Research map / 3D vision / NeRF and radiance fields
LightSpeed: Light and Fast Neural Light Fields on Mobile Devices
Aarush Gupta, Junli Cao, Chaoyang Wang, Ju Hu, Sergey Tulyakov, Jian Ren, Laszlo Attila Jeni
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 |
|---|---|---|---|
| Use a ray-space grid encoder instead of frequency encoding (ablation) | LightSpeed · Lego sub-scene, 15-layer decoder | PSNR: 28.84 -> 30.37 | Improved |
| Use a ray-space grid encoder instead of frequency encoding (ablation) | LightSpeed · Lego sub-scene, 30-layer decoder | PSNR: 30.63 -> 31.70 | Improved |
| Use a ray-space grid encoder instead of frequency encoding (ablation) | LightSpeed · Lego sub-scene, 60-layer decoder | PSNR: 32.16 -> 32.34 | Improved |
| Represent rays with light-slab coordinates and grids instead of Plücker coordinates (ablation) | LightSpeed · Lego sub-scene, 15-layer decoder | PSNR: 28.65 -> 30.37 | Improved |
| Represent rays with light-slab coordinates and grids instead of Plücker coordinates (ablation) | LightSpeed · Lego sub-scene, 30-layer decoder | PSNR: 30.84 -> 31.70 | Improved |
| Represent rays with light-slab coordinates and grids instead of Plücker coordinates (ablation) | LightSpeed · Lego sub-scene, 60-layer decoder | PSNR: 32.14 -> 32.34 | Improved |
| Increase pseudo-data from 2k to 6k samples per Lego sub-scene (ablation) | LightSpeed · Lego sub-scene | PSNR: 30.44 -> 31.2 | Improved |
| Use a splitting plane at radius/√2 rather than radius/√3 (ablation) | LightSpeed · Lego sub-scene | PSNR: 30.23 -> 30.44 | Improved |
| Train the five scene partitions in parallel | LightSpeed · Lego | training duration: 75 hours -> 15 hours | Improved |
| Use LightSpeed rather than MobileR2L | Neural light field rendering · Synthetic 360° | PSNR: 31.34 -> 32.23 | Improved |
| Use LightSpeed rather than MobileR2L | Neural light field rendering · Forward-facing | PSNR: 26.15 -> 26.50 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MobileNeRF — MobileNeRF: Exploiting the Polygon Rasterization Pipeline for Efficient Neural Field Rendering on Mobile Architectures (CVPR 2023) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning Neural Light Fields With Ray-Space Embedding (CVPR 2022 · similarity 0.90)
- Light Field Neural Rendering (CVPR 2022 · similarity 0.90)
- A Pre-convolved Representation for Plug-and-Play Neural Illumination Fields (AAAI 2024 · similarity 0.87)
- NeLF-Pro: Neural Light Field Probes for Multi-Scale Novel View Synthesis (CVPR 2024 · similarity 0.87)
- VideoRF: Rendering Dynamic Radiance Fields as 2D Feature Video Streams (CVPR 2024 · similarity 0.86)
- Neural Inverse Rendering from Propagating Light (CVPR 2025 · 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.
