Research map / 3D vision / Depth and stereo
Cost-volume multi-view stereo
Infers depth from multiple views with learned cost volumes, recurrent regularization and geometry-aware networks.
107 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Depth and stereo, keyword cluster “view stereo · photometric stereo · stereo matching”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 2 of 6 tracked benchmarks.
25 papers in 2023–24 and 32 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 8.2% to 7.8%.
Open questions the data points to
- Tanks and Temples intermediate (Mean F-score): best went from 67.03 (up to 2024) to 68.16 in 2025 — RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · 7 citations)
- ETH3D test, 2cm (F1): best went from 88.06 (up to 2024) to 89.6 in 2025 — DVP-MVS: Synergize Depth-Edge and Visibility Prior for Multi-View Stereo (AAAI 2025 · 16 citations)
- Papers report MVSFormer++ on Tanks-and-Temples intermediate (Mean F-score) anywhere from 59.33 to 67.18: comparisons against it depend on whose number you take.
What looks settled
- Tanks-and-Temples advanced (Mean F-score): best up to 2024 44.82, best in 2025–26 43.32 from 2 papers, at most 0.2% better, if at all.
No longer reported by papers here: DTU evaluation set (Overall (mm), last in 2024); DTU (Overall, last in 2024); DTU evaluation (Overall (mm), last in 2023).
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 · Semantic 3D Reconstruction With Continuous Regularization and Ray Potentials Using a Visibility Consistency Constraint (CVPR 2016 · 54 citations)
- 2017 · A Multi-View Stereo Benchmark With High-Resolution Images and Multi-Camera Videos (CVPR 2017 · 913 citations)
- 2018 · Pyramid Stereo Matching Network (CVPR 2018 · 1,860 citations)
- 2019 · Group-Wise Correlation Stereo Network (CVPR 2019 · 768 citations)
- 2020 · Cascade Cost Volume for High-Resolution Multi-View Stereo and Stereo Matching (CVPR 2020 · 876 citations)
- 2021 · PatchmatchNet: Learned Multi-View Patchmatch Stereo (CVPR 2021 · 424 citations)
- 2022 · TransMVSNet: Global Context-Aware Multi-View Stereo Network With Transformers (CVPR 2022 · 264 citations)
- 2023 · Iterative Geometry Encoding Volume for Stereo Matching (CVPR 2023 · 343 citations)
- 2024 · MoCha-Stereo: Motif Channel Attention Network for Stereo Matching (CVPR 2024 · 86 citations)
- 2025 · MonSter: Marry Monodepth to Stereo Unleashes Power (CVPR 2025 · 53 citations)
- 2026 · Dens3R: A Foundation Model for 3D Geometry Prediction (ICLR 2026 · 1 citation)
Earliest papers here
- Semantic 3D Reconstruction With Continuous Regularization and Ray Potentials Using a Visibility Consistency Constraint (CVPR 2016 · 54 citations)
- Just Look at the Image: Viewpoint-Specific Surface Normal Prediction for Improved Multi-View Reconstruction (CVPR 2016 · 26 citations)
- A Multi-View Stereo Benchmark With High-Resolution Images and Multi-Camera Videos (CVPR 2017 · 913 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.
Tanks and Temples intermediate · Mean F-score (higher is better)
8 papers, 2020–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 54.83 | Deep Stereo Using Adaptive Thin Volume Representation With Uncertainty Awareness (CVPR 2020 · 376 citations) |
| 2022 | 64.36 | Rethinking Depth Estimation for Multi-View Stereo: A Unified Representation (CVPR 2022 · 173 citations) |
| 2024 | 67.03 | MVSFormer++: Revealing the Devil in Transformer's Details for Multi-View Stereo (ICLR 2024 · 5 citations) |
| 2025 | 68.16 | RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · 7 citations) |
DTU evaluation set · Overall (mm) (lower is better)
6 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 0.352 | PatchmatchNet: Learned Multi-View Patchmatch Stereo (CVPR 2021 · 424 citations) |
| 2022 | 0.305 | TransMVSNet: Global Context-Aware Multi-View Stereo Network With Transformers (CVPR 2022 · 264 citations) |
| 2023 | 0.297 | Multi-View Stereo Representation Revist: Region-Aware MVSNet (CVPR 2023 · 28 citations) |
Tanks-and-Temples advanced · Mean F-score (higher is better)
6 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 37 | TransMVSNet: Global Context-Aware Multi-View Stereo Network With Transformers (CVPR 2022 · 264 citations) |
| 2024 | 44.82 | GoMVS: Geometrically Consistent Cost Aggregation for Multi-View Stereo (CVPR 2024 · 45 citations) |
DTU · Overall (lower is better)
5 papers, 2020–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 0.351 | Cost Volume Pyramid Based Depth Inference for Multi-View Stereo (CVPR 2020 · 354 citations) |
| 2022 | 0.298 | Generalized Binary Search Network for Highly-Efficient Multi-View Stereo (CVPR 2022 · 93 citations) |
| 2024 | 0.287 | GoMVS: Geometrically Consistent Cost Aggregation for Multi-View Stereo (CVPR 2024 · 45 citations) |
DTU evaluation · Overall (mm) (lower is better)
5 papers, 2020–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 0.355 | Cascade Cost Volume for High-Resolution Multi-View Stereo and Stereo Matching (CVPR 2020 · 876 citations) |
