---
title: "SuperGlue — paper by Sarlin et al. (2019) — summary and ideas"
description: "SuperGlue addresses data association in geometric vision: deciding which local features in two… 73 references, cited by 91 papers here."
authors: ["Paul-Edouard Sarlin", "Daniel DeTone", "Tomasz Malisiewicz", "Andrew Rabinovich"]
year: 2019
venue: "arXiv:1911.11763"
source: "https://arxiv.org/abs/1911.11763"
doi: "10.48550/arxiv.1911.11763"
arxiv: "1911.11763"
licence: "arXiv — non-exclusive license to distribute"
licence_url: "http://arxiv.org/licenses/nonexclusive-distrib/1.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/sarlin-2019-superglue-learning-feature-matching-graph/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/sarlin-2019-superglue-learning-feature-matching-graph/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2019](https://md.research.elamur.ai/free/en/papers/2019/) / SuperGlue: Learning Feature Matching with Graph Neural Networks

# SuperGlue: Learning Feature Matching with Graph Neural Networks

**Authors:** Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz and Andrew Rabinovich · **Year:** 2019 · **Venue:** [arXiv:1911.11763](https://md.research.elamur.ai/free/en/venues/ieee-computer-society-conference-on-computer-vision-and-pattern-recognition/) ([IEEE Computer Society Conference on Computer Vision and Pattern Recognition 2019](https://md.research.elamur.ai/free/en/venues/ieee-computer-society-conference-on-computer-vision-and-pattern-recognition/2019/)) · **Publisher:** [Institute of Electrical and Electronics Engineers](https://md.research.elamur.ai/free/en/publishers/institute-of-electrical-and-electronics-engineers/) · **Licence:** [arXiv — non-exclusive license to distribute](http://arxiv.org/licenses/nonexclusive-distrib/1.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/1911.11763) · [PDF](https://arxiv.org/pdf/1911.11763) · [DOI 10.48550/arxiv.1911.11763](https://doi.org/10.48550/arxiv.1911.11763) · [arXiv 1911.11763](https://arxiv.org/abs/1911.11763) · [OpenAlex](https://openalex.org/W2990957769)

**Citations:** 2,830 citations (OpenAlex, 2026-10-14) · [91](https://md.research.elamur.ai/free/en/papers/sarlin-2019-superglue-learning-feature-matching-graph/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
SuperGlue addresses data association in geometric vision: deciding which local features in two images depict the same 3D point, and which have no counterpart. Instead of designing better descriptors and then matching them with nearest-neighbour search and hand-tuned filters, the authors learn the matching step itself, as a "middle-end" between a feature front-end (e.g., SuperPoint or SIFT) and a pose-estimation or bundle-adjustment back-end. Each keypoint's descriptor is enriched with an MLP embedding of its position and confidence. A graph neural network then alternates attention within each image and across the two images to produce matching descriptors. Their pairwise scores, extended with a "dustbin" row and column for unmatched points, are converted into a partial soft assignment by differentiable optimal transport (Sinkhorn iterations). The whole system is trained end-to-end on annotated image pairs. Combined with SuperPoint, it beats handcrafted matchers and learned inlier classifiers on indoor and outdoor pose estimation, runs in real time on a GPU, and is released publicly. This makes it a practical upgrade for SfM and SLAM pipelines.

## Key points
- The architecture has two blocks: an attentional multiplex GNN, with self edges inside an image and cross edges between images, stacked for L alternating layers, and an optimal matching layer that runs the Sinkhorn algorithm for T iterations on an (M+1)×(N+1) score matrix.
- The output is a partial soft assignment matrix P ∈ [0,1]^(M×N). Its row and column sums are bounded by 1, so each keypoint gets at most one match, and occluded or non-repeatable points can stay unmatched.
- The keypoint encoder adds an MLP embedding of (x, y, detection confidence) to the visual descriptor. This is analogous to positional encodings in Transformers.
- Baselines include NN+distance and NN+OANet. Qualitative outdoor results are shown on the MegaDepth validation and PhotoTourism test sets, where SuperGlue often recovers poses with errors of a few degrees while the baselines fail. An indoor example reaches R 2.5°, t 1.4° with 59/68 inliers.
- Code and trained weights are available at github.com/magicleap/SuperGluePretrainedNetwork.

## Interesting ideas
- **Learn the cost, keep the solver.** A network predicts the costs, and a classical, differentiable optimization (Sinkhorn optimal transport) enforces the structural constraints. This hybrid could suit any alignment task that needs one-to-one structure, such as aligning entities, sentences, or records across two sources.
- **Dustbins for "no match."** Adding an extra row and column that can absorb unmatched elements is a cheap, principled way to let a soft assignment abstain. This is useful wherever the two sets only partially overlap.
- **Alternating self- and cross-attention over two sets.** Elements first gather context within their own set and then compare against the other set, repeatedly. This mimics a human looking back and forth between two images and transfers naturally to comparing two documents, tables, or maps.
- **Attention maps as a diagnostic.** By visualizing attention per layer, the authors show that easy points get resolved early, hard ones only in late layers, and attention spans tend to narrow with depth. Inspecting a model this way is a useful sanity check and also exposes failure modes, such as locking onto a repeated, self-similar structure.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [ETH Zurich](https://md.research.elamur.ai/free/en/universities/eth-zurich/) — University, Zurich, Switzerland · 84 works · 9,181 citations

Also on the byline: [Magic Leap](https://ror.org/058j17k65).

