Depth and stereo: research map
2,574 papers on Depth and stereo in 3D vision, from ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR (2016–2026), grouped into 4 clusters and 16 approaches. The busiest year so far is 2026.
What it is
Recovering 3D geometry from images: depth from one image (monocular depth), from two cameras (stereo) or from many views, plus optical flow, camera calibration, localisation and special imaging such as light fields and event cameras.
Terms you will see on this page:
- Monocular depth estimation
- predicting a depth map from a single image.
- Multi-view stereo (MVS)
- reconstructing dense 3D geometry from several calibrated images.
- Optical flow
- the apparent motion of each pixel between two frames.
Start reading with the most cited papers here:
- Structure-From-Motion Revisited (CVPR 2016 · 6,291 citations)
- Unsupervised Monocular Depth Estimation With Left-Right Consistency (CVPR 2017 · 3,325 citations)
Within 3D vision, its share shrank from 17.2% in 2023–24 to 14.8% in 2025–26 (306 → 409 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Depth and stereo in the interactive map
Working on something in this topic? Describe your idea in scime atlas to see which approach it falls under, the closest papers by meaning and how crowded the spot has become.
What changed recently
Approaches whose share of the topic moved most between 2023–24 and 2025–26 (ICML, NeurIPS, ICLR, CVPR and AAAI):
- Foundation models for monocular geometry rising: 5% → 10% of the topic (14 → 40 papers)
- Feed-forward 3D reconstruction rising: 10% → 14% of the topic (32 → 58 papers)
- Learned stereo with specialized sensors rising: 2% → 5% of the topic (5 → 19 papers)
- Dense feature matching for correspondence falling: 8% → 3% of the topic (23 → 14 papers)
- Non-line-of-sight imaging falling: 9% → 5% of the topic (27 → 20 papers)
Papers from 2025–2026 that are already cited most:
- VGGT: Visual Geometry Grounded Transformer (CVPR 2025 · 410 citations)
- Fast3R: Towards 3D Reconstruction of 1000+ Images in One Forward Pass (CVPR 2025 · 70 citations)
- MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training Supervision (CVPR 2025 · 69 citations)
- MonSter: Marry Monodepth to Stereo Unleashes Power (CVPR 2025 · 53 citations)
Approaches and key papers
Flow, matching and localization · 973 papers
Keywords: motion, flow, localization
Approaches in this cluster:
- Feed-forward 3D reconstruction (210 papers)
Reconstructs scenes and structure-from-motion in a single pass with learned priors, planes and dense matching. - Camera model and calibration for SfM (220 papers)
Handles camera calibration, distortion and minimal geometric solvers in large-scale structure from motion. - Learned optical and scene flow (217 papers)
Estimates motion fields with unrolled, unsupervised or joint learning of flow, depth and ego-motion. - Scene coordinate regression for localization (205 papers)
Localizes cameras by regressing scene coordinates or maps, with voting and 3D surface prediction. - Dense feature matching for correspondence (121 papers)
Learns dense or two-view correspondences with kernelized matching, cycle consistency and confidence estimation.
Most cited and most cited since 2024:
- Structure-From-Motion Revisited (CVPR 2016 · 6,291 citations)
- SuperGlue: Learning Feature Matching With Graph Neural Networks (CVPR 2020 · 2,845 citations)
- DUSt3R: Geometric 3D Vision Made Easy (CVPR 2024 · 545 citations)
- SplaTAM: Splat Track & Map 3D Gaussians for Dense RGB-D SLAM (CVPR 2024 · 438 citations)
Depth estimation · 691 papers
Keywords: depth estimation, monocular, depth map
Approaches in this cluster:
- Self-supervised monocular depth (243 papers)
Learns depth from unlabeled images or video using geometric constraints, semantic guidance and planarity priors. - Dense depth prediction and completion (175 papers)
Predicts consistent, high-resolution or sparse-input-completed depth from single images and video. - Foundation models for monocular geometry (133 papers)
Estimates geometry from single images with inductive biases, open-domain training and synthetic-real adaptation. - Depth from light field and defocus (140 papers)
Estimates depth using light fields, defocus, event cues and interference instead of standard RGB.
Most cited and most cited since 2024:
- Unsupervised Monocular Depth Estimation With Left-Right Consistency (CVPR 2017 · 3,325 citations)
- Unsupervised Learning of Depth and Ego-Motion From Video (CVPR 2017 · 2,885 citations)
- VGGT: Visual Geometry Grounded Transformer (CVPR 2025 · 410 citations)
- Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation (CVPR 2024 · 388 citations)
Computational imaging and light · 505 papers
Keywords: light, imaging, illumination
Approaches in this cluster:
- Neural light field and illumination imaging (171 papers)
Recovers scene properties with learned light fields, structured light and material estimation. - Non-line-of-sight imaging (177 papers)
Reconstructs hidden or transparent geometry from time-of-flight light transport. - Spike and event camera reconstruction (96 papers)
Reconstructs high-speed scenes from spike, single-photon and neuromorphic sensor data. - Snapshot hyperspectral imaging (61 papers)
Reconstructs spectral images from coded or dispersive snapshots with learned or physics-guided models.
Most cited and most cited since 2024:
- DeepView: View Synthesis With Learned Gradient Descent (CVPR 2019 · 434 citations)
- EPINET: A Fully-Convolutional Neural Network Using Epipolar Geometry for Depth From Light Field Images (CVPR 2018 · 317 citations)
- Boosting Spike Camera Image Reconstruction from a Perspective of Dealing with Spike Fluctuations (CVPR 2024 · 16 citations)
- Progressive Divide-and-Conquer via Subsampling Decomposition for Accelerated MRI (CVPR 2024 · 16 citations)
Stereo and multi-view stereo · 405 papers
Keywords: view stereo, photometric stereo, stereo matching
Approaches in this cluster:
- Cost-volume multi-view stereo (207 papers)
Infers depth from multiple views with learned cost volumes, recurrent regularization and geometry-aware networks. - Photometric stereo (98 papers)
Estimates surface normals from varying illumination, including uncalibrated and non-Lambertian cases. - Learned stereo with specialized sensors (100 papers)
Estimates depth from stereo pairs with active, polarization, normal or exposure cues.
Most cited and most cited since 2024:
- Pyramid Stereo Matching Network (CVPR 2018 · 1,860 citations)
- A Multi-View Stereo Benchmark With High-Resolution Images and Multi-Camera Videos (CVPR 2017 · 913 citations)
- MoCha-Stereo: Motif Channel Attention Network for Stereo Matching (CVPR 2024 · 86 citations)
- MonSter: Marry Monodepth to Stereo Unleashes Power (CVPR 2025 · 53 citations)
Related topics in 3D vision
- Gaussian splatting (1,340)
- Human pose estimation (2,063)
- NeRF and radiance fields (733)
- 3D generation (2,448)
- Face and shape (2,246)
