Research map / 3D vision / 3D generation
RoomPainter: View-Integrated Diffusion for Consistent Indoor Scene Texturing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes RoomPainter, a zero-shot, training-free framework that adapts SDXL with multi-view integrated sampling and related-view attention to texture indoor scene meshes in two stages (room-level then per-instance repaint).
Problem
Indoor scene texture synthesis from text suffers from cross-view inconsistency and seams in per-view inpainting methods, while optimization-based methods (score distillation) are computationally expensive and unstable.
Approach
Zero-shot multi-view integrated sampling that merges per-view UV textures at each denoising step with view-weighted dynamic merging and related-view attention, followed by a per-instance repaint sampling stage that fills occluded regions while preserving global consistency.
Key results
- RoomPainter achieves CLIP score 23.47 and aesthetic score 5.03, versus 21.93/4.85 for Text2Tex-C and 21.87/4.75 for SceneTex.
- Full method takes 46 minutes per scene, versus 2614.5 minutes for SceneTex and 70.75 for Text2Tex-C (Text2Tex-H is faster at 8.5 minutes but scores lowest at 21.58 CLIP).
- Removing MVRS drops CLIP score from 23.47 to 22.39 and aesthetic score from 5.03 to 4.40.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Related View-based Attention (ablation) | RoomPainter · 3D-FRONT scenes | CLIP Score (CLIP): 23.32 -> 23.47 | Helped |
| add Related View-based Attention (ablation) | RoomPainter · 3D-FRONT scenes | Aesthetic Score (AS): 5.01 -> 5.03 | Helped |
| add Multi-View Integrated Sampling (MVIS) (ablation) | RoomPainter · 3D-FRONT scenes | CLIP Score (CLIP): 23.27 -> 23.47 | Helped |
| add Multi-View Integrated Sampling (MVIS) (ablation) | RoomPainter · 3D-FRONT scenes | Aesthetic Score (AS): 5.01 -> 5.03 | Helped |
| add Multi-View Integrated Repaint Sampling (MVRS) (ablation) | RoomPainter · 3D-FRONT scenes | CLIP Score (CLIP): 22.39 -> 23.47 | Helped |
| add Multi-View Integrated Repaint Sampling (MVRS) (ablation) | RoomPainter · 3D-FRONT scenes | Aesthetic Score (AS): 4.40 -> 5.03 | Helped |
| use RoomPainter instead of Text2Tex-C | Indoor scene texturing · 3D-FRONT scenes | CLIP Score (CLIP): 21.93 -> 23.47 | Helped |
| use RoomPainter instead of Text2Tex-C | Indoor scene texturing · 3D-FRONT scenes | Aesthetic Score (AS): 4.85 -> 5.03 | Helped |
| use RoomPainter instead of SceneTex | Indoor scene texturing · 3D-FRONT scenes | Generation time (mins): 2614.50 -> 46.00 | Helped |
Limitations
- Quantitative evaluation is limited to 10 scenes and five text styles.
- Text2Tex-H is faster than the proposed method, which is slower at 46 minutes per scene.
- Evaluation relies on CLIP-based metrics without geometric or human consistency measures.
Experimental setup
Datasets: 3D-FRONT
Benchmarks: 3D-FRONT (10 scenes, 5 text styles)
Metrics: CLIP score CLIP aesthetic score generation time (minutes)
Baselines: Text2Tex-H Text2Tex-C SceneTex
Builds on: Text2Tex TEXTure SDXL ControlNet Multi-view attention sharing (Cao et al. style)
Closest papers by meaning
- SceneTex: High-Quality Texture Synthesis for Indoor Scenes via Diffusion Priors (CVPR 2024 · similarity 0.89)
- DiffuScene: Denoising Diffusion Models for Generative Indoor Scene Synthesis (CVPR 2024 · similarity 0.87)
- DiffInDScene: Diffusion-based High-Quality 3D Indoor Scene Generation (CVPR 2024 · similarity 0.87)
- FreeScene: Mixed Graph Diffusion for 3D Scene Synthesis from Free Prompts (CVPR 2025 · similarity 0.86)
- InstaInpaint: Instant 3D-Scene Inpainting with Masked Large Reconstruction Model (NeurIPS 2025 · similarity 0.86)
- Paint3D: Paint Anything 3D with Lighting-Less Texture Diffusion Models (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.