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 3,486 works · [⚙️ Engineering](https://md.research.elamur.ai/free/en/shelves/engineering/) · 188 works · [⚙️ Aerospace Engineering](https://md.research.elamur.ai/free/en/shelves/engineering/aerospace-engineering/) · 4 works

💡 Concepts (1): [💡 Natural language understanding](https://md.research.elamur.ai/free/en/concepts/natural-language-understanding/) · 52 works

## Cite

Cite as: Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz and Andrew Rabinovich. 2019. SuperGlue: Learning Feature Matching with Graph Neural Networks. arXiv:1911.11763.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{sarlin2019superglue,
  title = {{SuperGlue: Learning Feature Matching with Graph Neural Networks}},
  author = {Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz and Andrew Rabinovich},
  year = {2019},
  doi = {10.48550/arxiv.1911.11763},
  eprint = {1911.11763},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1911.11763}
}
```

- [← previous Language Models as Knowledge Bases?](https://md.research.elamur.ai/free/en/papers/petroni-2019-language-models-as-knowledge-bases/)
- [next → ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://md.research.elamur.ai/free/en/papers/lan-2019-albert-lite-bert-self-supervised/)
- [↑ back to papers of 2019](https://md.research.elamur.ai/free/en/papers/2019/)

## References

73 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Phototourism Challenge, CVPR 2019 Image Matching Workshop. [https://image-matching-workshop.github.io](https://image-matching-workshop.github.io). Accessed November 8, 2019. 7, 10 [↗](https://image-matching-workshop.github.io)
2. Relja Arandjelović and Andrew Zisserman. Three things everyone should know to improve object retrieval. In CVPR, 2012. 6 [find](https://scholar.google.com/scholar?q=Relja+Arandjelovic%CC%81+and+Andrew+Zisserman%2E+Three+things+everyone+should+know+to+improve+object+retrieval%2E+In+CVPR%2C+2012%2E+6)
3. Vassileios Balntas, Karel Lenc, Andrea Vedaldi, and Krystian Mikolajczyk. Hpatches: A benchmark and evaluation of handcrafted and learned local descriptors. In CVPR, 2017. 9 [find](https://scholar.google.com/scholar?q=Vassileios+Balntas%2C+Karel+Lenc%2C+Andrea+Vedaldi%2C+and+Krystian+Mikolajczyk%2E+Hpatches%3A+A+benchmark+and+evaluation+of+handcrafted+and+learned+local+descriptors%2E+In+CVPR%2C+2017%2E+9)
4. Peter W Battaglia, Jessica B Hamrick, Victor Bapst, Alvaro Sanchez-Gonzalez, Vinicius Zambaldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Ryan Faulkner, et al. [Relational inductive biases, deep learning, and graph networks](https://md.research.elamur.ai/free/en/papers/battaglia-2018-relational-inductive-biases-deep-learning/). arXiv:1806.01261, 2018. 2, 3
5. Alexander C Berg, Tamara L Berg, and Jitendra Malik. Shape matching and object recognition using low distortion correspondences. In CVPR, 2005. 2 [find](https://scholar.google.com/scholar?q=Alexander+C+Berg%2C+Tamara+L+Berg%2C+and+Jitendra+Malik%2E+Shape+matching+and+object+recognition+using+low+distortion+correspondences%2E+In+CVPR%2C+2005%2E+2)
6. JiaWang Bian, Wen-Yan Lin, Yasuyuki Matsushita, Sai-Kit Yeung, Tan-Dat Nguyen, and Ming-Ming Cheng. GMS: Grid-based motion statistics for fast, ultra-robust feature correspondence. In CVPR, 2017. 2, 6 [find](https://scholar.google.com/scholar?q=JiaWang+Bian%2C+Wen%2DYan+Lin%2C+Yasuyuki+Matsushita%2C+Sai%2DKit+Yeung%2C+Tan%2DDat+Nguyen%2C+and+Ming%2DMing+Cheng%2E+GMS%3A+Grid%2Dbased+motion+statistics+for+fast%2C+ultra%2Drobust+feature+correspondence%2E+In+CVPR%2C+2017%2E+2%2C+6)
7. Eric Brachmann and Carsten Rother. Neural-Guided RANSAC: Learning where to sample model hypotheses. In ICCV, 2019. 2, 5, 6, 11, 12 [find](https://scholar.google.com/scholar?q=Eric+Brachmann+and+Carsten+Rother%2E+Neural%2DGuided+RANSAC%3A+Learning+where+to+sample+model+hypotheses%2E+In+ICCV%2C+2019%2E+2%2C+5%2C+6%2C+11%2C+12)
