Pith. sign in

REVIEW 4 major objections 3 minor 42 references

CasP: Improving Semi-Dense Feature Matching Pipeline Leveraging Cascaded Correspondence Priors for Guidance

T0 review · 4 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims top-$k$ one-to-many priors at $1/16$ scale can localize the $1/8$ one-to-one search, yielding a matcher $\sim 2.2\times$ faster than ELoFTR with stronger cross-domain accuracy.

desk verdict The cascaded-prior pipeline is a real efficiency advance with a clean ablation, but the unmeasured recall of the fixed top-k prior stage is a genuine soft spot that needs addressing before this is fully convincing. read the letter →

arxiv 2507.17312 v2 pith:XGLED3U2 submitted 2025-07-23 cs.CV

Peiqi Chen , Lei Yu , Yi Wan , Yingying Pei , Xinyi Liu , Yongxiang Yao , Yingying Zhang , Lixiang Ru
show 4 more authors
Liheng Zhong Jingdong Chen Ming Yang Yongjun Zhang
This is my paper · ORCID
classification cs.CV
keywords semi-densefeaturematchingcascadedcorrespondencepriorsone-to-manyregion-basedselectivecross-attentionefficientrelativeposeestimationcross-domaingeneralizationvisuallocalization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that the matching stage — not the feature encoder — is the real efficiency bottleneck of semi-dense feature matching, and that the bottleneck can be removed without an accuracy penalty. CasP first computes coarse one-to-many correspondence priors at $1/16$ scale (the top-$k$ most similar tokens per token, with $k=8$), then restricts the one-to-one search at $1/8$ scale to those prior areas, bridged by a region-based selective cross-attention that sharpens feature discriminability among candidates. A training-inference decoupling applies the expensive dual-softmax operator only during training, replacing it with cheap partial softmax over the priors at inference. If the claim is right, it matters because the leading efficiency trick in prior work (dropping dual-softmax) measurably hurts accuracy, whereas CasP reports higher accuracy alongside $\sim$2.2$\times$/3.2$\times$ speedups over ELoFTR/LoFTR at 1152 resolution, with gains that grow as resolution rises — directly relevant to latency-sensitive SLAM and UAV image-matching.

What carries the argument

The load-bearing object is the cascaded correspondence prior: the top-$k$ one-to-many sets $\pi^A_{1/16}$ and $\pi^B_{1/16}$ obtained by row- and column-wise top-$k$ selection from the $1/16$-scale score matrix (Eq. 5). These sets convert global search into local search, and Eq. (14) makes mutual membership in the prior sets the acceptance condition for every final match, so the priors are at once the efficiency device and the accuracy ceiling. Around them sit three supporting mechanisms: RSCA (region-based selective cross-attention), which splits the $1/8$ feature map into cells and lets each query attend only to the $k\cdot r^2$ cells inside its priors; partial softmax (Eq. 12), which computes confidence only over prior positions at inference; and the training-inference decoupling, which keeps the full dual-softmax as a supervisory signal during training and omits it at inference. The hybrid interaction module (aggregated attention plus cross-CoC at $1/16$ and $1/32$ scales) and the two-stage homography refinement complete the pipeline.

What would settle it

On pairs with ground-truth poses and depth, rank the true $1/8$ correspondence inside the $1/16$ score matrix and check whether the top-8 set contains it: if a meaningful share of valid matches fall outside the top-8, the prior assumption caps achievable recall and increasing $k$ should recover those matches; if failures persist for matches already inside the top-8, the loss sits in RSCA, partial softmax, or refinement instead, and enlarging the prior set would not help.

