mlsanityData to 7 Oct 2026

Pixel-Perfect Structure-from-Motion with Featuremetric Refinement

arXiv 2108.08291 ↗First posted 18 Aug 2021cs.CVcited by 199

Not on the map. Why?

Abstract (arXiv)

Finding local features that are repeatable across multiple views is a cornerstone of sparse 3D reconstruction. The classical image matching paradigm detects keypoints per-image once and for all, which can yield poorly-localized features and propagate large errors to the final geometry. In this paper, we refine two key steps of structure-from-motion by a direct alignment of low-level image information from multiple views: we first adjust the initial keypoint locations prior to any geometric estimation, and subsequently refine points and camera poses as a post-processing. This refinement is robust to large detection noise and appearance changes, as it optimizes a featuremetric error based on dense features predicted by a neural network. This significantly improves the accuracy of camera poses and scene geometry for a wide range of keypoint detectors, challenging viewing conditions, and off-the-shelf deep features. Our system easily scales to large image collections, enabling pixel-perfect crowd-sourced localization at scale. Our code is publicly available at https://github.com/cvg/pixel-perfect-sfm as an add-on to the popular SfM software COLMAP.

Key facts. Cited by 199 papers in the map (data through 7 Oct 2026). Builds on 45 of its 104 references that are in the map. First posted on arXiv 18 Aug 2021.

Builds on

45 of 104 references.

References · 45

  1. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,267
  2. 2019SuperGlue: Learning Feature Matching with Graph Neural Networkscited by 1,437
  3. 2017SuperPoint: Self-Supervised Interest Point Detection and Descriptioncited by 1,313
  4. 2021LoFTR: Detector-Free Local Feature Matching with Transformerscited by 974
  5. 2017PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volumecited by 966
  6. 2016FlowNet 2.0: Evolution of Optical Flow Estimation with Deep Networkscited by 780
  7. 2018MVSNet: Depth Inference for Unstructured Multi-view Stereocited by 780
  8. 2018MegaDepth: Learning Single-View Depth Prediction from Internet Photoscited by 731
  9. 2018From Coarse to Fine: Robust Hierarchical Localization at Large Scalecited by 556
  10. 2016Direct Sparse Odometrycited by 479
  11. 2020DISK: Learning local features with policy gradientcited by 310
  12. 2016LIFT: Learned Invariant Feature Transformcited by 305
33 more
  1. 2019Recurrent MVSNet for High-resolution Multi-view Stereo Depth Inferencecited by 252
  2. 2020Image Matching across Wide Baselines: From Paper to Practicecited by 224
  3. 2019R2D2: Repeatable and Reliable Detector and Descriptorcited by 206
  4. 2021Back to the Feature: Learning Robust Camera Localization from Pixels to Posecited by 199
  5. 2018BA-Net: Dense Bundle Adjustment Networkcited by 173
  6. 2018Neighbourhood Consensus Networkscited by 173
  7. 2018LF-Net: Learning Local Features from Imagescited by 151
  8. 2019GLU-Net: Global-Local Universal Network for Dense Flow and Correspondencescited by 139
  9. 2020Patch2Pix: Epipolar-Guided Pixel-Level Correspondencescited by 128
  10. 2021Learning Accurate Dense Correspondences and When to Trust Themcited by 113
  11. 2020Fast-MVSNet: Sparse-to-Dense Multi-View Stereo With Learned Propagation and Gauss-Newton Refinementcited by 105
  12. 2020Efficient Neighbourhood Consensus Networks via Submanifold Sparse Convolutionscited by 92
  13. 2020Dual-Resolution Correspondence Networkscited by 90
  14. 2016Universal Correspondence Networkcited by 82
  15. 2020RANSAC-Flow: generic two-stage image alignmentcited by 48
  16. 2020Online Invariance Selection for Local Feature Descriptorscited by 41
  17. 2019GN-Net: The Gauss-Newton Loss for Multi-Weather Relocalizationcited by 32
  18. 2018LS-Net: Learning to Solve Nonlinear Least Squares for Monocular Stereocited by 28
  19. 2020LM-Reloc: Levenberg-Marquardt Based Direct Visual Relocalizationcited by 28
  20. 2016Photometric Bundle Adjustment for Vision-Based SLAMcited by 25
  21. 2020Multi-View Optimization of Local Feature Geometrycited by 24
  22. 2019UnsuperPoint: End-to-end Unsupervised Interest Point Detector and Descriptorcited by 22
  23. 2020S2DNet: Learning Accurate Correspondences for Sparse-to-Dense Feature Matchingcited by 22
  24. 2018Taking a Deeper Look at the Inverse Compositional Algorithmcited by 20
  25. 2018Deep ChArUco: Dark ChArUco Marker Pose Estimationcited by 18
  26. 2021Neural Reprojection Error: Merging Feature Learning and Camera Pose Estimationcited by 17
  27. 2019Neural Outlier Rejection for Self-Supervised Keypoint Learningcited by 16
  28. 2020Heatmap Regression via Randomized Roundingcited by 15
  29. 2020Efficient Initial Pose-graph Generation for Global SfMcited by 14
  30. 2020Deep Probabilistic Feature-metric Trackingcited by 12
  31. 2020Large Scale Photometric Bundle Adjustmentcited by 4
  32. 2017Semantic Texture for Robust Dense Trackingcited by 3
  33. 2019Optimal Multi-view Correction of Local Affine Framescited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 2; 2022: 25; 2023: 39; 2024: 48; 2025: 56; 2026: 29.