8. Cesar Cadena, Luca Carlone, Henry Carrillo, Yasir Latif, Davide Scaramuzza, José Neira, Ian Reid, and John J Leonard. Past, present, and future of simultaneous localization and mapping: Toward the robust-perception age. IEEE Transactions on robotics, 32(6):1309–1332, 2016. 1 [find](https://scholar.google.com/scholar?q=Cesar+Cadena%2C+Luca+Carlone%2C+Henry+Carrillo%2C+Yasir+Latif%2C+Davide+Scaramuzza%2C+Jose%CC%81+Neira%2C+Ian+Reid%2C+and+John+J+Leonard%2E+Past%2C+present%2C+and+future+of+simultaneous+localization+and+mapping%3A+Toward+the+ro)
9. Tibério S Caetano, Julian J McAuley, Li Cheng, Quoc V Le, and Alex J Smola. Learning graph matching. IEEE TPAMI, 31(6):1048–1058, 2009. 2 [find](https://scholar.google.com/scholar?q=Tibe%CC%81rio+S+Caetano%2C+Julian+J+McAuley%2C+Li+Cheng%2C+Quoc+V+Le%2C+and+Alex+J+Smola%2E+Learning+graph+matching%2E+IEEE+TPAMI%2C+31%286%29%3A1048%E2%80%931058%2C+2009%2E+2)
10. Jan Cech, Jiri Matas, and Michal Perdoch. Efficient sequential correspondence selection by cosegmentation. IEEE TPAMI, 32(9):1568–1581, 2010. 2 [find](https://scholar.google.com/scholar?q=Jan+Cech%2C+Jiri+Matas%2C+and+Michal+Perdoch%2E+Efficient+sequential+correspondence+selection+by+cosegmentation%2E+IEEE+TPAMI%2C+32%289%29%3A1568%E2%80%931581%2C+2010%2E+2)
11. Marvin M Chun. Contextual cueing of visual attention. Trends in cognitive sciences, 4(5):170–178, 2000. 3 [find](https://scholar.google.com/scholar?q=Marvin+M+Chun%2E+Contextual+cueing+of+visual+attention%2E+Trends+in+cognitive+sciences%2C+4%285%29%3A170%E2%80%93178%2C+2000%2E+3)
12. Marco Cuturi. Sinkhorn distances: Lightspeed computation of optimal transport. In NIPS, 2013. 2, 5 [find](https://scholar.google.com/scholar?q=Marco+Cuturi%2E+Sinkhorn+distances%3A+Lightspeed+computation+of+optimal+transport%2E+In+NIPS%2C+2013%2E+2%2C+5)
13. Angela Dai, Angel X Chang, Manolis Savva, Maciej Halber, Thomas Funkhouser, and Matthias Nießner. ScanNet: Richly-annotated 3D reconstructions of indoor scenes. In CVPR, 2017. 6 [find](https://scholar.google.com/scholar?q=Angela+Dai%2C+Angel+X+Chang%2C+Manolis+Savva%2C+Maciej+Halber%2C+Thomas+Funkhouser%2C+and+Matthias+Nie%C3%9Fner%2E+ScanNet%3A+Richly%2Dannotated+3D+reconstructions+of+indoor+scenes%2E+In+CVPR%2C+2017%2E+6)
14. Angela Dai, Matthias Nießner, Michael Zollhöfer, Shahram Izadi, and Christian Theobalt. Bundlefusion: Real-time globally consistent 3d reconstruction using on-the-fly surface reintegration. ACM Transactions on Graphics (ToG), 36(3):24, 2017. 11 [find](https://scholar.google.com/scholar?q=Angela+Dai%2C+Matthias+Nie%C3%9Fner%2C+Michael+Zollho%CC%88fer%2C+Shahram+Izadi%2C+and+Christian+Theobalt%2E+Bundlefusion%3A+Real%2Dtime+globally+consistent+3d+reconstruction+using+on%2Dthe%2Dfly+surface+reintegration%2E+ACM+Trans)
15. Haowen Deng, Tolga Birdal, and Slobodan Ilic. PPFNet: Global context aware local features for robust 3D point matching. In CVPR, 2018. 2 [find](https://scholar.google.com/scholar?q=Haowen+Deng%2C+Tolga+Birdal%2C+and+Slobodan+Ilic%2E+PPFNet%3A+Global+context+aware+local+features+for+robust+3D+point+matching%2E+In+CVPR%2C+2018%2E+2)
16. Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Deep image homography estimation. In RSS Workshop: Limits and Potentials of Deep Learning in Robotics, 2016. 5 [find](https://scholar.google.com/scholar?q=Daniel+DeTone%2C+Tomasz+Malisiewicz%2C+and+Andrew+Rabinovich%2E+Deep+image+homography+estimation%2E+In+RSS+Workshop%3A+Limits+and+Potentials+of+Deep+Learning+in+Robotics%2C+2016%2E+5)
17. Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Self-improving visual odometry. arXiv:1812.03245, 2018. 6 — [arXiv:1812.03245](https://arxiv.org/abs/1812.03245) [find](https://scholar.google.com/scholar?q=Daniel+DeTone%2C+Tomasz+Malisiewicz%2C+and+Andrew+Rabinovich%2E+Self%2Dimproving+visual+odometry%2E+arXiv%3A1812%2E03245%2C+2018%2E+6+%E2%80%94+%5BarXiv%3A1812%2E03245%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F1812%2E03245%29)
18. Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. SuperPoint: Self-supervised interest point detection and description. In CVPR Workshop on Deep Learning for Visual SLAM, 2018. 2, 4, 5, 6, 9 [find](https://scholar.google.com/scholar?q=Daniel+DeTone%2C+Tomasz+Malisiewicz%2C+and+Andrew+Rabinovich%2E+SuperPoint%3A+Self%2Dsupervised+interest+point+detection+and+description%2E+In+CVPR+Workshop+on+Deep+Learning+for+Visual+SLAM%2C+2018%2E+2%2C+4%2C+5%2C+6%2C+9)
