Research map / 3D vision / 3D generation
Repurposing Geometric Foundation Models for Multi-view Diffusion
Wooseok Jang, Seonghu Jeon, Jisang Han, Jinhyeok Choi, Minkyung Kwon, Seungryong Kim, Saining Xie, Sainan 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 |
|---|---|---|---|
| synthesize up to level 1 instead of level 0 only (ablation) | GLD · Re10K/DL3DV boundary selection | PSNR: 12.55 -> 13.61 | Improved |
| synthesize up to level 1 instead of level 0 only (ablation) | GLD · Re10K/DL3DV boundary selection | AbsRel: 0.267 -> 0.191 | Improved |
| extend synthesis boundary from level 1 to level 2 (ablation) | GLD · Re10K/DL3DV boundary selection | PSNR: 13.61 -> 13.35 | Hurt |
| extend synthesis boundary from level 1 to level 3 (ablation) | GLD · Re10K/DL3DV boundary selection | AbsRel: 0.191 -> 0.260 | Hurt |
| cascade level-0 generation on level-1 output instead of independent (ablation) | GLD · RealEstate10K (N=4) | PSNR: 18.81 -> 19.00 | Improved |
| cascade level-0 generation on level-1 output instead of independent (ablation) | GLD · RealEstate10K (N=4) | ATE: 0.197 -> 0.182 | Improved |
| use DA3 geometric latent instead of VAE latent (ablation) | GLD · DL3DV (N=1) | ATE: 0.880 -> 0.237 | Improved |
| use DA3 geometric latent instead of DINO latent (ablation) | GLD · Mip-NeRF 360 (N=2) | PSNR: 13.718 -> 14.542 | Improved |
| use DA3 geometric latent instead of VAE latent (ablation) | GLD · DL3DV (N=4) | PSNR: 16.37 -> 17.09 | Improved |
| train from scratch with GLD latent instead of MVGenMaster from scratch (ablation) | GLD · Mip-NeRF 360 (out-of-domain, N=2, from scratch) | PSNR: 11.217 -> 14.542 | Improved |
| use DA3 decoder on DA3 features instead of DINO RAE decoder (ablation) | DA3 decoder · Re10K reconstruction | PSNR: 26.78 -> 35.41 | Improved |
| use VGGT backbone instead of DA3 (ablation) | GLD · Re10K/DL3DV | quality | Mixed |
Closest papers by meaning
- SweetDreamer: Aligning Geometric Priors in 2D diffusion for Consistent Text-to-3D (ICLR 2024 · similarity 0.91)
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.90)
- MVGenMaster: Scaling Multi-View Generation from Any Image via 3D Priors Enhanced Diffusion Model (CVPR 2025 · similarity 0.89)
- WildFusion: Learning 3D-Aware Latent Diffusion Models in View Space (ICLR 2024 · similarity 0.89)
- Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models (CVPR 2025 · similarity 0.88)
- LT3SD: Latent Trees for 3D Scene Diffusion (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.
