SuperGlue: Learning Feature Matching With Graph Neural Networks
By Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz, Andrew Rabinovich. 2019
2,839 citations recorded by the source corpus.
This paper introduces SuperGlue, a neural network that matches two sets of local features by jointly finding correspondences and rejecting non-matchable points. Assignments are estimated by solving a differentiable optimal transport problem, whose costs are predicted by a graph neural network. We introduce a flexible context aggregation mechanism based on attention, enabling SuperGlue to reason about the underlying 3D scene and feature assignments jointly. Compared to traditional, hand-designed heuristics, our technique learns priors over geometric transformations and regularities of the 3D world through end-to-end training from image pairs. SuperGlue outperforms other learned approaches and achieves state-of-the-art results on the task of pose estimation in challenging real-world indoor and outdoor environments. The proposed method performs matching in real-time on a modern GPU and can be readily integrated into modern SfM or SLAM systems. The code and trained weights are publicly available at github.com/magicleap/SuperGluePretrainedNetwork.
Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition, Graph Learning
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Aerospace Engineering
- Marc Pollefeys
- Andreas Geiger
- Jitendra Malik
- Pascal Fua
- Eduard Trulls
- Jia Deng
- Kristen Grauman
- Luc Van Gool
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Dacheng Tao
- Jitendra Malik
- Ming-Hsuan Yang
- Trevor Darrell
- Kristen Grauman
- Xiaogang Wang
- Andrew Zisserman
In Graph Learning
- Bernard Ghanem
- Jia Deng
- Zachary Teed
- Marc Pollefeys
- Allan Jabri
- Dacheng Tao
- Antonio Torralba
- Long Quan
Citations in the corpus
- SimVLM: Simple Visual Language Model Pretraining with Weak Supervision (2021, background): “, 2017) has pushed the state of the art on a wide range of natural language processing (NLP) tasks (Rajpurkar et al., 2016; Wang et al., 2018; Sarlin et al., 2020).”
- Ego4D: Around the World in 3,000 Hours of Egocentric Video (2021, background): “[195] Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich.”
- Lifelong Graph Learning (2020, critical): “We next show that the performance of SuperGlue [36], which used a regular graph attention model for feature matching, can be simply improved by changing the graph-attention matcher to our proposed FGN.”
- COTR: Correspondence Transformer for Matching Across Images (2021, methodology): “The current state of the art builds on attentional graph neural networks [54].”
- Learning Target Candidate Association to Keep Track of What Not to Track (2021, methodology): “The adopted architecture [38] uses a Graph Neural Network (GNN) with message passing that sends the messages in an alternating fashion across self or cross edges to produce a new feature representation for each node after every layer.”
- GMFlow: Learning Optical Flow via Global Matching (2021, methodology): “The top-performing sparse methods (e.g., SuperGlue [34] and LoFTR [38]) adopt Transformers [41] to reason about the mutual relationship between feature descriptors, and the correspondences are extracted with an explicit matching layer (e.g., softmax layer [43]).”
- Space-Time Correspondence as a Contrastive Random Walk (2020, background): “by assuming soft bijection [18, 31, 115, 86] – it is unclear how they should extend to the multi-hop setting, where such heuristics may not always be desirable at each intermediate step.”
- CoFiNet: Reliable Coarse-to-fine Correspondences for Robust Point Cloud Registration (2021, methodology): “Recently, a coarse-to-fine mechanism is leveraged by our 2D counterparts [23, 24, 25] to avoid direct keypoint detection, which shows superiority over the state-of-the-art detection-based method [26].”
- Back to the Feature: Learning Robust Camera Localization from Pixels to Pose (2021, methodology): “The former matches dense deep descriptors and relies on a dense reference 3D model, while hloc matches SuperPoint [23] features with SuperGlue [71] and builds a sparse 3D SfM reference point cloud.”
- Continuous 3D Perception Model with Persistent State (2025, neutral): “…replacing hand-crafted components with learning-based alternatives (e.g., substituting conventional feature descriptors [6, 47] with learned ones [17, 20, 63, 73]), integrating data-driven priors [13, 77–79, 93, 97, 105] into the systems, or optimizing entire systems end-to-end [76, 78, 83, 94].”
- A review of multimodal image matching: Methods and applications (2021, methodology): “Apart from learning with multilayer perceptron (MLP), another method conducted this task with graph convolutional networks (GCN) [346].”
- iNeRF: Inverting Neural Radiance Fields for Pose Estimation (2020, methodology): “than a strong feature-based baseline, using SuperGlue [30].”
- LightGlue: Local Feature Matching at Light Speed (2023, methodology): “Following best practices in benchmarking [31] and unlike past works [56, 68], we use a state-of-the-art robust estimator [3] and extensively tune the inlier threshold for each method separately.”
- Image Matching from Handcrafted to Deep Features: A Survey (2020, background): “Sarlin et al. (2020) proposed SuperGlue, to match two sets of local features by jointly finding correspondences and rejecting non-matchable points.”
- Text Data Augmentation for Deep Learning (2021, background): “…about meaning versus form could be to look at recently developed benchmarks in language processing such as the distinction between GLUE [31] and SuperGLUE [32] tasks that predominantly test an understanding of form to knowledge-intensive tasks such as KILT [33] that better probe for meaning.”
- A Tale of Two Features: Stable Diffusion Complements DINO for Zero-Shot Semantic Correspondence (2023, background): “Deep models [14, 37, 42, 49, 63, 68] learn effective features for dense correspondence that can be robust to photometric and geometric changes such as rotation, scaling, or perspective transformation.”
- VGGT: Visual Geometry Grounded Transformer (2025, neutral): “Two-View matching comparison on ScanNet-1500 [9, 60] .”