mlsanityData to 7 Oct 2026

DISK: Learning local features with policy gradient

arXiv 2006.13566 ↗First posted 24 Jun 2020cs.CV cs.LGcited by 310

Not on the map. Why?

Abstract (arXiv)

Local feature frameworks are difficult to learn in an end-to-end fashion, due to the discreteness inherent to the selection and matching of sparse keypoints. We introduce DISK (DIScrete Keypoints), a novel method that overcomes these obstacles by leveraging principles from Reinforcement Learning (RL), optimizing end-to-end for a high number of correct feature matches. Our simple yet expressive probabilistic model lets us keep the training and inference regimes close, while maintaining good enough convergence properties to reliably train from scratch. Our features can be extracted very densely while remaining discriminative, challenging commonly held assumptions about what constitutes a good keypoint, as showcased in Fig. 1, and deliver state-of-the-art results on three public benchmarks.

Key facts. Cited by 310 papers in the map (data through 7 Oct 2026). First posted on arXiv 24 Jun 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 15; 2022: 31; 2023: 63; 2024: 59; 2025: 74; 2026: 68.

310 papers.

Sort:

By area:

  1. 2023DUSt3R: Geometric 3D Vision Made Easycited by 1,526
  2. 2025VGGT: Visual Geometry Grounded Transformercited by 1,475
  3. 2021LoFTR: Detector-Free Local Feature Matching with Transformerscited by 974
  4. 2023LightGlue: Local Feature Matching at Light Speedcited by 583
  5. 2023Emergent Correspondence from Image Diffusioncited by 437
  6. 2023A Tale of Two Features: Stable Diffusion Complements DINO for Zero-Shot Semantic Correspondencecited by 297
  7. 2023RoMa: Robust Dense Feature Matchingcited by 296
  8. 2023Visual Geometry Grounded Deep Structure From Motioncited by 225
  9. 2021Pixel-Perfect Structure-from-Motion with Featuremetric Refinementcited by 199
  10. 2021COTR: Correspondence Transformer for Matching Across Imagescited by 182
  11. 2022OnePose: One-Shot Object Pose Estimation without CAD Modelscited by 155
  12. 2024CoTracker3: Simpler and Better Point Tracking by Pseudo-Labelling Real Videoscited by 122
38 more
  1. 2023ALIKED: A Lighter Keypoint and Descriptor Extraction Network via Deformable Transformationcited by 120
  2. 2024Efficient LoFTR: Semi-Dense Local Feature Matching with Sparse-Like Speedcited by 119
  3. 2023OnePose++: Keypoint-Free One-Shot Object Pose Estimation without CAD Modelscited by 114
  4. 2024XFeat: Accelerated Features for Lightweight Image Matchingcited by 113
  5. 2022Map-free Visual Relocalization: Metric Pose Relative to a Single Imagecited by 112
  6. 2023Accelerated Coordinate Encoding: Learning to Relocalize in Minutes using RGB and Posescited by 107
  7. 2025VGGT-Long: Chunk it, Loop it, Align it -- Pushing VGGT's Limits on Kilometer-scale Long RGB Sequencescited by 107
  8. 2023OrienterNet: Visual Localization in 2D Public Maps with Neural Matchingcited by 84
  9. 2021CATs: Cost Aggregation Transformers for Visual Correspondencecited by 77
  10. 2021ALIKE: Accurate and Lightweight Keypoint Detection and Descriptor Extractioncited by 77
  11. 2024GIM: Learning Generalizable Image Matcher From Internet Videoscited by 72
  12. 2023Detector-Free Structure from Motioncited by 71
  13. 2024Deep Patch Visual SLAMcited by 68
  14. 2023DeDoDe: Detect, Don't Describe -- Describe, Don't Detect for Local Feature Matchingcited by 62
  15. 2023Dense Optical Tracking: Connecting the Dotscited by 58
  16. 2024MegaScenes: Scene-Level View Synthesis at Scalecited by 58
  17. 2024OmniGlue: Generalizable Feature Matching with Foundation Model Guidancecited by 55
  18. 2023SiLK -- Simple Learned Keypointscited by 54
  19. 2022CATs++: Boosting Cost Aggregation with Convolutions and Transformerscited by 53
  20. 2024GaussNav: Gaussian Splatting for Visual Navigationcited by 49
  21. 2025Dynamic Point Maps: A Versatile Representation for Dynamic 3D Reconstructioncited by 43
  22. 2025MatchAnything: Universal Cross-Modality Image Matching with Large-Scale Pre-Trainingcited by 40
  23. 2022The 8-Point Algorithm as an Inductive Bias for Relative Pose Prediction by ViTscited by 38
  24. 2023SNAP: Self-Supervised Neural Maps for Visual Positioning and Semantic Understandingcited by 38
  25. 2024DeDoDe v2: Analyzing and Improving the DeDoDe Keypoint Detectorcited by 38
  26. 2023Rethinking Optical Flow from Geometric Matching Consistent Perspectivecited by 33
  27. 2022TopicFM: Robust and Interpretable Topic-Assisted Feature Matchingcited by 32
  28. 2022Adaptive Assignment for Geometry Aware Local Feature Matchingcited by 30
  29. 2022The One Where They Reconstructed 3D Humans and Environments in TV Showscited by 30
  30. 2023PMatch: Paired Masked Image Modeling for Dense Geometric Matchingcited by 29
  31. 2022Decoupling Makes Weakly Supervised Local Feature Bettercited by 28
  32. 2023SFD2: Semantic-guided Feature Detection and Descriptioncited by 28
  33. 2023360Loc: A Dataset and Benchmark for Omnidirectional Visual Localization with Cross-device Queriescited by 28
  34. 2022A case for using rotation invariant features in state of the art feature matcherscited by 27
  35. 2025FG2: Fine-Grained Cross-View Localization by Fine-Grained Feature Matchingcited by 27
  36. 2022Self-Supervised Equivariant Learning for Oriented Keypoint Detectioncited by 26
  37. 2022ReDFeat: Recoupling Detection and Description for Multimodal Feature Learningcited by 26
  38. 2023HSCNet++: Hierarchical Scene Coordinate Classification and Regression for Visual Localization with Transformercited by 26

50 most cited of 310 All 310 (JSON)