19. Mihai Dusmanu, Ignacio Rocco, Tomas Pajdla, Marc Pollefeys, Josef Sivic, Akihiko Torii, and Torsten Sattler. D2-Net: A trainable CNN for joint detection and description of local features. In CVPR, 2019. 2, 6, 7, 10, 12 [find](https://scholar.google.com/scholar?q=Mihai+Dusmanu%2C+Ignacio+Rocco%2C+Tomas+Pajdla%2C+Marc+Pollefeys%2C+Josef+Sivic%2C+Akihiko+Torii%2C+and+Torsten+Sattler%2E+D2%2DNet%3A+A+trainable+CNN+for+joint+detection+and+description+of+local+features%2E+In+CVPR%2C+201)
20. Patrick Ebel, Anastasiia Mishchuk, Kwang Moo Yi, Pascal Fua, and Eduard Trulls. Beyond cartesian representations for local descriptors. In ICCV, 2019. 2 [find](https://scholar.google.com/scholar?q=Patrick+Ebel%2C+Anastasiia+Mishchuk%2C+Kwang+Moo+Yi%2C+Pascal+Fua%2C+and+Eduard+Trulls%2E+Beyond+cartesian+representations+for+local+descriptors%2E+In+ICCV%2C+2019%2E+2)
21. Martin A Fischler and Robert C Bolles. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography. Communications of the ACM, 24(6):381–395, 1981. 2 [find](https://scholar.google.com/scholar?q=Martin+A+Fischler+and+Robert+C+Bolles%2E+Random+sample+consensus%3A+a+paradigm+for+model+fitting+with+applications+to+image+analysis+and+automated+cartography%2E+Communications+of+the+ACM%2C+24%286%29%3A381%E2%80%93395%2C+19)
22. Jonas Gehring, Michael Auli, David Grangier, Denis Yarats, and Yann N Dauphin. [Convolutional sequence to sequence learning](https://md.research.elamur.ai/free/en/papers/gehring-2017-convolutional-sequence-sequence-learning/). In ICML, 2017. 3
23. Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. [Neural message passing for quantum chemistry](https://md.research.elamur.ai/free/en/papers/gilmer-2017-neural-message-passing-quantum-chemistry/). In ICML, 2017. 2, 3
24. Richard Hartley and Andrew Zisserman. Multiple view geometry in computer vision. Cambridge university press, 2003. 6, 11 [find](https://scholar.google.com/scholar?q=Richard+Hartley+and+Andrew+Zisserman%2E+Multiple+view+geometry+in+computer+vision%2E+Cambridge+university+press%2C+2003%2E+6%2C+11)
25. Diederik P Kingma and Jimmy Ba. [Adam: A method for stochastic optimization](https://md.research.elamur.ai/free/en/papers/kingma-2014-adam-method-stochastic-optimization/). arXiv:1412.6980, 2014. 12
26. Juho Lee, Yoonho Lee, Jungtaek Kim, Adam Kosiorek, Seungjin Choi, and Yee Whye Teh. Set Transformer: A framework for attention-based permutation-invariant neural networks. In ICML, 2019. 2 [find](https://scholar.google.com/scholar?q=Juho+Lee%2C+Yoonho+Lee%2C+Jungtaek+Kim%2C+Adam+Kosiorek%2C+Seungjin+Choi%2C+and+Yee+Whye+Teh%2E+Set+Transformer%3A+A+framework+for+attention%2Dbased+permutation%2Dinvariant+neural+networks%2E+In+ICML%2C+2019%2E+2)
27. Marius Leordeanu and Martial Hebert. A spectral technique for correspondence problems using pairwise constraints. In ICCV, 2005. 2 [find](https://scholar.google.com/scholar?q=Marius+Leordeanu+and+Martial+Hebert%2E+A+spectral+technique+for+correspondence+problems+using+pairwise+constraints%2E+In+ICCV%2C+2005%2E+2)
28. Yujia Li, Chenjie Gu, Thomas Dullien, Oriol Vinyals, and Pushmeet Kohli. Graph matching networks for learning the similarity of graph structured objects. In ICML, 2019. 2 [find](https://scholar.google.com/scholar?q=Yujia+Li%2C+Chenjie+Gu%2C+Thomas+Dullien%2C+Oriol+Vinyals%2C+and+Pushmeet+Kohli%2E+Graph+matching+networks+for+learning+the+similarity+of+graph+structured+objects%2E+In+ICML%2C+2019%2E+2)
29. Zhengqi Li and Noah Snavely. MegaDepth: Learning single-view depth prediction from internet photos. In CVPR, 2018. 7 [find](https://scholar.google.com/scholar?q=Zhengqi+Li+and+Noah+Snavely%2E+MegaDepth%3A+Learning+single%2Dview+depth+prediction+from+internet+photos%2E+In+CVPR%2C+2018%2E+7)