Watch

Extended reading notes

Core claim

CasP establishes that a semi-dense matcher can get its search cost from global to local by making the coarse stage produce guidance rather than final matches. One-to-many matching at $1/16$ scale selects, for each token, the top-$k$ ($k=8$) positions in the other view via the score matrix (Eq. 5); one-to-one matching at $1/8$ scale then scores each query only against the $r^2$ sub-cells of its priors, with Eq. (14) guaranteeing an accepted match $(i,j)$ only when $j$ lies in the prior set of $i$ and $i$ in the prior set of $j$. Because each query attends to $k\cdot r^2 = 32$ candidates instead of the whole map, and because the pipeline can use a much cheaper low-level backbone (0.8M parameters in the lite model), the match-stage cost drops sharply while the coarse priors keep the search geometrically plausible. The paper supports this with pose-estimation AUC gains on indoor cross-domain benchmarks (58.7 versus 53.6 for ELoFTR on SN-1500 at 20°) and a lite-model runtime of 108 ms versus 238 ms for ELoFTR on MD-1500, alongside homography and visual localization results.

Load-bearing premise

The pipeline presumes that every true $1/8$-scale correspondence is among the top-8 priors chosen at $1/16$ scale: Eq. (14) rejects any match whose partner is outside the prior set on either side, so if coarse ranking misses the true correspondence (large scale change, occlusion, or repetitive texture), no later stage can recover it, and the fixed $k=8$ never adapts to the difficulty of the pair.

Editorial extensions

If this is right

  • Matching-stage cost scales with the number of retained priors rather than the full feature map, so the speedup over single-stage global-search matchers widens as input resolution increases.
  • A lite variant with 0.8M low-level parameters runs about 2.2x faster than ELoFTR and 3.2x faster than LoFTR at 1152 resolution in FP32, with further gains under FP16, which is directly usable in latency-sensitive systems.
  • The common shortcut of deleting the dual-softmax operator to gain speed is shown to degrade accuracy, while substituting cascaded prior guidance improves accuracy and runtime at once.
  • Cross-domain pose estimation on indoor scenes improves substantially (AUC@20° of 58.7 versus 53.6 for ELoFTR on SN-1500, with consistent gains on ETH3D), indicating better transfer than single-stage global search.
  • The full model matches or exceeds the accuracy of the strongest semi-dense and near-dense baselines on relative pose, homography (HPatches), and visual localization while using far less runtime.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because Eq. (14) makes the top-8 prior the hard acceptance gate, an adaptive $k$ that grows with scene ambiguity (for instance, where $1/16$ score peaks are flat or multi-modal) is the most direct lever on the accuracy ceiling the paper leaves untested.
  • The same two-phase prior structure could be transplanted into dense matchers (kernelized or correlation-based) that currently pay full-map search costs, possibly transferring the resolution-growing speedup to a different accuracy regime.
  • The training-inference decoupling suggests dual-softmax acts mainly as a learning signal rather than a filtering device; a top-$k$-aware loss trained without any softmax might remove the train/test mismatch and simplify the pipeline further.
  • The cross-domain gains hint that prior-restricted search acts as geometric regularization; a direct test would be comparing CasP against a global-search variant with identical capacity on texture-sparse indoor scenes, isolating the prior constraint as the cause of the transfer.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 3 minor

Summary. CasP proposes a semi-dense feature matching pipeline that decomposes matching into two phases: a coarse one-to-many search at 1/16 resolution that selects top-k correspondence priors, followed by a one-to-one matching stage at 1/8 resolution whose search range is restricted to those priors. A region-based selective cross-attention (RSCA) module is inserted between the phases to enhance features at prior positions, and a partial-softmax operator reduces inference cost. The training procedure uses dual-softmax supervision and injects ground-truth correspondences into the one-to-many confidence matrix, while inference omits the dual-softmax and relies on raw score top-k selection. The paper reports state-of-the-art or competitive accuracy on MD-1500, SN-1500, HPatches, Aachen, InLoc, and ETH3D, with speedups of roughly 2.2x over ELoFTR at 1152x1152 for the lite model.