199 papers.

Sort:

By area:

  1. 2023DUSt3R: Geometric 3D Vision Made Easycited by 1,526
  2. 2025VGGT: Visual Geometry Grounded Transformercited by 1,475
  3. 2024Grounding Image Matching in 3D with MASt3Rcited by 960
  4. 2023LightGlue: Local Feature Matching at Light Speedcited by 583
  5. 2021Mega-NeRF: Scalable Construction of Large-Scale NeRFs for Virtual Fly-Throughscited by 396
  6. 2025Fast3R: Towards 3D Reconstruction of 1000+ Images in One Forward Passcited by 356
  7. 20243D Reconstruction with Spatial Memorycited by 269
  8. 2023Visual Geometry Grounded Deep Structure From Motioncited by 225
  9. 2024Global Structure-from-Motion Revisitedcited by 215
  10. 2024A Hierarchical 3D Gaussian Representation for Real-Time Rendering of Very Large Datasetscited by 214
  11. 2022Deep Patch Visual Odometrycited by 186
  12. 2025FLARE: Feed-forward Geometry, Appearance and Camera Estimation from Uncalibrated Sparse Viewscited by 176
38 more
  1. 2022DKM: Dense Kernelized Feature Matching for Geometry Estimationcited by 174
  2. 2024MASt3R-SfM: a Fully-Integrated Solution for Unconstrained Structure-from-Motioncited by 161
  3. 2025Streaming 4D Visual Geometry Transformercited by 147
  4. 2023PoseDiffusion: Solving Pose Estimation via Diffusion-aided Bundle Adjustmentcited by 146
  5. 2024Cameras as Rays: Pose Estimation via Ray Diffusioncited by 129
  6. 2022MatchFormer: Interleaving Attention in Transformers for Feature Matchingcited by 126
  7. 2024Efficient LoFTR: Semi-Dense Local Feature Matching with Sparse-Like Speedcited by 119
  8. 2023OnePose++: Keypoint-Free One-Shot Object Pose Estimation without CAD Modelscited by 114
  9. 2025VGGT-Long: Chunk it, Loop it, Align it -- Pushing VGGT's Limits on Kilometer-scale Long RGB Sequencescited by 107
  10. 2022RelPose: Predicting Probabilistic Relative Rotation for Single Objects in the Wildcited by 102
  11. 2024SLAM3R: Real-Time Dense Scene Reconstruction from Monocular RGB Videoscited by 100
  12. 2022LaMAR: Benchmarking Localization and Mapping for Augmented Realitycited by 98
  13. 2024Stereo4D: Learning How Things Move in 3D from Internet Stereo Videoscited by 89
  14. 2023OrienterNet: Visual Localization in 2D Public Maps with Neural Matchingcited by 84
  15. 2022DFNet: Enhance Absolute Pose Regression with Direct Feature Matchingcited by 82
  16. 2024Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localizationcited by 81
  17. 2024Large Spatial Model: End-to-end Unposed Images to Semantic 3Dcited by 73
  18. 2023Detector-Free Structure from Motioncited by 71
  19. 2025Pow3R: Empowering Unconstrained 3D Reconstruction with Camera and Scene Priorscited by 69
  20. 2023DBARF: Deep Bundle-Adjusting Generalizable Neural Radiance Fieldscited by 60
  21. 2022DeepLSD: Line Segment Detection and Refinement with Deep Image Gradientscited by 58
  22. 2022SparsePose: Sparse-View Camera Pose Regression and Refinementcited by 53
  23. 2025Light3R-SfM: Towards Feed-forward Structure-from-Motioncited by 47
  24. 2024SplineGS: Robust Motion-Adaptive Spline for Real-Time Dynamic 3D Gaussians from Monocular Videocited by 44
  25. 20233D Line Mapping Revisitedcited by 39
  26. 2025Matrix3D: Large Photogrammetry Model All-in-Onecited by 39
  27. 2025AerialMegaDepth: Learning Aerial-Ground Reconstruction and View Synthesiscited by 39
  28. 2023SNAP: Self-Supervised Neural Maps for Visual Positioning and Semantic Understandingcited by 38
  29. 2023Neural Refinement for Absolute Pose Regression with Feature Synthesiscited by 37
  30. 2023Efficient Large-scale Scene Representation with a Hybrid of High-resolution Grid and Plane Featurescited by 36
  31. 2023Tracking and Mapping in Medical Computer Vision: A Reviewcited by 36
  32. 2023LU-NeRF: Scene and Pose Estimation by Synchronizing Local Unposed NeRFscited by 33
  33. 2022AligNeRF: High-Fidelity Neural Radiance Fields via Alignment-Aware Trainingcited by 31
  34. 2024Robust Incremental Structure-from-Motion with Hybrid Featurescited by 31
  35. 2023POPE: 6-DoF Promptable Pose Estimation of Any Object, in Any Scene, with One Referencecited by 30
  36. 2023Deep Event Visual Odometrycited by 29
  37. 2022PlaneFormers: From Sparse View Planes to 3D Reconstructioncited by 26
  38. 2024Local Feature Matching Using Deep Learning: A Surveycited by 26

50 most cited of 199 All 199 (JSON)