30. Eliane Maria Loiola, Nair Maria Maia de Abreu, Paulo Oswaldo Boaventura-Netto, Peter Hahn, and Tania Querido. A survey for the quadratic assignment problem. European journal of operational research, 176(2):657–690, 2007. 2 [find](https://scholar.google.com/scholar?q=Eliane+Maria+Loiola%2C+Nair+Maria+Maia+de+Abreu%2C+Paulo+Oswaldo+Boaventura%2DNetto%2C+Peter+Hahn%2C+and+Tania+Querido%2E+A+survey+for+the+quadratic+assignment+problem%2E+European+journal+of+operational+research%2C+1)
31. David G Lowe. Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision, 60(2):91–110, 2004. 2, 6 [find](https://scholar.google.com/scholar?q=David+G+Lowe%2E+Distinctive+image+features+from+scale%2Dinvariant+keypoints%2E+International+Journal+of+Computer+Vision%2C+60%282%29%3A91%E2%80%93110%2C+2004%2E+2%2C+6)
32. Zixin Luo, Tianwei Shen, Lei Zhou, Jiahui Zhang, Yao Yao, Shiwei Li, Tian Fang, and Long Quan. ContextDesc: Local descriptor augmentation with cross-modality context. In CVPR, 2019. 2, 3, 5, 6 [find](https://scholar.google.com/scholar?q=Zixin+Luo%2C+Tianwei+Shen%2C+Lei+Zhou%2C+Jiahui+Zhang%2C+Yao+Yao%2C+Shiwei+Li%2C+Tian+Fang%2C+and+Long+Quan%2E+ContextDesc%3A+Local+descriptor+augmentation+with+cross%2Dmodality+context%2E+In+CVPR%2C+2019%2E+2%2C+3%2C+5%2C+6)
33. Kwang Moo Yi, Eduard Trulls, Yuki Ono, Vincent Lepetit, Mathieu Salzmann, and Pascal Fua. Learning to find good correspondences. In CVPR, 2018. 2, 5, 6, 11, 12 [find](https://scholar.google.com/scholar?q=Kwang+Moo+Yi%2C+Eduard+Trulls%2C+Yuki+Ono%2C+Vincent+Lepetit%2C+Mathieu+Salzmann%2C+and+Pascal+Fua%2E+Learning+to+find+good+correspondences%2E+In+CVPR%2C+2018%2E+2%2C+5%2C+6%2C+11%2C+12)
34. Peter J Mucha, Thomas Richardson, Kevin Macon, Mason A Porter, and Jukka-Pekka Onnela. Community structure in time-dependent, multiscale, and multiplex networks. Science, 328(5980):876–878, 2010. 3 [find](https://scholar.google.com/scholar?q=Peter+J+Mucha%2C+Thomas+Richardson%2C+Kevin+Macon%2C+Mason+A+Porter%2C+and+Jukka%2DPekka+Onnela%2E+Community+structure+in+time%2Ddependent%2C+multiscale%2C+and+multiplex+networks%2E+Science%2C+328%285980%29%3A876%E2%80%93878%2C+2010%2E+3)
35. James Munkres. Algorithms for the assignment and transportation problems. Journal of the society for industrial and applied mathematics, 5(1):32–38, 1957. 5 [find](https://scholar.google.com/scholar?q=James+Munkres%2E+Algorithms+for+the+assignment+and+transportation+problems%2E+Journal+of+the+society+for+industrial+and+applied+mathematics%2C+5%281%29%3A32%E2%80%9338%2C+1957%2E+5)
36. Vincenzo Nicosia, Ginestra Bianconi, Vito Latora, and Marc Barthelemy. Growing multiplex networks. Physical review letters, 111(5):058701, 2013. 3 [find](https://scholar.google.com/scholar?q=Vincenzo+Nicosia%2C+Ginestra+Bianconi%2C+Vito+Latora%2C+and+Marc+Barthelemy%2E+Growing+multiplex+networks%2E+Physical+review+letters%2C+111%285%29%3A058701%2C+2013%2E+3)
37. Yuki Ono, Eduard Trulls, Pascal Fua, and Kwang Moo Yi. LF-Net: Learning local features from images. In NeurIPS, 2018. 2, 6, 7, 12 [find](https://scholar.google.com/scholar?q=Yuki+Ono%2C+Eduard+Trulls%2C+Pascal+Fua%2C+and+Kwang+Moo+Yi%2E+LF%2DNet%3A+Learning+local+features+from+images%2E+In+NeurIPS%2C+2018%2E+2%2C+6%2C+7%2C+12)
38. Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in PyTorch. In NIPS Workshops, 2017. 5 [find](https://scholar.google.com/scholar?q=Adam+Paszke%2C+Sam+Gross%2C+Soumith+Chintala%2C+Gregory+Chanan%2C+Edward+Yang%2C+Zachary+DeVito%2C+Zeming+Lin%2C+Alban+Desmaison%2C+Luca+Antiga%2C+and+Adam+Lerer%2E+Automatic+differentiation+in+PyTorch%2E+In+NIPS+Workshops)
39. Gabriel Peyré and Marco Cuturi. Computational optimal transport. Foundations and Trends R in Machine Learning, 11(5-6):355–607, 2019. 2, 4 [find](https://scholar.google.com/scholar?q=Gabriel+Peyre%CC%81+and+Marco+Cuturi%2E+Computational+optimal+transport%2E+Foundations+and+Trends+R+in+Machine+Learning%2C+11%285%2D6%29%3A355%E2%80%93607%2C+2019%2E+2%2C+4)
40. Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. PointNet: Deep learning on point sets for 3D classification and segmentation. In CVPR, 2017. 2 [find](https://scholar.google.com/scholar?q=Charles+R+Qi%2C+Hao+Su%2C+Kaichun+Mo%2C+and+Leonidas+J+Guibas%2E+PointNet%3A+Deep+learning+on+point+sets+for+3D+classification+and+segmentation%2E+In+CVPR%2C+2017%2E+2)
41. Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on point sets in a metric space. In NIPS, 2017. 2 [find](https://scholar.google.com/scholar?q=Charles+Ruizhongtai+Qi%2C+Li+Yi%2C+Hao+Su%2C+and+Leonidas+J+Guibas%2E+Pointnet%2B%2B%3A+Deep+hierarchical+feature+learning+on+point+sets+in+a+metric+space%2E+In+NIPS%2C+2017%2E+2)
42. Filip Radenović, Ahmet Iscen, Giorgos Tolias, Yannis Avrithis, and Ondřej Chum. Revisiting Oxford and Paris: Large-scale image retrieval benchmarking. In CVPR, 2018. 5 [find](https://scholar.google.com/scholar?q=Filip+Radenovic%CC%81%2C+Ahmet+Iscen%2C+Giorgos+Tolias%2C+Yannis+Avrithis%2C+and+Ondr%CC%8Cej+Chum%2E+Revisiting+Oxford+and+Paris%3A+Large%2Dscale+image+retrieval+benchmarking%2E+In+CVPR%2C+2018%2E+5)
43. Rahul Raguram, Jan-Michael Frahm, and Marc Pollefeys. A comparative analysis of RANSAC techniques leading to adaptive real-time random sample consensus. In ECCV, 2008. 2 [find](https://scholar.google.com/scholar?q=Rahul+Raguram%2C+Jan%2DMichael+Frahm%2C+and+Marc+Pollefeys%2E+A+comparative+analysis+of+RANSAC+techniques+leading+to+adaptive+real%2Dtime+random+sample+consensus%2E+In+ECCV%2C+2008%2E+2)
44. René Ranftl and Vladlen Koltun. Deep fundamental matrix estimation. In ECCV, 2018. 2, 5 [find](https://scholar.google.com/scholar?q=Rene%CC%81+Ranftl+and+Vladlen+Koltun%2E+Deep+fundamental+matrix+estimation%2E+In+ECCV%2C+2018%2E+2%2C+5)
45. Jerome Revaud, Philippe Weinzaepfel, César De Souza, Noe Pion, Gabriela Csurka, Yohann Cabon, and Martin Humenberger. R2D2: Repeatable and reliable detector and descriptor. In NeurIPS, 2019. 2, 5, 9, 10 [find](https://scholar.google.com/scholar?q=Jerome+Revaud%2C+Philippe+Weinzaepfel%2C+Ce%CC%81sar+De+Souza%2C+Noe+Pion%2C+Gabriela+Csurka%2C+Yohann+Cabon%2C+and+Martin+Humenberger%2E+R2D2%3A+Repeatable+and+reliable+detector+and+descriptor%2E+In+NeurIPS%2C+2019%2E+2%2C+5%2C+9%2C)
46. Ignacio Rocco, Mircea Cimpoi, Relja Arandjelović, Akihiko Torii, Tomas Pajdla, and Josef Sivic. Neighbourhood consensus networks. In NeurIPS, 2018. 2, 5 [find](https://scholar.google.com/scholar?q=Ignacio+Rocco%2C+Mircea+Cimpoi%2C+Relja+Arandjelovic%CC%81%2C+Akihiko+Torii%2C+Tomas+Pajdla%2C+and+Josef+Sivic%2E+Neighbourhood+consensus+networks%2E+In+NeurIPS%2C+2018%2E+2%2C+5)
47. Ethan Rublee, Vincent Rabaud, Kurt Konolige, and Gary R Bradski. ORB: An efficient alternative to SIFT or SURF. In ICCV, 2011. 6 [find](https://scholar.google.com/scholar?q=Ethan+Rublee%2C+Vincent+Rabaud%2C+Kurt+Konolige%2C+and+Gary+R+Bradski%2E+ORB%3A+An+efficient+alternative+to+SIFT+or+SURF%2E+In+ICCV%2C+2011%2E+6)
48. Paul-Edouard Sarlin, Cesar Cadena, Roland Siegwart, and Marcin Dymczyk. From coarse to fine: Robust hierarchical localization at large scale. In CVPR, 2019. 10 [find](https://scholar.google.com/scholar?q=Paul%2DEdouard+Sarlin%2C+Cesar+Cadena%2C+Roland+Siegwart%2C+and+Marcin+Dymczyk%2E+From+coarse+to+fine%3A+Robust+hierarchical+localization+at+large+scale%2E+In+CVPR%2C+2019%2E+10)
49. Torsten Sattler, Bastian Leibe, and Leif Kobbelt. SCRAM-SAC: Improving RANSAC’s efficiency with a spatial consistency filter. In ICCV, 2009. 2 [find](https://scholar.google.com/scholar?q=Torsten+Sattler%2C+Bastian+Leibe%2C+and+Leif+Kobbelt%2E+SCRAM%2DSAC%3A+Improving+RANSAC%E2%80%99s+efficiency+with+a+spatial+consistency+filter%2E+In+ICCV%2C+2009%2E+2)
50. Torsten Sattler, Will Maddern, Carl Toft, Akihiko Torii, Lars Hammarstrand, Erik Stenborg, Daniel Safari, Masatoshi Okutomi, Marc Pollefeys, Josef Sivic, et al. Benchmarking 6dof outdoor visual localization in changing conditions. In CVPR, 2018. 10 [find](https://scholar.google.com/scholar?q=Torsten+Sattler%2C+Will+Maddern%2C+Carl+Toft%2C+Akihiko+Torii%2C+Lars+Hammarstrand%2C+Erik+Stenborg%2C+Daniel+Safari%2C+Masatoshi+Okutomi%2C+Marc+Pollefeys%2C+Josef+Sivic%2C+et+al%2E+Benchmarking+6dof+outdoor+visual+locali)