Significance. If the central claim holds, CasP addresses a real bottleneck: global dual-softmax matching at fine scales dominates runtime at high resolution. The idea of cascaded one-to-many priors is well motivated, and the ablations on ELoFTR with and without the cascaded matching module provide a clean attribution of the gains to the proposed pipeline. The evaluation is broad, the code is promised to be released, and the efficiency numbers are reported on standard benchmarks. The main risk is that the one-to-many recall at inference is never measured, leaving a potential training-inference gap unquantified; this is fixable and does not invalidate the overall approach.

major comments (4)
  1. [Sec. 3.3 and Fig. 4] The training procedure injects ground-truth correspondences into the one-to-many stage, and Fig. 4 states that during training the top-k priors include ground-truth correspondences for RSCA learning. At inference, the priors are selected purely from the raw score matrix S1/16. This creates an oracle training signal: RSCA is trained on prior sets that always contain the true 1/8-scale match, while at inference a true match that is not in the top-k of S1/16 is unrecoverable. The paper reports final pose AUC and ablation numbers, but never reports the recall of the one-to-many stage, i.e., the fraction of true correspondences whose 1/8-scale match lies inside the top-k priors selected without ground-truth injection. Please report this recall on the evaluation datasets, and preferably compare results with and without the ground-truth injection during training. If the recall is high (e.g., >99%), the concern is largely resolved; if it is not, the central claim of accuracy preservation needs qualification.
  2. [Sec. 3.3, Eq. (14)] Equation (14) makes it explicit that a match is accepted only if the true correspondence lies in both the query-side and key-side prior sets. The choice k=8 is fixed and no sensitivity analysis is provided. The paper gives a geometric lower-bound argument (k >= 4) but no empirical evidence that k=8 covers the true correspondence under scale changes larger than 2x, repetitive texture, or occlusion. Please add an ablation varying k (e.g., 4, 8, 16, 32) and report the one-to-many recall as a function of k. Without this, the accuracy claims are contingent on an unexamined assumption about the top-k prior coverage.
  3. [Sec. 4.2, Table 2] The central efficiency claim rests on average runtime numbers, but the paper does not specify the number of runs, variance, warm-up, or whether the baseline runtimes were measured on the same machine and software stack. The runtime for AffineFormer is explicitly inferred from the original paper, which makes the comparison inhomogeneous. Please provide a precise measurement protocol, including repeated runs with standard deviation, and clearly distinguish measured versus cited runtimes. This is important because the speedup is the paper's headline contribution.
  4. [Sec. 3.3 and Sec. 3.5] The one-to-one matching stage uses full dual-softmax during training but partial softmax during inference. The paper does not analyze whether this decoupling changes confidence calibration or the effect of the threshold theta on the final match set. Please specify the value of theta used in all experiments and, if possible, report the sensitivity of pose AUC to theta. This is a smaller point than the prior-recall issue, but it is a necessary detail for reproducibility.
minor comments (3)
  1. [Sec. 4.1] The threshold theta and the window size w are defined in Sec. 3.3 and Sec. 3.4, but only w is given in the implementation details. Please state the value of theta used for all experiments.
  2. [Sec. 4.2] The evaluation protocol says images are resized to align with the corresponding accuracy metrics, but the exact target resolutions for each benchmark are not stated. Please specify the resizing rules for MD-1500, SN-1500, and the zero-shot ETH3D benchmarks.
  3. [Table 2] The table contains formatting artifacts such as broken lines and placeholder underlines in the sparse-method rows. These should be cleaned up in the camera-ready version.

Circularity Check

0 steps flagged · score 2.0 of 10

No load-bearing circularity: the central cascaded-matching claim is externally benchmarked; EcoMatcher self-citation and teacher forcing are disclosed and non-reductive.

full rationale

The derivation chain is self-contained. The cascaded pipeline defines one-to-many priors as top-k selections from S1/16 (Eq. 5) and one-to-one matches as partial softmax restricted to those priors (Eqs. 12-14); this is a method definition, not a reduction of output to input. Accuracy and speed are evaluated on external benchmarks (MD-1500, SN-1500, HPatches, Aachen, InLoc, ETH3D) with standard estimators, so the reported AUCs do not reduce to any parameter fitted to those benchmarks. The only self-citation for EcoMatcher [7] supports context-cluster and cross-CoC interaction components, which are not the central claim; the ablation in Table 6 (EL+CM) isolates the cascaded matching module on ELoFTR, providing independent support. The training-time injection of ground-truth correspondences into P1/16 (Sec. 3.3) is clearly disclosed teacher forcing; it does not by construction force inference-time top-k priors from S1/16 alone to contain the truth, so it is a train/inference distribution shift rather than a circular fit. The unreported recall of the one-to-many prior stage (whether Eq. 14's top-k coverage contains true matches) is a legitimate robustness concern, not a circularity. Score 2 reflects the minor EcoMatcher self-citation and the disclosed teacher-forcing step, neither load-bearing.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The method's empirical claims rest on hand-chosen hyperparameters (k, loss weights, window size, threshold) and on the coarse-prior coverage assumption. There are no free parameters fitted to benchmark targets in the sense of a derivation, but exact reproduction requires the omitted threshold and sensitivity checks.

