Research map / Vision-language and multimodal / CLIP and zero-shot
Beyond Sequential Tools: A Unified VLM Agent System for Photographic Post-Processing via Dynamic Multi-Expert Fusion
Honglin Xiong, Chenjie Zhu, Jianbiao Ding, Zixuan Ni, Wei Li, Zhenpeng Mi, Qian Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add VLM-enhanced prompts to the Flux-Kontext backbone (ablation) | Flux-Kontext · Real-1000 Group 1 | PSNR: 17.69 -> 19.37 | Improved |
| add VLM-enhanced prompts to the Flux-Kontext backbone (ablation) | Flux-Kontext · Real-1000 Group 1 | MANIQA: 0.2863 -> 0.2847 | No clear effect |
| add specialized expert LoRA module (ablation) | Flux-Kontext + VLM prompt · Real-1000 Group 1 | PSNR: 19.37 -> 22.90 | Improved |
| add specialized expert LoRA module (ablation) | Flux-Kontext + VLM prompt · Real-1000 Group 1 | MUSIQ: 51.35 -> 60.67 | Improved |
| add DPO-based alignment of weight allocation branch (ablation) | K-V only expert fusion · Real-1000 Group 2 | PSNR: 20.12 -> 21.10 | Improved |
| add dynamic VLM-allocated weights instead of fixed uniform weights (ablation) | K-V only expert fusion · Real-1000 Group 2 | PSNR: 19.56 -> 21.10 | Improved |
| restrict LoRA adaptation to K and V instead of Q, K, V (ablation) | expert LoRAs with DPO weights · Real-1000 Group 2 | PSNR: 20.45 -> 21.10 | Improved |
| add DPO alignment on top of heuristic weights (ablation) | K-V only expert fusion · Real-1000 Group 2 | MUSIQ: 54.72 -> 55.84 | Improved |
| use full system instead of Flux-Kontext baseline | Flux-Kontext · Real-1000 Group 3 | PSNR: 12.51 -> 19.25 | Improved |
| use single-pass fused experts instead of sequential agent tools | AgenticIR · Real-1000 Group 2 | PSNR: 15.37 -> 21.10 | Improved |
| use full system instead of 4K Agent on MiO100 Group C | 4K Agent · MiO100 Group C | CLIPIQA: 0.5233 -> 0.4742 | Hurt |
Closest papers by meaning
- An Intelligent Agentic System for Complex Image Restoration Problems (ICLR 2025 · similarity 0.88)
- Hybrid Agents for Image Restoration (CVPR 2026 · similarity 0.88)
- Complexity Experts are Task-Discriminative Learners for Any Image Restoration (CVPR 2025 · similarity 0.87)
- RestoreAgent: Autonomous Image Restoration Agent via Multimodal Large Language Models (NeurIPS 2024 · similarity 0.87)
- Continuous Expert Assembly: Instance-Conditioned Low-Rank Residuals for All-in-One Image Restoration (NeurIPS 2026 · similarity 0.86)
- OPERA: An Agent for Image Restoration with End-to-End Joint Planning–Execution Optimization (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
