mlsanityData to 7 Oct 2026

End2End Multi-View Feature Matching with Differentiable Pose Optimization

arXiv 2205.01694 ↗First posted 3 May 2022cs.CVcited by 19

Not on the map. Why?

Abstract (arXiv)

Erroneous feature matches have severe impact on subsequent camera pose estimation and often require additional, time-costly measures, like RANSAC, for outlier rejection. Our method tackles this challenge by addressing feature matching and pose optimization jointly. To this end, we propose a graph attention network to predict image correspondences along with confidence weights. The resulting matches serve as weighted constraints in a differentiable pose estimation. Training feature matching with gradients from pose optimization naturally learns to down-weight outliers and boosts pose estimation on image pairs compared to SuperGlue by 6.7% on ScanNet. At the same time, it reduces the pose estimation time by over 50% and renders RANSAC iterations unnecessary. Moreover, we integrate information from multiple views by spanning the graph across multiple frames to predict the matches all at once. Multi-view matching combined with end-to-end training improves the pose estimation metrics on Matterport3D by 18.5% compared to SuperGlue.

Key facts. Cited by 19 papers in the map (data through 7 Oct 2026). Builds on 34 of its 63 references that are in the map. First posted on arXiv 3 May 2022.

Builds on

34 of 63 references.

References · 34

  1. 2017Attention Is All You Needcited by 59,932
  2. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  3. 2017Neural Message Passing for Quantum Chemistrycited by 3,460
  4. 2017ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenescited by 3,088
  5. 2013Sinkhorn Distances: Lightspeed Computation of Optimal Transportation Distancescited by 2,277
  6. 2019SuperGlue: Learning Feature Matching with Graph Neural Networkscited by 1,437
  7. 2017Matterport3D: Learning from RGB-D Data in Indoor Environmentscited by 1,354
  8. 2017SuperPoint: Self-Supervised Interest Point Detection and Descriptioncited by 1,313
  9. 2021LoFTR: Detector-Free Local Feature Matching with Transformerscited by 974
  10. 2015Convolutional Networks on Graphs for Learning Molecular Fingerprintscited by 751
  11. 2018MegaDepth: Learning Single-View Depth Prediction from Internet Photoscited by 731
  12. 2017Convolutional Sequence to Sequence Learningcited by 550
22 more
  1. 2019D2-Net: A Trainable CNN for Joint Detection and Description of Local Featurescited by 324
  2. 2020DISK: Learning local features with policy gradientcited by 310
  3. 2016LIFT: Learned Invariant Feature Transformcited by 305
  4. 2020Image Matching across Wide Baselines: From Paper to Practicecited by 224
  5. 2019R2D2: Repeatable and Reliable Detector and Descriptorcited by 206
  6. 2016DeMoN: Depth and Motion Network for Learning Monocular Stereocited by 200
  7. 2021Pixel-Perfect Structure-from-Motion with Featuremetric Refinementcited by 199
  8. 2021COTR: Correspondence Transformer for Matching Across Imagescited by 182
  9. 2017Learning to Find Good Correspondencescited by 176
  10. 2018BA-Net: Dense Bundle Adjustment Networkcited by 173
  11. 2018Neighbourhood Consensus Networkscited by 173
  12. 2018LF-Net: Learning Local Features from Imagescited by 151
  13. 2019Learning Two-View Correspondences and Geometry Using Order-Aware Networkcited by 148
  14. 2019Neural-Guided RANSAC: Learning Where to Sample Model Hypothesescited by 129
  15. 2020ASLFeat: Learning Local Features of Accurate Shape and Localizationcited by 129
  16. 2018MAGSAC: marginalizing sample consensuscited by 122
  17. 20193DRegNet: A Deep Neural Network for 3D Point Registrationcited by 96
  18. 2020Efficient Neighbourhood Consensus Networks via Submanifold Sparse Convolutionscited by 92
  19. 2020Dual-Resolution Correspondence Networkscited by 90
  20. 2014Robust Camera Location Estimation by Convex Programmingcited by 41
  21. 2019Reinforced Feature Points: Optimizing Feature Detection and Description for a High-Level Taskcited by 39
  22. 2018RegNet: Learning the Optimization of Direct Image-to-Image Pose Registrationcited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 1; 2023: 3; 2024: 6; 2025: 5; 2026: 4.

19 papers.

Sort:

  1. 2024OmniGlue: Generalizable Feature Matching with Foundation Model Guidancecited by 55
  2. 2024Matching 2D Images in 3D: Metric Relative Pose from Metric Correspondencescited by 25
  3. 2023Deep Learning for Camera Calibration and Beyond: A Surveycited by 20
  4. 2024Learning to Make Keypoints Sub-Pixel Accuratecited by 13
  5. 2024FAR: Flexible, Accurate and Robust 6DoF Relative Camera Pose Estimationcited by 12
  6. 2023Learning Accurate Template Matching with Differentiable Coarse-to-Fine Correspondence Refinementcited by 9
  7. 2025CoMatcher: Multi-View Collaborative Feature Matchingcited by 7
  8. 2023Automated Camera Calibration via Homography Estimation with GNNscited by 6
  9. 2024Multi-Session SLAM with Differentiable Wide-Baseline Pose Optimizationcited by 6
  10. 2022Self-Supervised Correspondence Estimation via Multiview Registrationcited by 5
  11. 2025DINO-VO: A Feature-based Visual Odometry Leveraging a Visual Foundation Modelcited by 5
  12. 2026MV-RoMa: From Pairwise Matching into Multi-View Track Reconstructioncited by 2
7 more
  1. 2026Diff3R: Feed-forward 3D Gaussian Splatting with Uncertainty-aware Differentiable Optimizationcited by 2
  2. 2026MultiCam: On-the-fly Multi-Camera Pose Estimation Using Spatiotemporal Overlaps of Known Objectscited by 1
  3. 2024Matching Semantically Similar Non-Identical Objectscited by 0
  4. 2025SteerPose: Simultaneous Extrinsic Camera Calibration and Matching from Articulationcited by 0
  5. 2025C-DOG: Multi-View Multi-instance Feature Association Using Connected δ-Overlap Graphscited by 0
  6. 2025EndoMatcher: Generalizable Endoscopic Image Matcher via Multi-Domain Pre-training for Robot-Assisted Surgerycited by 0
  7. 2026IUP-Pose: Decoupled Iterative Uncertainty Propagation for Real-time Relative Pose Regression via Implicit Dense Alignment v1cited by 0