Research map / 3D vision / Face and shape
Fast Training of Neural Lumigraph Representations using Meta Learning
Alexander William Bergman, Petr Kellnhofer, Gordon Wetzstein
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 |
|---|---|---|---|
| Apply MetaNLR++ instead of NLR++ | NLR++ · DTU | time to 25dB PSNR: 3.2 min. -> 1.9 min. | Improved |
| Apply MetaNLR++ instead of NLR++ | NLR++ · DTU | time to 30dB PSNR: 37.4 min. -> 22.5 min. | Improved |
| Apply MetaNLR++ instead of NLR++ | NLR++ · DTU | Maximum PSNR: 31.02dB -> 30.57dB | Hurt |
| Apply MetaNLR++ instead of NLR++ | NLR++ · NLR dataset | PSNR after 30 minutes: 28.22 -> 31.12 | Improved |
| Apply MetaNLR++ instead of NLR++ | NLR++ · NLR dataset | LPIPS after 30 minutes: 0.145 -> 0.083 | Improved |
| Apply MetaNLR++ instead of NLR++ | NLR++ · NLR dataset | Convergence PSNR: 35.54 -> 37.55 | Improved |
| Apply MetaNLR++ instead of NLR++ | NLR++ · NLR dataset | Convergence LPIPS: 0.046 -> 0.034 | Improved |
| Add meta initialization to all networks rather than no meta initialization (ablation) | CNN E/D with learned Γ · DTU, after 10 minutes | PSNR: 27.32dB -> 29.91dB | Improved |
| Add meta initialization to Φ only rather than no meta initialization (ablation) | CNN E/D with learned Γ · DTU, after 10 minutes | PSNR: 27.32dB -> 28.48dB | Improved |
| Add meta initialization to all networks rather than no meta initialization (ablation) | CNN E/D with fixed Γ · DTU, after 10 minutes | PSNR: 28.62dB -> 29.88dB | Improved |
| Add learned Γ instead of fixed Γ (ablation) | RGB pixel features without meta initialization · DTU, after 10 minutes | PSNR: 27.85dB -> 26.82dB | Hurt |
| Add learned Γ instead of fixed Γ (ablation) | CNN E/D without meta initialization · DTU, after 10 minutes | PSNR: 28.62dB -> 27.32dB | Hurt |
Closest papers by meaning
- Neural Lumigraph Rendering (CVPR 2021 · similarity 0.87)
- Light Field Networks: Neural Scene Representations with Single-Evaluation Rendering (NeurIPS 2021 · similarity 0.87)
- MetaSDF: Meta-Learning Signed Distance Functions (NeurIPS 2020 · similarity 0.87)
- Efficient-3Dim: Learning a Generalizable Single-image Novel-view Synthesizer in One Day (ICLR 2024 · similarity 0.85)
- SketchINR: A First Look into Sketches as Implicit Neural Representations (CVPR 2024 · similarity 0.84)
- MetaNeRV: Meta Neural Representations for Videos with Spatial-Temporal Guidance (AAAI 2025 · similarity 0.84)
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.