51. Torsten Sattler, Tobias Weyand, Bastian Leibe, and Leif Kobbelt. Image retrieval for image-based localization revisited. In BMVC, 2012. 10 [find](https://scholar.google.com/scholar?q=Torsten+Sattler%2C+Tobias+Weyand%2C+Bastian+Leibe%2C+and+Leif+Kobbelt%2E+Image+retrieval+for+image%2Dbased+localization+revisited%2E+In+BMVC%2C+2012%2E+10)
52. Johannes Lutz Schönberger and Jan-Michael Frahm. Structure-from-motion revisited. In CVPR, 2016. 7, 10 [find](https://scholar.google.com/scholar?q=Johannes+Lutz+Scho%CC%88nberger+and+Jan%2DMichael+Frahm%2E+Structure%2Dfrom%2Dmotion+revisited%2E+In+CVPR%2C+2016%2E+7%2C+10)
53. Johannes Lutz Schönberger, Enliang Zheng, Marc Pollefeys, and Jan-Michael Frahm. Pixelwise view selection for unstructured multi-view stereo. In ECCV, 2016. 7 [find](https://scholar.google.com/scholar?q=Johannes+Lutz+Scho%CC%88nberger%2C+Enliang+Zheng%2C+Marc+Pollefeys%2C+and+Jan%2DMichael+Frahm%2E+Pixelwise+view+selection+for+unstructured+multi%2Dview+stereo%2E+In+ECCV%2C+2016%2E+7)
54. Eli Shechtman and Michal Irani. Matching local self-similarities across images and videos. In CVPR, 2007. 3 [find](https://scholar.google.com/scholar?q=Eli+Shechtman+and+Michal+Irani%2E+Matching+local+self%2Dsimilarities+across+images+and+videos%2E+In+CVPR%2C+2007%2E+3)
55. Richard Sinkhorn and Paul Knopp. Concerning nonnegative matrices and doubly stochastic matrices. Pacific Journal of Mathematics, 1967. 2, 5 [find](https://scholar.google.com/scholar?q=Richard+Sinkhorn+and+Paul+Knopp%2E+Concerning+nonnegative+matrices+and+doubly+stochastic+matrices%2E+Pacific+Journal+of+Mathematics%2C+1967%2E+2%2C+5)
56. Bart Thomee, David A Shamma, Gerald Friedland, Benjamin Elizalde, Karl Ni, Douglas Poland, Damian Borth, and Li-Jia Li. [YFCC100M: The new data in multimedia research](https://md.research.elamur.ai/free/en/papers/thomee-2015-yfcc100m-new-data-multimedia-research/). Communications of the ACM, 59(2):64–73, 2016. 7, 10
57. Lorenzo Torresani, Vladimir Kolmogorov, and Carsten Rother. Feature correspondence via graph matching: Models and global optimization. In ECCV, 2008. 2 [find](https://scholar.google.com/scholar?q=Lorenzo+Torresani%2C+Vladimir+Kolmogorov%2C+and+Carsten+Rother%2E+Feature+correspondence+via+graph+matching%3A+Models+and+global+optimization%2E+In+ECCV%2C+2008%2E+2)
58. Tomasz Trzcinski, Jacek Komorowski, Lukasz Dabala, Konrad Czarnota, Grzegorz Kurzejamski, and Simon Lynen. SConE: Siamese constellation embedding descriptor for image matching. In ECCV Workshops, 2018. 3 [find](https://scholar.google.com/scholar?q=Tomasz+Trzcinski%2C+Jacek+Komorowski%2C+Lukasz+Dabala%2C+Konrad+Czarnota%2C+Grzegorz+Kurzejamski%2C+and+Simon+Lynen%2E+SConE%3A+Siamese+constellation+embedding+descriptor+for+image+matching%2E+In+ECCV+Workshops%2C+2018)
59. Tinne Tuytelaars and Luc J Van Gool. Wide baseline stereo matching based on local, affinely invariant regions. In BMVC, 2000. 2 [find](https://scholar.google.com/scholar?q=Tinne+Tuytelaars+and+Luc+J+Van+Gool%2E+Wide+baseline+stereo+matching+based+on+local%2C+affinely+invariant+regions%2E+In+BMVC%2C+2000%2E+2)
60. Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. Instance normalization: The missing ingredient for fast stylization. arXiv:1607.08022, 2016. 2, 5 — [arXiv:1607.08022](https://arxiv.org/abs/1607.08022) [find](https://scholar.google.com/scholar?q=Dmitry+Ulyanov%2C+Andrea+Vedaldi%2C+and+Victor+Lempitsky%2E+Instance+normalization%3A+The+missing+ingredient+for+fast+stylization%2E+arXiv%3A1607%2E08022%2C+2016%2E+2%2C+5+%E2%80%94+%5BarXiv%3A1607%2E08022%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F1607%2E)
61. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. [Attention is all you need](https://md.research.elamur.ai/free/en/papers/vaswani-2017-attention-all-need/). In NIPS, 2017. 1, 2, 3, 4, 5
62. Petar Velikovi, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Li, and Yoshua Bengio. Graph attention networks. In ICLR, 2018. 2 [find](https://scholar.google.com/scholar?q=Petar+Velikovi%2C+Guillem+Cucurull%2C+Arantxa+Casanova%2C+Adriana+Romero%2C+Pietro+Li%2C+and+Yoshua+Bengio%2E+Graph+attention+networks%2E+In+ICLR%2C+2018%2E+2)