free parameters (5)
  • k (number of correspondence priors) = 8
    Chosen by hand in Sec. 4.1; controls the recall and computation trade-off of the entire second matching phase.
  • Loss weights (lambda1..lambda4) = 0.5, 0.5, 0.25, 1.0
    Set in Sec. 4.1 without a sensitivity study; they determine the balance of coarse and fine supervision.
  • Window size w = 5
    Set in Sec. 4.1 for local patch extraction in the refinement module.
  • Confidence threshold theta = not reported
    Used in Sec. 3.3 to filter coarse matches; its value is omitted, which is a gap for exact reproduction.
  • Interaction and RSCA repeats N1/16, N1/8 = 2, 2
    Architecture hyperparameters chosen in Sec. 4.1 without an ablation on their values.
assumptions (3)
  • domain assumption Each token at 1/16 scale corresponds to at most (16/8)^2 = 4 tokens in the 1/8 feature map, so k >= 4 priors can cover the true match.
    Stated in Sec. 3.3; if large scale changes or occlusions push the true match outside the top-k, Eq. (14) cannot recover it.
  • standard math Partial softmax over prior positions is a valid confidence measure for supervision and filtering.
    Used in Eqs. (12)-(13) to define P1/8 and the coarse loss; it is a standard normalization with no calibration evidence.
  • domain assumption Models trained on MegaDepth outdoor data transfer to indoor and zero-shot scenes.
    All cross-domain claims in Sec. 4 rely on this; generalization is measured but not derived.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CasP: Improving Semi-Dense Feature Matching Pipeline Leveraging Cascaded Correspondence Priors for Guidance." pith.science (2026). https://pith.science/paper/XGLED3U2

@misc{pith2026250717312,
  author       = {Pith},
  title        = {Pith review of: CasP: Improving Semi-Dense Feature Matching Pipeline Leveraging Cascaded Correspondence Priors for Guidance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XGLED3U2}},
  note         = {Machine review of arXiv:2507.17312}
}
abstract

Semi-dense feature matching methods have shown strong performance in challenging scenarios. However, the existing pipeline relies on a global search across the entire feature map to establish coarse matches, limiting further improvements in accuracy and efficiency. Motivated by this limitation, we propose a novel pipeline, CasP, which leverages cascaded correspondence priors for guidance. Specifically, the matching stage is decomposed into two progressive phases, bridged by a region-based selective cross-attention mechanism designed to enhance feature discriminability. In the second phase, one-to-one matches are determined by restricting the search range to the one-to-many prior areas identified in the first phase. Additionally, this pipeline benefits from incorporating high-level features, which helps reduce the computational costs of low-level feature extraction. The acceleration gains of CasP increase with higher resolution, and our lite model achieves a speedup of $\sim2.2\times$ at a resolution of 1152 compared to the most efficient method, ELoFTR. Furthermore, extensive experiments demonstrate its superiority in geometric estimation, particularly with impressive cross-domain generalization. These advantages highlight its potential for latency-sensitive and high-robustness applications, such as SLAM and UAV systems. Code is available at https://github.com/pq-chen/CasP.

Figures

Figures reproduced from arXiv: 2507.17312 by the authors.