| 2022 | 0.295 | WT-MVSNet: Window-based Transformers for Multi-view Stereo (NeurIPS 2022 · 17 citations) |
ETH3D test, 2cm · F1 (higher is better)
4 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 87.44 | Adaptive Patch Deformation for Textureless-Resilient Multi-View Stereo (CVPR 2023 · 39 citations) |
| 2024 | 88.06 | SD-MVS: Segmentation-Driven Deformation Multi-View Stereo with Spherical Refinement and EM Optimization (AAAI 2024 · 28 citations) |
| 2025 | 89.6 | DVP-MVS: Synergize Depth-Edge and Visibility Prior for Multi-View Stereo (AAAI 2025 · 16 citations) |
What has been tried
80 papers here report 835 measured changes (a component added, removed or swapped, compared with a baseline): helped 74%, no clear effect 3%, hurt 20%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add MCA to feature maps | Scene Flow test ablation · EPE (px) | 0.458 -> 0.449 | Helped | MoCha-Stereo: Motif Channel Attention Network for Stereo Matching (CVPR 2024 · 86 citations) |
| add MGR-based monocular-stereo disparity fusion instead of convolution-based fusion | Scene Flow test · EPE (px) | 0.46 -> 0.43 | Helped | MonSter: Marry Monodepth to Stereo Unleashes Power (CVPR 2025 · 53 citations) |
| replace GCA with standard 3D convolution | DTU · Overall | 0.287 -> 0.315 | Hurt | GoMVS: Geometrically Consistent Cost Aggregation for Multi-View Stereo (CVPR 2024 · 45 citations) |
| add LSGP | UnrealStereo4K overall · AvgErr | 18.81 -> 6.28 | Helped | LoS: Local Structure-Guided Stereo Matching (CVPR 2024 · 34 citations) |
| add rendering after causal prediction | TUM RGB ATE RMSE (11 sequences) · ATE RMSE [cm] | 6.0 -> 4.5 | Helped | MUSt3R: Multi-view Network for Stereo 3D Reconstruction (CVPR 2025 · 32 citations) |
| add matching-cost patch deformation and multi-scale consistency | ETH3D · F1 | 81.27 -> 86.94 | Helped | SD-MVS: Segmentation-Driven Deformation Multi-View Stereo with Spherical Refinement and EM Optimization (AAAI 2024 · 28 citations) |
| Add intra-image features | SceneFlow test · EPE (px) | 0.62 -> 0.54 | Helped | IINet: Implicit Intra-inter Information Fusion for Real-Time Stereo Matching (AAAI 2024 · 24 citations) |
| add the depth-edge aligned prior | ETH3D, 2cm · F1 | 86.53 -> 88.67 | Helped | DVP-MVS: Synergize Depth-Edge and Visibility Prior for Multi-View Stereo (AAAI 2025 · 16 citations) |
| Add edge aggregation and edge refinement | ETH3D 2cm · F1 | 86.53 -> 88.09 | Helped | MSP-MVS: Multi-Granularity Segmentation Prior Guided Multi-View Stereo (AAAI 2025 · 13 citations) |
| Replace fixed sampling with the Plane Indicator | DTU testing set · Overall (mm) | 0.342 -> 0.322 | Helped | Learning Deformable Hypothesis Sampling for Accurate PatchMatch Multi-View Stereo (AAAI 2024 · 12 citations) |
| add intra-view geometric loss | SceneFlow · EPE (px) | 0.765 -> 0.615 | Helped | Learning Intra-view and Cross-view Geometric Knowledge for Stereo Matching (CVPR 2024 · 8 citations) |
| Add R-SA | DTU test · Overall error (mm) | 0.318 -> 0.299 | Helped | RRT-MVS: Recurrent Regularization Transformer for Multi-View Stereo (AAAI 2025 · 7 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 |
|---|---|---|---|---|
| MVSFormer++ | Tanks-and-Temples intermediate | Mean F-score | 59.33 – 67.18 | 5 papers, e.g. CVPR 2024 and AAAI 2025 |
What reviewers will expect
Datasets and baselines named in the 14 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ETH3D · 4 of 14 papers
- ARKitScenes · 4 of 14 papers
- ScanNet++ · 4 of 14 papers
- BlendedMVS · 3 of 14 papers
- WildRGB-D · 3 of 14 papers
- CO3D · 3 of 14 papers
- Tanks & Temples · 2 of 14 papers
- Tanks and Temples · 2 of 14 papers
Baselines
- MVSTER · 4 of 14 papers
- PatchMatchNet · 3 of 14 papers
- EPNet · 3 of 14 papers
- ACMM · 3 of 14 papers
- ACMP · 3 of 14 papers
- ACMMP · 3 of 14 papers
- SD-MVS · 3 of 14 papers
- APD-MVS · 3 of 14 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 57 papers; from OpenAlex and the papers' affiliations).
- Huazhong University of Science and Technology · 4 papers
- Chinese Academy of Sciences · 4 papers
- Harbin Institute of Technology · 3 papers
- Chinese Academy of Agricultural Sciences · 3 papers
- Ministry of Agriculture and Rural Affairs · 3 papers
- Zhejiang University · 3 papers
- Wuhan University · 2 papers
- Nanjing University of Science and Technology · 2 papers
- University of Science and Technology Beijing · 2 papers
- The Chinese University of Hong Kong · 2 papers
Use it
Download BibTeX of 40 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
- Photometric stereo (43 papers)
- Learned stereo with specialized sensors (53 papers)
Elsewhere in the topic
- Feed-forward 3D reconstruction (133 papers)
- Self-supervised monocular depth (116 papers)
- Scene coordinate regression for localization (112 papers)
- Learned optical and scene flow (108 papers)
- Neural light field and illumination imaging (107 papers)
- Camera model and calibration for SfM (99 papers)
- Non-line-of-sight imaging (93 papers)
- Dense depth prediction and completion (85 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).