Research map / Video / Video generation
Refacade: Editing Object with Given Reference Texture
Youze Huang, Penghui Ruan, Bojia Zi, Xianbiao Qi, Jianan Wang, Rong Xiao
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 |
|---|---|---|---|
| replace Canny/HED/gray/depth structure conditions with texture-remover untextured video (ablation) | Refacade stage 1 · Video benchmark | foreground DINO score: 0.1906 (Canny) -> 0.2622 | Improved |
| add jigsaw permutation of the reference image (ablation) | Refacade stage 1 (Canny) · Video benchmark | background MSE: 68.83 -> 30.65 | Improved |
| add jigsaw permutation of the reference image (ablation) | Refacade stage 1 (Canny) · Video benchmark | foreground LPIPS: 0.7674 -> 0.7347 | Improved |
| add jigsaw permutation of the reference image (ablation) | Refacade stage 1 (Canny) · Video benchmark | GPT-5 score: 2.10 -> 2.42 | Improved |
| add stage-2 high-quality finetuning on Pexels (ablation) | Refacade · Video benchmark | foreground DINO score: 0.2622 -> 0.3241 | Improved |
| add stage-2 high-quality finetuning on Pexels (ablation) | Refacade · Video benchmark | EWarp: 1.3510 -> 1.4248 | Hurt |
| use small jigsaw patches (2%) instead of 5-10% (ablation) | Refacade jigsaw · Video benchmark | foreground DINO score: 0.2615 (10%) -> 0.2158 (2%) | Hurt |
| use large jigsaw patches (50-100%) instead of 5-10% (ablation) | Refacade jigsaw · Video benchmark | EWarp / texture fidelity: EWarp 1.4495 (10%) -> 1.3970 (100%); DINO 0.2615 -> 0.2380 | Mixed |
| jigsaw patch size changed between 2% and 100% (ablation) | Refacade jigsaw · Video benchmark | background LPIPS: 0.0326 for all patch sizes | No clear effect |
| distill texture remover with DMD2 from 50 to 3 sampling steps | Texture remover · Inference cost | sampling steps: 50 -> 3 steps | Improved |
| use Refacade instead of general-purpose and inpainting editors | Object Retexture · Image benchmark | user preference: 0.8944 vs 0.2236 (Flux-Kontext-Image) | Improved |
Closest papers by meaning
- Generating Non-Stationary Textures using Self-Rectification (CVPR 2024 · similarity 0.87)
- 3DOT: Texture Transfer for 3DGS Objects from a Single Reference Image (NeurIPS 2025 · similarity 0.86)
- Move Anything with Layered Scene Diffusion (CVPR 2024 · similarity 0.86)
- TextureDreamer: Image-Guided Texture Synthesis Through Geometry-Aware Diffusion (CVPR 2024 · similarity 0.86)
- CTRL-D: Controllable Dynamic 3D Scene Editing with Personalized 2D Diffusion (CVPR 2025 · similarity 0.86)
- PAIR Diffusion: A Comprehensive Multimodal Object-Level Image Editor (CVPR 2024 · 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.