Figure 1
Figure 1. Matching Accuracy and Efficiency Comparisons. The runtime and AUC@20° accuracy are measured across two bench￾marks and compared with LoFTR and ELoFTR (EL). The acceler￾ation gains of ELoFTR diminish with increasing resolution since the matching stage occupies a significant portion of runtime. The novel cascaded matching pipeline is more efficient and robust than merely removing the dual-softmax (DS) operator. cal pi… view at source ↗
Figure 2
Figure 2. Schematic Diagram of Cascaded Matching. (a) One￾to-many correspondence priors are selected with global search at a coarse scale and shown as same-colored patches (purple denotes potential common priors). (b) One-to-one matches are determined with prior guidance at the target scale and shown as same-numbered patches, with a region-based attention mechanism, RSCA, applied only at prior token positions. ducing enhanced… view at source ↗
Figure 3
Figure 3. Pipeline Overview. (1) Low-level (1/2 to 1/8) and high-level (1/16 to 1/32) feature maps are extracted for local and global descriptions, respectively. (2) High-level feature maps are transformed by a hybrid interaction module composed of sub-modules that operate at different scales to enhance complementarity. (3) A cascaded matching module establishes one-to-one correspondences at the 1/8 scale, with the search ran… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Training-Inference Decoupling Cascaded Matching. During training, the top-k priors selected by one-to-many matching include ground-truth correspondences for RSCA learning, and the DS operator is applied in both matching stages for supervision. During inference, one-to-…
Figure 5
Figure 5. Figure 5: Qualitative Results. Two challenging image pairs are selected for qualitative analysis and compared with LoFTR and ELoFTR. One pair lacks texture details due to illumination changes, while the other undergoes significant viewpoint variations. Method AUC on HPatches ↑ @…
Figure 6
Figure 6. Figure 6: Visualization of One-to-Many Matching. For each token, k = 8 priors are displayed in the corresponding image. additional advanced modules in our method further enhances accuracy. 4) Our pipeline requires fewer GMACs, delivers faster runtime, and uses less memory on MD-…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 40 canonical work pages

  1. [1]

    Hpatches: A benchmark and evaluation of handcrafted and learned local descriptors

    Vassileios Balntas, Karel Lenc, Andrea Vedaldi, and Krystian Mikolajczyk. Hpatches: A benchmark and evaluation of handcrafted and learned local descriptors. In CVPR, pages 5173–5182, 2017. 7

    work page 2017

  2. [2]

    Magsac++, a fast, reliable and accurate robust estima- tor

    Daniel Barath, Jana Noskova, Maksym Ivashechkin, and Jiri Matas. Magsac++, a fast, reliable and accurate robust estima- tor. In CVPR, pages 1304–1312, 2020. 8

    work page 2020

  3. [3]

    Improving transformer-based image matching by cascaded capturing spatially informative keypoints

    Chenjie Cao and Yanwei Fu. Improving transformer-based image matching by cascaded capturing spatially informative keypoints. In ICCV, pages 12129–12139, 2023. 3

    work page 2023

  4. [4]

    Learning to match features with seeded graph matching network

    Hongkai Chen, Zixin Luo, Jiahui Zhang, Lei Zhou, Xuyang Bai, Zeyu Hu, Chiew-Lan Tai, and Long Quan. Learning to match features with seeded graph matching network. In ICCV, pages 6301–6310, 2021. 2

    work page 2021

  5. [5]

    Aspanformer: Detector-free image matching with adaptive span transformer

    Hongkai Chen, Zixin Luo, Lei Zhou, Yurun Tian, Mingmin Zhen, Tian Fang, David Mckinnon, Yanghai Tsin, and Long Quan. Aspanformer: Detector-free image matching with adaptive span transformer. In ECCV, pages 20–36, 2022. 1, 3, 4, 6, 7

    work page 2022

  6. [6]

    Affine-based deformable attention and selective fusion for semi-dense matching

    Hongkai Chen, Zixin Luo, Yurun Tian, Xuyang Bai, Ziyu Wang, Lei Zhou, Mingmin Zhen, Tian Fang, David McKin- non, Yanghai Tsin, et al. Affine-based deformable attention and selective fusion for semi-dense matching. In CVPRW, pages 4254–4263, 2024. 1, 3, 6, 7

    work page 2024

  7. [7]

