- Blind Face Video Restoration with Temporal Consistent Generative Prior and Degradation-Aware PromptJingfan Tan, Hyunhee Park, Ying Zhang, Tao Wang et al.ACM MM 2024 · 10 citations
- FMA-Net: Flow-Guided Dynamic Filtering and Iterative Feature Refinement with Multi-Attention for Joint Video Super-Resolution and DeblurringGeunhyuk Youk, Jihyong Oh, Munchurl KimCVPR 2024 · 16 citations
