Research map / 3D vision / Gaussian splatting
Lagrangian Hashing for Compressed Neural Field Representations
Shrisudhan Govindarajan, Zeno Sambugaro, Akhmedkhan Shabanov, Towaki Takikawa, Weiwei Sun, Daniel Rebain, Nicola Conci, Kwang Moo Yi, Andrea Tagliasacchi
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 |
|---|---|---|---|
| add Gaussian mixture (K=4) per hash bucket in finest levels (ablation) | InstantNGP hash grid · Tanks & Temples | PSNR: 27.49 -> 27.94 | Improved |
| increase Gaussians per bucket from 4 to 8 (ablation) | Lagrangian Hashing · Tanks & Temples | PSNR: 27.94 -> 27.99 | No clear effect |
| use 2 Lagrangian levels instead of none (ablation) | Lagrangian Hashing · Tanks & Temples | PSNR: 27.49 -> 27.94 | Improved |
| increase Lagrangian levels from 2 to 8 (ablation) | Lagrangian Hashing · Tanks & Temples | PSNR: 27.94 -> 28.03 | No clear effect |
| make Gaussian standard deviation learnable (ablation) | Lagrangian Hashing · Tanks & Temples | PSNR: 27.92 -> 27.86 | Hurt |
| use Lagrangian Hashing instead of InstantNGP at about half the parameters | NeRF Synthetic · NeRF Synthetic | PSNR: 33.11 -> 33.12 | No clear effect |
| use Lagrangian Hashing instead of InstantNGP at matched parameters | Gigapixel images · Gigapixel images (avg) | PSNR: 39.60 -> 40.71 | Improved |
| use Lagrangian Hashing instead of CompactNGP at 0.18M parameters | NeRF Synthetic · NeRF Synthetic | PSNR: 30.66 -> 30.72 | Improved |
| use Lagrangian Hashing instead of 3DGS at 0.18M parameters | NeRF Synthetic · NeRF Synthetic | PSNR: 19.91 -> 30.72 | Improved |
| use Lagrangian Hashing instead of 3DGS at ~12M parameters | NeRF Synthetic · NeRF Synthetic | PSNR: 33.73 vs 33.25 | Hurt |
Closest papers by meaning
- Neural Fields as Distributions: Signal Processing Beyond Euclidean Space (CVPR 2024 · similarity 0.83)
- Rectified Lagrangian for Out-of-Distribution Detection in Modern Hopfield Networks (AAAI 2025 · similarity 0.82)
- Preconditioners for the Stochastic Training of Neural Fields (CVPR 2025 · similarity 0.82)
- LORE: Lagrangian-Optimized Robust Embeddings for Visual Encoders (NeurIPS 2025 · similarity 0.82)
- Neural Physical Simulation with Multi-Resolution Hash Grid Encoding (AAAI 2024 · similarity 0.82)
- A Global Geometric Analysis of Maximal Coding Rate Reduction (ICML 2024 · 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.