    Ecomatcher: Efficient clustering oriented matcher for detector-free image matching

    Peiqi Chen, Lei Yu, Yi Wan, Yongjun Zhang, Jian Wang, Liheng Zhong, Jingdong Chen, and Ming Yang. Ecomatcher: Efficient clustering oriented matcher for detector-free image matching. In ECCV, pages 344–360, 2024. 2, 3, 4

    work page 2024

  8. [8]

    Scannet: Richly- annotated 3d reconstructions of indoor scenes

    Angela Dai, Angel X Chang, Manolis Savva, Maciej Halber, Thomas Funkhouser, and Matthias Nießner. Scannet: Richly- annotated 3d reconstructions of indoor scenes. In CVPR, pages 5828–5839, 2017. 6

    work page 2017

Show all 42 references
  1. [9]

    Superpoint: Self-supervised interest point detection and description

    Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabi- novich. Superpoint: Self-supervised interest point detection and description. In CVPRW, pages 224–236, 2018. 1, 2, 6, 7

    2018

  2. [10]

    Repvgg: Making vgg-style convnets great again

    Xiaohan Ding, Xiangyu Zhang, Ningning Ma, Jungong Han, Guiguang Ding, and Jian Sun. Repvgg: Making vgg-style convnets great again. In CVPR, pages 13733–13742, 2021. 3

    2021

  3. [11]

    Dkm: Dense kernelized feature matching for geometry estimation

    Johan Edstedt, Ioannis Athanasiadis, M˚arten Wadenb¨ack, and Michael Felsberg. Dkm: Dense kernelized feature matching for geometry estimation. In CVPR, pages 17765–17775, 2023. 6, 7

    2023

  4. [12]

    Roma: Robust dense feature matching

    Johan Edstedt, Qiyu Sun, Georg B¨okman, M˚arten Wadenb¨ack, and Michael Felsberg. Roma: Robust dense feature matching. In CVPR, pages 19790–19800, 2024. 6

    2024

  5. [13]

    Top- icfm: Robust and interpretable topic-assisted feature match- ing

    Khang Truong Giang, Soohwan Song, and Sungho Jo. Top- icfm: Robust and interpretable topic-assisted feature match- ing. In AAAI, pages 2447–2455, 2023. 2, 7

    2023

  6. [14]

    Silk: Simple learned keypoints

    Pierre Gleize, Weiyao Wang, and Matt Feiszli. Silk: Simple learned keypoints. In ICCV, pages 22499–22508, 2023. 2

    2023

  7. [15]

    Detector-free struc- ture from motion

    Xingyi He, Jiaming Sun, Yifan Wang, Sida Peng, Qixing Huang, Hujun Bao, and Xiaowei Zhou. Detector-free struc- ture from motion. In CVPR, pages 21594–21603, 2024. 1

    2024

  8. [16]

    Adaptive assignment for geometry aware local feature match- ing

    Dihe Huang, Ying Chen, Yong Liu, Jianlin Liu, Shang Xu, Wenlong Wu, Yikang Ding, Fan Tang, and Chengjie Wang. Adaptive assignment for geometry aware local feature match- ing. In CVPR, pages 5425–5434, 2023. 4

    2023

  9. [17]

    Transformers are rnns: Fast autoregressive transformers with linear attention

    Angelos Katharopoulos, Apoorv Vyas, Nikolaos Pappas, and Franc ¸ois Fleuret. Transformers are rnns: Fast autoregressive transformers with linear attention. In ICML, pages 5156– 5165, 2020. 2

    2020

  10. [18]

    Megadepth: Learning single- view depth prediction from internet photos

    Zhengqi Li and Noah Snavely. Megadepth: Learning single- view depth prediction from internet photos. In CVPR, pages 2041–2050, 2018. 6

    2018

  11. [19]

    Pixel-perfect structure-from-motion with featuremetric refinement

    Philipp Lindenberger, Paul-Edouard Sarlin, Viktor Larsson, and Marc Pollefeys. Pixel-perfect structure-from-motion with featuremetric refinement. In CVPR, pages 5987–5997, 2021. 1

