Research map / 3D vision / NeRF and radiance fields
Inverse rendering with neural fields
Recovers geometry, reflectance and lighting from multi-view images for relighting and view synthesis.
110 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2020–2026) · an approach in NeRF and radiance fields, keyword cluster “rendering · field · synthesis”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Few recent reports The tracked benchmarks have at most one report from 2025–26 that could beat the old record, too few to say whether they are still moving.
48 papers in 2023–24 and 33 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 18.0% to 36.7%.
Open questions the data points to
- Papers report NeRF on Real Forward-Facing (LPIPS) anywhere from 0.199 to 0.25: comparisons against it depend on whose number you take.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2020 · A Neural Rendering Framework for Free-Viewpoint Relighting (CVPR 2020 · 47 citations)
- 2021 · IBRNet: Learning Multi-View Image-Based Rendering (CVPR 2021 · 722 citations)
- 2022 · Neural 3D Video Synthesis From Multi-View Video (CVPR 2022 · 380 citations)
- 2023 · HexPlane: A Fast Representation for Dynamic Scenes (CVPR 2023 · 425 citations)
- 2024 · Structure-Aware Sparse-View X-ray 3D Reconstruction (CVPR 2024 · 77 citations)
- 2025 · Sparse Voxels Rasterization: Real-time High-fidelity Radiance Field Rendering (CVPR 2025 · 16 citations)
- 2026 · PEERING INTO THE UNKNOWN: ACTIVE VIEW SELECTION WITH NEURAL UNCERTAINTY MAPS FOR 3D RECONSTRUCTION (ICLR 2026 · 0 citations)
Earliest papers here
- A Neural Rendering Framework for Free-Viewpoint Relighting (CVPR 2020 · 47 citations)
- Neural Sparse Voxel Fields (NeurIPS 2020 · 11 citations)
- IBRNet: Learning Multi-View Image-Based Rendering (CVPR 2021 · 722 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
NeRF Synthetic · PSNR (higher is better)
5 papers, 2022–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 32.32 | DIVeR: Real-Time and Accurate Neural Radiance Fields With Deterministic Integration for Volume Rendering (CVPR 2022 · 58 citations) |
| 2023 | 35.02 | Multiscale Tensor Decomposition and Rendering Equation Encoding for View Synthesis (CVPR 2023 · 16 citations) |
Tanks & Temples · PSNR (higher is better)
5 papers, 2022–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 28.46 | NeurMiPs: Neural Mixture of Planar Experts for View Synthesis (CVPR 2022 · 23 citations) |
| 2023 | 29.05 | Multiscale Tensor Decomposition and Rendering Equation Encoding for View Synthesis (CVPR 2023 · 16 citations) |
Tanks & Temples · LPIPS (lower is better)
4 papers, 2022–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 0.089 | NeurMiPs: Neural Mixture of Planar Experts for View Synthesis (CVPR 2022 · 23 citations) |
What has been tried
92 papers here report 913 measured changes (a component added, removed or swapped, compared with a baseline): helped 75%, no clear effect 5%, hurt 17%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add LS-MSA to the baseline | X3D NVS average · PSNR | 37.97 -> 47.95 | Helped | Structure-Aware Sparse-View X-ray 3D Reconstruction (CVPR 2024 · 77 citations) |
| Add Gaussian encoding initialization | Eyeful Tower dataset · PSNR | 32.52 -> 32.58 | Helped | SpecNeRF: Gaussian Directional Encoding for Specular Reflections (CVPR 2024 · 19 citations) |
| use spatially adaptive β(x) instead of a learned global β | Eyeful Tower · FPS | 28.79 -> 45.78 | Helped | HybridNeRF: Efficient Neural Rendering via Adaptive Volumetric Surfaces (CVPR 2024 · 17 citations) |
