Research map / 3D vision / 3D generation
LaRP: Efficient Multi-View Inpainting with Latent Reprojection Priors
Gaoyang Zhang, Xinguo Liu
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 |
|---|---|---|---|
| use LaRP latent reprojection instead of ControlNet on reprojected pixels (ablation) | LaRP · SPIn-NeRF | FID: 40.12 -> 34.84 | Improved |
| use LaRP conditioning instead of cross-attention (ablation) | LaRP · SPIn-NeRF | MEt3R (MASt3R): 0.1735 -> 0.1109 | Improved |
| keep UNet decoder locked instead of unlocking it (ablation) | LaRP · SPIn-NeRF | FID: 37.18 -> 34.84 | Improved |
| train on two-view data instead of single-view CC12M (ablation) | LaRP · SPIn-NeRF | FID: 44.15 -> 34.84 | Improved |
| add latent dropout to single-view training (ablation) | LaRP · SPIn-NeRF | FID: 44.15 -> 43.29 | No clear effect |
| add FPS-based view selection (ablation) | LaRP · SPIn-NeRF | FID: 36.19 -> 34.84 | Improved |
| add 3D-aware mask preparation (ablation) | LaRP · SPIn-NeRF | LPIPS: 0.2789 -> 0.2458 | Improved |
| train on all object categories instead of single category (ablation) | LaRP · SPIn-NeRF | FID: 35.02 -> 34.84 | No clear effect |
| use LaRP cloned-encoder design instead of ControlNet (ablation) | LaRP · training convergence | steps to sudden convergence: ~7,000 -> ~2,000 steps | Improved |
| use LaRP instead of MVInpainter-F for training cost | LaRP · training | total training time: 3 days -> 14 hours | Improved |
| use LaRP+NeRF 20 min instead of MALD-NeRF 960 min | LaRP · SPIn-NeRF | time: 960 min -> 20 min | Improved |
| use LaRP inpaintings with 4 min NeRF training instead of 20 min (ablation) | LaRP · SPIn-NeRF | LPIPS: 0.2458 -> 0.3006 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- AuraFusion360 — AuraFusion360: Augmented Unseen Region Alignment for Reference-based 360deg Unbounded Scene Inpainting (CVPR 2025) · better on 1 of 2 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
- MVIP-NeRF: Multi-view 3D Inpainting on NeRF Scenes via Diffusion Prior (CVPR 2024 · similarity 0.89)
- MVInpainter: Learning Multi-View Consistent Inpainting to Bridge 2D and 3D Editing (NeurIPS 2024 · similarity 0.89)
- RAD: Region-Aware Diffusion Models for Image Inpainting (CVPR 2025 · similarity 0.86)
- InstaInpaint: Instant 3D-Scene Inpainting with Masked Large Reconstruction Model (NeurIPS 2025 · similarity 0.86)
- In-N-Out: Lifting 2D Diffusion Prior for 3D Object Removal via Tuning-Free Latents Alignment (NeurIPS 2024 · similarity 0.86)
- Instant3dit: Multiview Inpainting for Fast Editing of 3D Objects (CVPR 2025 · similarity 0.85)
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.
