Research map / 3D vision / Depth and stereo
Foundation models for monocular geometry
Estimates geometry from single images with inductive biases, open-domain training and synthetic-real adaptation.
81 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Depth and stereo, keyword cluster “depth estimation · monocular · depth map”
State History Tried Baselines Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
14 papers in 2023–24 and 40 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 4.6% to 9.8%.
Open questions the data points to
- Papers report DUSt3R on DTU (rel) anywhere from 3.3 to 3.9: 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
- 2016 · A Hole Filling Approach Based on Background Reconstruction for View Synthesis in 3D Video (CVPR 2016 · 63 citations)
- 2017 · Unsupervised Learning of Depth and Ego-Motion From Video (CVPR 2017 · 2,885 citations)
- 2018 · CodeSLAM — Learning a Compact, Optimisable Representation for Dense Visual SLAM (CVPR 2018 · 401 citations)
- 2019 · Pushing the Boundaries of View Extrapolation With Multiplane Images (CVPR 2019 · 348 citations)
- 2020 · RoutedFusion: Learning Real-Time Depth Map Fusion (CVPR 2020 · 88 citations)
- 2021 · TransformerFusion: Monocular RGB Scene Reconstruction using Transformers (NeurIPS 2021 · 81 citations)
- 2022 · BNV-Fusion: Dense 3D Reconstruction Using Bi-Level Neural Volume Fusion (CVPR 2022 · 36 citations)
- 2023 · HairStep: Transfer Synthetic to Real Using Strand and Depth Maps for Single-View 3D Hair Modeling (CVPR 2023 · 30 citations)
- 2024 · SelfOcc: Self-Supervised Vision-Based 3D Occupancy Prediction (CVPR 2024 · 65 citations)
- 2025 · VGGT: Visual Geometry Grounded Transformer (CVPR 2025 · 410 citations)
- 2026 · Depth Anything 3: Recovering the Visual Space from Any Views (ICLR 2026 · 2 citations)
Earliest papers here
- A Hole Filling Approach Based on Background Reconstruction for View Synthesis in 3D Video (CVPR 2016 · 63 citations)
- Multiview Image Completion With Space Structure Propagation (CVPR 2016 · 29 citations)
- Real-Time Depth Refinement for Specular Objects (CVPR 2016 · 20 citations)
What has been tried
48 papers here report 480 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 5%, hurt 16%, 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 |
|---|---|---|---|---|
| replace alternating attention with cross-attention | ETH3D · Overall | 0.709 -> 1.061 | Hurt | VGGT: Visual Geometry Grounded Transformer (CVPR 2025 · 410 citations) |
| replace affine-invariant point maps with SI-Log depth and a dense ray map | Ablation evaluation (ViT-Base; local point map) · Relp | 7.16 -> 9.19 | Hurt | MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training Supervision (CVPR 2025 · 69 citations) |
| Add L_mvs to the density-field architecture | nuScenes occupancy · IoU | 18.90 -> 36.68 | Helped | SelfOcc: Self-Supervised Vision-Based 3D Occupancy Prediction (CVPR 2024 · 65 citations) |
| add per-pixel ray direction as input | Sintel · mean angular error | 36.6 -> 36.0 | Helped | Rethinking Inductive Biases for Surface Normal Estimation (CVPR 2024 · 58 citations) |
| decouple metric scale from relative geometry using a CLS-token-conditioned MLP instead of entangled SI-Log prediction | 10-dataset ablation average · metric point Relp | 10.0 -> 9.20 | Helped | MoGe-2: Accurate Monocular Geometry with Metric Scale and Sharp Details (NeurIPS 2025 · 27 citations) |
| Add Pseudo Building Bbox Calculator (PBC) | BN30/70 · F1-score | 61.35 -> 62.32 | Helped | 3D Building Reconstruction from Monocular Remote Sensing Images with Multi-level Supervisions (CVPR 2024 · 14 citations) |
| Add the X2,2 pointmap prediction head | Habitat · Avg. | 77.7 -> 78.3 | Helped | Pow3R: Empowering Unconstrained 3D Reconstruction with Camera and Scene Priors (CVPR 2025 · 13 citations) |
| add depth-map joint training to incident-map estimation | Waymo · e_f | 0.188 -> 0.145 | Helped | DiffCalib: Reformulating Monocular Camera Calibration as Diffusion-Based Dense Incident Map Generation (AAAI 2025 · 5 citations) |
| Vary Dmin from 15 to 45 | Natural image test set · MSE | Helped | Mask-Homo: Pseudo Plane Mask-Guided Unsupervised Multi-Homography Estimation (AAAI 2024 · 5 citations) | |
| Fine-tune on Occ3D occupancy GT | nuScenes validation · MAE | 1.93 -> 1.89 | Helped | Accurate Training Data for Occupancy Map Prediction in Automated Driving Using Evidence Theory (CVPR 2024 · 4 citations) |
| add iterative reweighting | complex shapes · discontinuity reconstruction | Helped | Discontinuity-preserving Normal Integration with Auxiliary Edges (CVPR 2024 · 2 citations) | |
| add sparse points | Tanks and Temples · F1-score | 0.397 -> 0.382 | Hurt | GeoSVR: Taming Sparse Voxels for Geometrically Accurate Surface Reconstruction (NeurIPS 2025 · 1 citation) |
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 |
|---|---|---|---|---|
| DUSt3R | DTU | rel | 3.3 – 3.9 | 2 papers, e.g. NeurIPS 2025 and CVPR 2025 |
What reviewers will expect
Datasets and baselines named in the 15 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Hypersim · 5 of 15 papers
- ARKitScenes · 5 of 15 papers
- ScanNet++ · 5 of 15 papers
- KITTI · 3 of 15 papers
- NYUv2 · 3 of 15 papers
- Waymo · 3 of 15 papers
- ETH3D · 3 of 15 papers
- MegaDepth · 3 of 15 papers
Baselines
- DUSt3R · 5 of 15 papers
- MASt3R · 5 of 15 papers
- VGGT · 4 of 15 papers
- Depth Pro · 3 of 15 papers
- UniDepth V2 · 3 of 15 papers
- WildCamera · 2 of 15 papers
- LeReS · 2 of 15 papers
- Metric3D V2 · 2 of 15 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 54 papers; from OpenAlex and the papers' affiliations).
- University of Science and Technology of China · 4 papers
- Tsinghua University · 4 papers
- Zhejiang University · 3 papers
- Microsoft · 3 papers
- Zhejiang University of Technology · 2 papers
- Hong Kong University of Science and Technology · 2 papers
- Wuhan University · 2 papers
Use it
Download BibTeX of 36 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
- Self-supervised monocular depth (116 papers)
- Dense depth prediction and completion (85 papers)
- Depth from light field and defocus (60 papers)
Elsewhere in the topic
- Feed-forward 3D reconstruction (133 papers)
- Scene coordinate regression for localization (112 papers)
- Learned optical and scene flow (108 papers)
- Neural light field and illumination imaging (107 papers)
- Cost-volume multi-view stereo (107 papers)
- Camera model and calibration for SfM (99 papers)
- Non-line-of-sight imaging (93 papers)
- Dense feature matching for correspondence (80 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).