| Use error rather than uncertainty as the information signal, with the vMF sampler and relaxation | TanksAndTemples · PSNR | Helped | NeRF Director: Revisiting View Selection in Neural Volume Rendering (CVPR 2024 · 17 citations) | |
| use adaptive voxel sizes instead of uniform 512³ voxels | Mip-NeRF360 bonsai and bicycle · LPIPS | 0.326 -> 0.200 | Helped | Sparse Voxels Rasterization: Real-time High-fidelity Radiance Field Rendering (CVPR 2025 · 16 citations) |
| Add cubemap far-field features rather than analytical encoding | Materials · PSNR | 28.54 -> 30.27 | Helped | Neural Directional Encoding for Efficient and Accurate View-Dependent Appearance Modeling (CVPR 2024 · 14 citations) |
| Add sequence-level appearance embeddings | Argoverse 2 residential, single sequence · PSNR | 19.70 -> 22.22 | Helped | Multi-Level Neural Scene Graphs for Dynamic Urban Environments (CVPR 2024 · 10 citations) |
| remove the self-consistency loss Lcon | 6 synthetic scenes, far-field + flashlight · albedo PSNR | 27.70 -> 28.05 | Hurt | VMINer: Versatile Multi-view Inverse Rendering with Near- and Far-field Light Sources (CVPR 2024 · 10 citations) |
| add Stage 2 refinement after Stage 1 | DeepFashion3D #1 · Chamfer distance | 7.22 -> 4.55 | Helped | 2S-UDF: A Novel Two-stage UDF Learning Method for Robust Non-watertight Model Reconstruction from Multi-view Images (CVPR 2024 · 9 citations) |
| use geometry field splatting instead of original 2DGS splatting | DTU · Chamfer Distance | 0.70 -> 0.63 | Helped | Geometry Field Splatting with Gaussian Surfels (CVPR 2025 · 7 citations) |
| replace the learned kernel with an interpolation kernel | 2D image fitting · PSNR | 56.19 -> 48.12 | Hurt | Grounding and Enhancing Grid-based Models for Neural Fields (CVPR 2024 · 7 citations) |
| Add Personalized Fields during pre-training | 5s self-reconstruction ablation · Sync-C | 2.019 -> 4.816 | Helped | InsTaG: Learning Personalized 3D Talking Head from Few-Second Video (CVPR 2025 · 6 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| NeRF | Real Forward-Facing | LPIPS | 0.199 – 0.25 | 2 papers, e.g. CVPR 2022 and CVPR 2021 |
| 3DGS | Tanks&Temples | PSNR | 22.22 – 23.75 | 2 papers, e.g. CVPR 2024 and CVPR 2025 |
What reviewers will expect
Datasets and baselines named in the 23 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Mip-NeRF 360 · 4 of 23 papers
- ScanNet · 3 of 23 papers
- DTU · 3 of 23 papers
- Replica · 2 of 23 papers
- Stanford-ORB · 2 of 23 papers
- Tanks&Temples · 2 of 23 papers
- Deep Blending · 2 of 23 papers
- Objaverse · 2 of 23 papers
Baselines
- 3DGS · 5 of 23 papers
- Neuralangelo · 4 of 23 papers
- Zip-NeRF · 4 of 23 papers
- 2DGS · 3 of 23 papers
- NeRF · 3 of 23 papers
- NeuS · 3 of 23 papers
- SuGaR · 2 of 23 papers
- GS-LRM · 2 of 23 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 81 papers; from OpenAlex and the papers' affiliations).
- Zhejiang University · 6 papers
- Google · 5 papers
- Meta · 4 papers
- Carnegie Mellon University · 4 papers
- University of Toronto · 3 papers
- Peking University · 3 papers
- Adobe · 3 papers
- Hong Kong University of Science and Technology · 3 papers
- Nanjing University · 2 papers
- National University of Defense Technology · 2 papers
Use it
Download BibTeX of 39 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Few-shot neural rendering architectures (53 papers)
- 3D-aware generative radiance fields (20 papers)
Elsewhere in the topic
- Sparse-input NeRF regularization (106 papers)
- Dynamic and editable NeRF (88 papers)
- Single-view and streaming radiance fields (49 papers)
- NeRF for scene decomposition and understanding (30 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).