Research map / Image generation / Fast diffusion sampling
GeodesicNVS: Probability Density Geodesic Flow Matching for Novel View Synthesis
Xuqin Wang, Tao Wu, Yanfeng Zhang, Lu Liu, Mingwei Sun, Yongliang Wang, Niclas Zeller, Daniel Cremers
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 |
|---|---|---|---|
| replace noise-to-data with data-to-data coupling (linear interpolant) | flow matching NVS · Objaverse (100 NFE) | FID: 5.5093 -> 5.4324 | Improved |
| replace noise-to-data with data-to-data coupling | flow matching NVS · GSO30 (100 NFE) | FID: 15.2763 -> 15.0543 (CLIP-S 89.8226 -> 89.5713) | Hurt |
| replace noise-to-data with data-to-data coupling at 10 NFE | flow matching NVS · Objaverse (10 NFE) | FID: 5.7768 -> 5.8223 | No clear effect |
| replace linear interpolant with probability density geodesic interpolant | Linear-D2D-FM · Objaverse (reduced training, LVIS subset) | FID: 11.8124 -> 10.4010 | Improved |
| replace linear interpolant with probability density geodesic interpolant | Linear-D2D-FM · Objaverse (reduced training, LVIS subset) | CLIP-S: 94.3476 -> 92.3368 | Hurt |
| replace learned-metric Metric FM with density geodesic | Metric FM · Objaverse (reduced training, LVIS subset) | FID: 11.9609 -> 10.4010 | Improved |
| optimize geodesic path (vs DDIM-initialized path) (ablation) | geodesic interpolant · Interpolant path study (6 objects) | AOFM: 6.4773 -> 13.6968 | Improved |
| use geodesic instead of linear interpolant | interpolants · Interpolant path study (6 objects) | AOFM: 1.0411 -> 13.6968 | Improved |
Closest papers by meaning
- MoCam: Unified Novel View Synthesis via Structured Denoising Dynamics (ECCV 2026 · similarity 0.89)
- Metric Flow Matching for Smooth Interpolations on the Data Manifold (NeurIPS 2024 · similarity 0.87)
- GenRec: Knowing Where to Reconstruct and Where to Generate (NeurIPS 2026 · similarity 0.87)
- Rectifying Categorical Flows on Statistical Manifolds for One-Step Generation (NeurIPS 2026 · similarity 0.86)
- Around the World in 80 Timesteps: A Generative Approach to Global Visual Geolocation (CVPR 2025 · similarity 0.86)
- MultiDiff: Consistent Novel View Synthesis from a Single Image (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.
