Research map / 3D vision / NeRF and radiance fields
Mega-NERF: Scalable Construction of Large-Scale NeRFs for Virtual Fly-Throughs
Haithem Turki, Deva Ramanan, Mahadev Satyanarayanan
Check 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 spatially partitioned NeRF submodules with geometry-aware pixel assignment | Mega-NeRF · Mill 19 - Rubble | PSNR | Improved |
| Add per-image appearance embeddings (ablation) | Mega-NeRF · Mill 19, Quad 6k, UrbanScene3D | PSNR: 20.42 -> 22.34; 16.16 -> 18.08; 19.45 -> 23.60 | Improved |
| Use ellipsoid foreground bounds and altitude-based ray limits instead of NeRF++ unit-sphere bounds (ablation) | Mega-NeRF · Mill 19, Quad 6k, UrbanScene3D | PSNR: 22.14 -> 22.34; 18.02 -> 18.08; 23.42 -> 23.60 | Improved |
| Use spatially-aware sparse connections instead of fully connected layers (ablation) | Mega-NeRF · Mill 19, UrbanScene3D | PSNR: 21.63 -> 22.34; 22.44 -> 23.60 | Improved |
| Train submodules on their respective pixel partitions instead of jointly on the full dataset (ablation) | Mega-NeRF · Mill 19, Quad 6k, UrbanScene3D | PSNR: 21.10 -> 22.34; 17.43 -> 18.08; 21.45 -> 23.60 | Improved |
| Prune pixel assignments using estimated ray intersections with surfaces (ablation) | Mega-NeRF · Mill 19, Quad 6k, UrbanScene3D | PSNR: 22.50 -> 22.76; 18.13 -> 18.16; 23.65 -> 23.87 | Improved |
| Increase the number of submodules from 1 to 16 at fixed channel count (ablation) | Mega-NeRF · Mill 19 - Rubble | PSNR: 21.75 -> 23.34; 22.60 -> 24.52; 23.40 -> 25.68 | Improved |
| Increase channels per submodule from 128 to 512 (ablation) | Mega-NeRF · Mill 19 - Rubble | PSNR: 21.75 -> 23.40; 22.61 -> 24.53; 23.08 -> 25.11; 23.34 -> 25.68 | Improved |
| Use Mega-NeRF-Dynamic temporal-coherence rendering instead of conventional full ray sampling | Mega-NeRF · Mill 19, Quad 6k, UrbanScene3D | render time (s): 101 -> 3.96; 174 -> 2.91; 122 -> 3.219 | Improved |
Later papers that compared with this method
5 later accepted papers report MegaNeRF as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- HUG: Hierarchical Urban Gaussian Splatting with Block-Based Reconstruction for Large-Scale Aerial Scenes (ICCV 2025 · better on 0 of 1 shared result)
- City-on-Web: Real-time Neural Rendering of Large-scale Scenes on the Web (ECCV 2024 · better on 1 of 2 shared results)
- UE4-NeRF:Neural Radiance Field for Real-Time Rendering of Large-Scale Scene (NeurIPS 2023 · better on 5 of 5 shared results)
- Progressively Optimized Local Radiance Fields for Robust View Synthesis (CVPR 2023 · better on 3 of 3 shared results)
- Grid-Guided Neural Radiance Fields for Large Urban Scenes (CVPR 2023 · better on 2 of 2 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
- Block-NeRF: Scalable Large Scene Neural View Synthesis (CVPR 2022 · similarity 0.91)
- NeRF On-the-go: Exploiting Uncertainty for Distractor-free NeRFs in the Wild (CVPR 2024 · similarity 0.89)
- A General Implicit Framework for Fast NeRF Composition and Rendering (AAAI 2024 · similarity 0.89)
- Rad-NeRF: Ray-decoupled Training of Neural Radiance Field (NeurIPS 2024 · similarity 0.89)
- Multiplane Prior Guided Few-Shot Aerial Scene Rendering (CVPR 2024 · similarity 0.89)
- NerfBaselines: Consistent and Reproducible Evaluation of Novel View Synthesis Methods (NeurIPS 2025 · similarity 0.88)
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.