63. Cédric Villani. Optimal transport: old and new, volume 338. Springer Science & Business Media, 2008. 2 [find](https://scholar.google.com/scholar?q=Ce%CC%81dric+Villani%2E+Optimal+transport%3A+old+and+new%2C+volume+338%2E+Springer+Science+%26+Business+Media%2C+2008%2E+2)
64. Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In CVPR, 2018. 2 [find](https://scholar.google.com/scholar?q=Xiaolong+Wang%2C+Ross+Girshick%2C+Abhinav+Gupta%2C+and+Kaiming+He%2E+Non%2Dlocal+neural+networks%2E+In+CVPR%2C+2018%2E+2)
65. Yue Wang and Justin M Solomon. Deep Closest Point: Learning representations for point cloud registration. In ICCV, 2019. 2 [find](https://scholar.google.com/scholar?q=Yue+Wang+and+Justin+M+Solomon%2E+Deep+Closest+Point%3A+Learning+representations+for+point+cloud+registration%2E+In+ICCV%2C+2019%2E+2)
66. Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E. Sarma, Michael M. Bronstein, and Justin M. Solomon. Dynamic Graph CNN for learning on point clouds. ACM Transactions on Graphics, 2019. 2 [find](https://scholar.google.com/scholar?q=Yue+Wang%2C+Yongbin+Sun%2C+Ziwei+Liu%2C+Sanjay+E%2E+Sarma%2C+Michael+M%2E+Bronstein%2C+and+Justin+M%2E+Solomon%2E+Dynamic+Graph+CNN+for+learning+on+point+clouds%2E+ACM+Transactions+on+Graphics%2C+2019%2E+2)
67. Jianxiong Xiao, Andrew Owens, and Antonio Torralba. SUN3D: A database of big spaces reconstructed using SfM and object labels. In ICCV, 2013. 11 [find](https://scholar.google.com/scholar?q=Jianxiong+Xiao%2C+Andrew+Owens%2C+and+Antonio+Torralba%2E+SUN3D%3A+A+database+of+big+spaces+reconstructed+using+SfM+and+object+labels%2E+In+ICCV%2C+2013%2E+11)
68. Tsun-Yi Yang, Duy-Kien Nguyen, Huub Heijnen, and Vassileios Balntas. UR2KiD: Unifying retrieval, keypoint detection, and keypoint description without local correspondence supervision. arXiv:2001.07252, 2020. 10 — [arXiv:2001.07252](https://arxiv.org/abs/2001.07252) [find](https://scholar.google.com/scholar?q=Tsun%2DYi+Yang%2C+Duy%2DKien+Nguyen%2C+Huub+Heijnen%2C+and+Vassileios+Balntas%2E+UR2KiD%3A+Unifying+retrieval%2C+keypoint+detection%2C+and+keypoint+description+without+local+correspondence+supervision%2E+arXiv%3A2001%2E07252)
69. Kwang Moo Yi, Eduard Trulls, Vincent Lepetit, and Pascal Fua. LIFT: Learned invariant feature transform. In ECCV, 2016. 2, 6 [find](https://scholar.google.com/scholar?q=Kwang+Moo+Yi%2C+Eduard+Trulls%2C+Vincent+Lepetit%2C+and+Pascal+Fua%2E+LIFT%3A+Learned+invariant+feature+transform%2E+In+ECCV%2C+2016%2E+2%2C+6)
70. Manzil Zaheer, Satwik Kottur, Siamak Ravanbakhsh, Barnabas Poczos, Ruslan R Salakhutdinov, and Alexander J Smola. Deep sets. In NIPS, 2017. 2 [find](https://scholar.google.com/scholar?q=Manzil+Zaheer%2C+Satwik+Kottur%2C+Siamak+Ravanbakhsh%2C+Barnabas+Poczos%2C+Ruslan+R+Salakhutdinov%2C+and+Alexander+J+Smola%2E+Deep+sets%2E+In+NIPS%2C+2017%2E+2)
71. Jiahui Zhang, Dawei Sun, Zixin Luo, Anbang Yao, Lei Zhou, Tianwei Shen, Yurong Chen, Long Quan, and Hongen Liao. Learning two-view correspondences and geometry using order-aware network. In ICCV, 2019. 2, 5, 6, 10, 11, 12 [find](https://scholar.google.com/scholar?q=Jiahui+Zhang%2C+Dawei+Sun%2C+Zixin+Luo%2C+Anbang+Yao%2C+Lei+Zhou%2C+Tianwei+Shen%2C+Yurong+Chen%2C+Long+Quan%2C+and+Hongen+Liao%2E+Learning+two%2Dview+correspondences+and+geometry+using+order%2Daware+network%2E+In+ICCV%2C+2019)
72. Li Zhang, Xiangtai Li, Anurag Arnab, Kuiyuan Yang, Yunhai Tong, and Philip HS Torr. Dual graph convolutional network for semantic segmentation. In BMVC, 2019. 2 [find](https://scholar.google.com/scholar?q=Li+Zhang%2C+Xiangtai+Li%2C+Anurag+Arnab%2C+Kuiyuan+Yang%2C+Yunhai+Tong%2C+and+Philip+HS+Torr%2E+Dual+graph+convolutional+network+for+semantic+segmentation%2E+In+BMVC%2C+2019%2E+2)
73. Yimeng Zhang, Zhaoyin Jia, and Tsuhan Chen. Image retrieval with geometry-preserving visual phrases. In CVPR, 2011. 3 [find](https://scholar.google.com/scholar?q=Yimeng+Zhang%2C+Zhaoyin+Jia%2C+and+Tsuhan+Chen%2E+Image+retrieval+with+geometry%2Dpreserving+visual+phrases%2E+In+CVPR%2C+2011%2E+3)

*Updated 2 November 2026*