Research map / Image generation / Editing and style transfer
ReDesign: Recovering Editable Design Structures from Raster Images via Agentic Decomposition
Jooyeol Yun, Jintae Park, Hyesu Lim, Junha Hyung, Hyungjin Chung, Choo Jaegul
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 lineal memory (ablation) | ReDesign · Figma (subset) | PQ: 39.09 -> 44.76 | Improved |
| add lineal memory (ablation) | ReDesign · Figma (subset) | L1: 0.0542 -> 0.0474 | Improved |
| use parallel tree expansion instead of sequential execution (ablation) | ReDesign · Figma (subset) | PQ: 42.00 -> 44.76 | Improved |
| use parallel tree expansion instead of sequential execution (ablation) | ReDesign · Figma (subset) | Tool calls: 36.8 -> 28.8 | Improved |
| increase tree depth from 2 to full depth (ablation) | ReDesign · Figma (subset) | PQ: 31.92 -> 44.76 | Improved |
| increase tree depth from 2 to full depth (ablation) | ReDesign · Figma (subset) | PSNR: 28.43 -> 25.56 | Hurt |
| increase tree depth from 3 to full depth (ablation) | ReDesign · Figma (subset) | F1: 0.4939 -> 0.5194 | Improved |
| increase tree depth from 3 to full depth (ablation) | ReDesign · Figma (subset) | API cost: $0.020 -> $0.027 | Hurt |
| replace Gemini-3 flash with GPT-5 mini as controller (ablation) | ReDesign · Figma (909 files) | PQ: 45.37 -> 43.39 | Mixed |
| use stepwise graceful verification instead of terminal verification | ReDesign · Figma (909 files) | PSNR and time | Improved |
| use tree-structured parallel expansion instead of serial tool use | ReDesign · Figma (909 files) | speedup: 7.1x speedup | Improved |
| use hierarchical expansion with verification instead of linear tool chain | ReDesign vs Tool Agent · Figma (909 files) | PSNR: 13.923 -> 26.286 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LayerD — LayerD: Decomposing Raster Graphic Designs into Layers (ICCV 2025) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Concept Lancet: Image Editing with Compositional Representation Transplant (CVPR 2025 · similarity 0.85)
- Decomposition of Graphic Design with Unified Multimodal Model (ICML 2025 · similarity 0.85)
- Pattern Analogies: Learning to Perform Programmatic Image Edits by Analogy (CVPR 2025 · similarity 0.84)
- LayerCraft: Enhancing Text-to-Image Generation with CoT Reasoning and Layered Object Integration (NeurIPS 2025 · similarity 0.84)
- PixPerfect: Seamless Latent Diffusion Local Editing with Discriminative Pixel-Space Refinement (NeurIPS 2025 · similarity 0.84)
- VisionDirector: Vision-Language Guided Closed-Loop Refinement for Generative Image Synthesis (CVPR 2026 · similarity 0.83)
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.
