Research map / 3D vision / Depth and stereo
Light Field Networks: Neural Scene Representations with Single-Evaluation Rendering
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Represent the scene as a neural light field with one network evaluation per ray rather than volumetric rendering | LFN · 256×256 image rendering | clock time (ms): 2.1 vs 120 (SRN); 2.1 vs $30e3 (pixelNeRF) | Helped |
| Represent a 360-degree light field with an LFN instead of a six-plane Lumigraph | LFN · 360-degree light field storage at 256×256×17×17 | storage (MB): 1.6 MB vs 146 MB | Helped |
| Use LFN instead of globally conditioned SRN or DVR | single-shot multi-class reconstruction · ShapeNet 13 classes single-shot | PSNR (mean): LFN 24.95; SRN 23.28; DVR 22.70 | Helped |
| Use LFN instead of pixelNeRF | single-shot novel view synthesis · ShapeNet single-class | PSNR: LFNs perform 1dB worse than pixelNeRF | Hurt |
| Use LFN instead of pixelNeRF | single-shot novel view synthesis · ShapeNet multi-class | PSNR: LFNs perform 2dB worse than pixelNeRF | Hurt |
| Use LFN instead of SRN | class-specific single-shot reconstruction · ShapeNet cars | PSNR | Helped |
| Use LFN instead of SRN | class-specific single-shot reconstruction · ShapeNet chairs | PSNR | Hurt |
| Use LFN instead of SRN | sparse reconstructed geometry · ShapeNet cars and chairs | mean L1 depth error | Helped |
| Train the light-field prior with fewer scene examples | LFN multi-view consistency · ShapeNet | multi-view consistency | Hurt |
Closest papers by meaning
- Learning Neural Light Fields With Ray-Space Embedding (CVPR 2022 · similarity 0.92)
- Light Field Neural Rendering (CVPR 2022 · similarity 0.92)
- Neural Point Light Fields (CVPR 2022 · similarity 0.91)
- NeLF-Pro: Neural Light Field Probes for Multi-Scale Novel View Synthesis (CVPR 2024 · similarity 0.89)
- DyLiN: Making Light Field Networks Dynamic (CVPR 2023 · similarity 0.89)
- S$^3$-NeRF: Neural Reflectance Field from Shading and Shadow under a Single Viewpoint (NeurIPS 2022 · similarity 0.89)
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.