Research map / 3D vision / NeRF and radiance fields
NGP-RT: Fusing Multi-Level Hash Features with Lightweight Attention for Real-Time Novel View Synthesis
Yubin Hu, Xiaoyang Guo, Yang Xiao, Jingwei Huang, Yong-Jin Liu
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 SUM feature aggregation with lightweight separate-attention (spatially variant) (ablation) | NGP-RT (L=4) · Mip-NeRF 360 | PSNR: 25.51 -> 26.05 | Improved |
| replace MLP aggregation with lightweight attention (ablation) | NGP-RT (L=4) · Mip-NeRF 360 | FPS: 14.7 -> 61.8 FPS; PSNR 26.22 -> 26.05 | Improved |
| make attention parameters spatially variant instead of invariant (ablation) | NGP-RT (L=4) · Mip-NeRF 360 | PSNR: 25.73 -> 26.05 | Improved |
| use separate attention parameters for density and color instead of shared (ablation) | NGP-RT (L=4) · Mip-NeRF 360 | PSNR: 25.89 -> 26.05 | Improved |
| increase fine-grained hash levels from L=2 to L=4 (ablation) | NGP-RT · Mip-NeRF 360 | PSNR / FPS: PSNR 25.64 -> 26.05; FPS 108 -> 61.8 | Mixed |
| add pre-computed occupancy distance grid to ray marching (ablation) | NGP-RT · Mip-NeRF 360 | render time: 9.98 -> 9.26 ms (L=2); marching points 85.1 -> 46.7 | Improved |
| occupancy distance grid effect on image quality (ablation) | NGP-RT · Mip-NeRF 360 | PSNR: 25.67 -> 25.64 (L=2) | No clear effect |
| use multi-level hash features instead of MERF tri-plane features (outdoor scenes) | NGP-RT · Mip-NeRF 360 outdoor scenes | PSNR: 23.19 -> 22.76 | Hurt |
| use multi-level hash features instead of MERF tri-plane features (indoor scenes) | NGP-RT · Mip-NeRF 360 indoor scenes | PSNR: 27.80 -> 29.25 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Zip-NeRF — Zip-NeRF: Anti-Aliased Grid-Based Neural Radiance Fields (ICCV 2023) · better on 0 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
- KiloNeRF: Speeding Up Neural Radiance Fields With Thousands of Tiny MLPs (ICCV 2021 · similarity 0.85)
- ASH: A Modern Framework for Parallel Spatial Hashing in 3D Perception (TPAMI 2022 · similarity 0.84)
- HashPoint: Accelerated Point Searching and Sampling for Neural Rendering (CVPR 2024 · similarity 0.83)
- XScale-NVS: Cross-Scale Novel View Synthesis with Hash Featurized Manifold (CVPR 2024 · similarity 0.83)
- 4K4D: Real-Time 4D View Synthesis at 4K Resolution (CVPR 2024 · similarity 0.82)
- LinPrim: Linear Primitives for Differentiable Volumetric Rendering (NeurIPS 2025 · similarity 0.82)
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.
