Research map / 3D vision / NeRF and radiance fields
KiloNeRF: Speeding Up Neural Radiance Fields With Thousands of Tiny MLPs
Christian Reiser, Songyou Peng, Yiyi Liao, Andreas Geiger
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 one large NeRF MLP by thousands of tiny MLPs (with ESS and ERT already applied) (ablation) | NeRF · Lego bulldozer | render time (ms): 788 -> 22 | Improved |
| add empty space skipping and early ray termination to NeRF (ablation) | NeRF · Lego bulldozer | render time (ms): 56185 -> 788 | Improved |
| replace NeRF with KiloNeRF | NeRF · Synthetic-NeRF | render time (ms): 56185 -> 26 | Improved |
| replace NSVF with KiloNeRF | NSVF · Synthetic-NeRF | render time (ms): 4344 -> 26 | Improved |
| replace NeRF with KiloNeRF | NeRF · Synthetic-NeRF | PSNR: 31.01 -> 31.00 | No clear effect |
| replace NSVF with KiloNeRF | NSVF · Synthetic-NSVF | PSNR: 35.13 -> 33.37 | Hurt |
| use thousands of tiny MLPs instead of a single tiny MLP (ablation) | KiloNeRF · Lego bulldozer | LPIPS | Improved |
| use full grid resolution 10x16x10 instead of half 5x8x5 (ablation) | KiloNeRF · Lego bulldozer | LPIPS | Improved |
| use full number of hidden units instead of half (ablation) | KiloNeRF · Lego bulldozer | LPIPS | Improved |
| add final fine-tuning phase (ablation) | KiloNeRF · Lego bulldozer | LPIPS | Improved |
| add distillation from a pretrained NeRF (ablation) | KiloNeRF · Lego bulldozer | image artifacts | Improved |
| add L2 weight regularization on last two layers (ablation) | KiloNeRF · Lego bulldozer | image artifacts | Improved |
Later papers that compared with this method
5 later accepted papers report KiloNeRF as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Progressive Volume Distillation with Active Learning for Efficient NeRF Architecture Conversion (IJCV 2026 · better on 2 of 2 shared results)
- Learning Neural Duplex Radiance Fields for Real-Time View Synthesis (CVPR 2023 · better on 6 of 8 shared results)
- NeurMiPs: Neural Mixture of Planar Experts for View Synthesis (CVPR 2022 · better on 3 of 4 shared results)
- DIVeR: Real-Time and Accurate Neural Radiance Fields With Deterministic Integration for Volume Rendering (CVPR 2022 · better on 0 of 1 shared result)
- R2L: Distilling Neural Radiance Field to Neural Light Field for Efficient Novel View Synthesis (ECCV 2022 · better on 1 of 1 shared result)
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
- NeRFLight: Fast and Light Neural Radiance Fields Using a Shared Feature Grid (CVPR 2023 · similarity 0.92)
- PyNeRF: Pyramidal Neural Radiance Fields (NeurIPS 2023 · similarity 0.92)
- Learning Neural Duplex Radiance Fields for Real-Time View Synthesis (CVPR 2023 · similarity 0.91)
- Ced-NeRF: A Compact and Efficient Method for Dynamic Neural Radiance Fields (AAAI 2024 · similarity 0.90)
- NeRF-VPT: Learning Novel View Representations with Neural Radiance Fields via View Prompt Tuning (AAAI 2024 · similarity 0.89)
- VideoRF: Rendering Dynamic Radiance Fields as 2D Feature Video Streams (CVPR 2024 · 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.
