Research map / 3D vision / 3D generation
GenRC: Generative 3D Room Completion from Sparse Image Collections
Ming-Feng Li, Yueh-Feng Ku, Hong-Xuan Yen, Chi Liu, Yu-Lun Liu, Albert Y Chen, Cheng-Hao Kuo, Min Sun
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 panorama inpainting instead of iterative novel-view filling (ablation) | GenRC · ScanNet (5% input) | depth MSE: 0.44 -> 0.27 | Improved |
| use E-Diffusion instead of plain MultiDiffusion for panorama inpainting (ablation) | GenRC · ScanNet (5% input) | PSNR: 13.8 -> 14.4 | Improved |
| add texture refinement (last 20 of 50 steps with MultiDiffusion) (ablation) | GenRC · ScanNet (10% input) | depth MSE: 0.21 -> 0.13 | Improved |
| add texture refinement (ablation) | GenRC · ScanNet (5% input) | PSNR: 14.6 -> 14.4 | No clear effect |
| add active sampling (3 panorama candidates, pick lowest depth MSE) (ablation) | GenRC · ScanNet (5% input) | depth MSE: 0.46 -> 0.27 | Improved |
| add textual inversion for the prompt token (ablation) | GenRC · ScanNet (5% input) | CLIP score: 0.792 -> 0.794 | Improved |
| add textual inversion for the prompt token (ablation) | GenRC · ScanNet (5% input) | PSNR: 14.7 -> 14.4 | Hurt |
| use 8 views in E-Diffusion instead of 6 or 16 (ablation) | GenRC · ScanNet (5% input) | depth MSE: 0.29 (6 views), 0.32 (16 views) -> 0.27 | Improved |
| use 20 texture refinement steps instead of 10 or 30 (ablation) | GenRC · ScanNet (5% input) | depth MSE: 0.37 (F=10), 0.35 (F=30) -> 0.27 | Improved |
| replace the Stable Diffusion-based GenRC by RGBD2 trained on ScanNet when moving to ARKitScenes | RGBD2 · ARKitScenes (5% input) | PSNR: 12.2 vs GenRC 13.2 | Hurt |
| use GenRC with shuffled instead of in-order camera trajectories | GenRC · ScanNet (3% input) | PSNR: 12.9 -> 12.7 | No clear effect |
| shuffle camera trajectories for RGBD2 | RGBD2 · ScanNet (3% input) | depth MSE: 0.72 -> 1.72 | Hurt |
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.
- RGBD2 — RGBD2: Generative Scene Synthesis via Incremental View Inpainting Using RGBD Diffusion Models (CVPR 2023) · better on 9 of 11 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
- GenFusion: Closing the Loop between Reconstruction and Generation via Videos (CVPR 2025 · similarity 0.86)
- GenPC: Zero-shot Point Cloud Completion via 3D Generative Priors (CVPR 2025 · similarity 0.86)
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · similarity 0.86)
- G3DR: Generative 3D Reconstruction in ImageNet (CVPR 2024 · similarity 0.86)
- NeRFiller: Completing Scenes via Generative 3D Inpainting (CVPR 2024 · similarity 0.86)
- CM-EVS: Sparse Panoramic RGB-D-Pose Data for Complete Scene Coverage (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.
