Research map / 3D vision / 3D generation
Real3D: Towards Scaling Large Reconstruction Models with Real Images
Hanwen Jiang, Qixing Huang, Georgios Pavlakos
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 input-view rendering loss L_in (ablation) | Real3D · CO3D | PSNR: 18.44 -> 18.60 | Improved |
| add semantic-level self-supervision L_sem (ablation) | Real3D · CO3D | PSNR: 18.60 -> 18.81 | Improved |
| add pixel-level cycle-consistency loss L_pix (ablation) | Real3D · CO3D | PSNR: 18.81 -> 19.18 | Improved |
| use naive CLIP semantic loss on all novel views instead of no semantic loss (ablation) | Real3D · CO3D | PSNR: 18.60 -> 17.89 | Hurt |
| use regularized L_sem instead of naive semantic loss (ablation) | Real3D · CO3D | PSNR: 17.89 -> 18.81 | Improved |
| add stop-gradient on intermediate rendering in pixel loss (ablation) | Real3D · CO3D | PSNR: 17.78 -> 19.18 | Improved |
| add training curriculum to pixel loss (ablation) | Real3D · CO3D | PSNR: 18.63 -> 19.18 | Improved |
| curate clean data instead of using all data (ablation) | Real3D · CO3D | PSNR: 18.79 -> 19.18 | Improved |
| replace multi-view real data (MVImgNet) with single-view self-training on MVImgNet eval | LRM · MVImgNet | PSNR gain: 0.410 -> 0.720 | Improved |
| self-training instead of multi-view real data | LRM · WildImages | PSNR gain: 0.260 -> 0.720 | Improved |
| add Real3D self-training to TripoSR | TripoSR · MVImgNet (novel views) | PSNR: 19.81 -> 20.53 | Improved |
| add Real3D self-training to TripoSR | TripoSR · mesh quality | Chamfer-L1: 0.321 -> 0.275 | Improved |
Closest papers by meaning
- LRM-Zero: Training Large Reconstruction Models with Synthesized Data (NeurIPS 2024 · similarity 0.91)
- LRM: Large Reconstruction Model for Single Image to 3D (ICLR 2024 · similarity 0.91)
- MegaSynth: Scaling Up 3D Scene Reconstruction with Synthesized Data (CVPR 2025 · similarity 0.90)
- Large Spatial Model: End-to-end Unposed Images to Semantic 3D (NeurIPS 2024 · similarity 0.89)
- GTR: Improving Large 3D Reconstruction Models through Geometry and Texture Refinement (ICLR 2025 · similarity 0.88)
- LIRM: Large Inverse Rendering Model for Progressive Reconstruction of Shape, Materials and View-dependent Radiance Fields (CVPR 2025 · similarity 0.88)
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.