    2021

  12. [20]

    Lightglue: Local feature matching at light speed

    Philipp Lindenberger, Paul-Edouard Sarlin, and Marc Polle- feys. Lightglue: Local feature matching at light speed. In ICCV, pages 17627–17638, 2023. 1, 2, 6

    2023

  13. [21]

    Distinctive image features from scale- invariant keypoints

    David G Lowe. Distinctive image features from scale- invariant keypoints. IJCV, 60:91–110, 2004. 1, 2

    2004

  14. [22]

    Image as set of points

    Xu Ma, Yuqian Zhou, Huan Wang, Can Qin, Bin Sun, Chang Liu, and Yun Fu. Image as set of points. In ICLR, 2023. 3

    2023

  15. [23]

    Pats: Patch area transportation with subdivision for local feature matching

    Junjie Ni, Yijin Li, Zhaoyang Huang, Hongsheng Li, Hujun Bao, Zhaopeng Cui, and Guofeng Zhang. Pats: Patch area transportation with subdivision for local feature matching. In CVPR, pages 17776–17786, 2023. 3

    2023

  16. [24]

    From coarse to fine: Robust hierarchical localization at large scale

    Paul-Edouard Sarlin, Cesar Cadena, Roland Siegwart, and Marcin Dymczyk. From coarse to fine: Robust hierarchical localization at large scale. In CVPR, pages 12716–12725,

  17. [25]

    Superglue: Learning feature match- ing with graph neural networks

    Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Superglue: Learning feature match- ing with graph neural networks. In CVPR, pages 4938–4947,

  18. [26]

    Back to the feature: Learning robust camera localization from pixels to pose

    Paul-Edouard Sarlin, Ajaykumar Unagar, M ˚ans Larsson, Hugo Germain, Carl Toft, Viktor Larsson, Marc Pollefeys, Vincent Lepetit, Lars Hammarstrand, Fredrik Kahl, and Torsten Sattler. Back to the feature: Learning robust camera localization from pixels to pose. In CVPR, pages 3...

  19. [27]

    Benchmarking 6dof outdoor visual localization in changing conditions

    Torsten Sattler, Will Maddern, Carl Toft, Akihiko Torii, Lars Hammarstrand, Erik Stenborg, Daniel Safari, Masatoshi Oku- tomi, Marc Pollefeys, Josef Sivic, et al. Benchmarking 6dof outdoor visual localization in changing conditions. In CVPR, pages 8601–8610, 2018. 7

    2018

  20. [28]

    Structure- from-motion revisited

    Johannes L Schonberger and Jan-Michael Frahm. Structure- from-motion revisited. In CVPR, pages 4104–4113, 2016. 1, 6 9

    2016

  21. [29]

    A multi-view stereo benchmark with high- resolution images and multi-camera videos

    Thomas Schops, Johannes L Schonberger, Silvano Galliani, Torsten Sattler, Konrad Schindler, Marc Pollefeys, and An- dreas Geiger. A multi-view stereo benchmark with high- resolution images and multi-camera videos. In CVPR, pages 3260–3269, 2017. 8

    2017

  22. [30]

    Gim: Learning generalizable image matcher from internet videos

    Xuelun Shen, Zhipeng Cai, Wei Yin, Matthias M¨uller, Zijun Li, Kaixuan Wang, Xiaozhi Chen, and Cheng Wang. Gim: Learning generalizable image matcher from internet videos. In ICLR, 2024. 8

    2024

  23. [31]

    Clustergnn: Cluster-based coarse-to- fine graph neural network for efficient feature matching

    Yan Shi, Jun-Xiong Cai, Yoli Shavit, Tai-Jiang Mu, Wensen Feng, and Kai Zhang. Clustergnn: Cluster-based coarse-to- fine graph neural network for efficient feature matching. In CVPR, pages 12517–12526, 2022. 2

    2022

  24. [32]

    Loftr: Detector-free local feature matching with transformers

    Jiaming Sun, Zehong Shen, Yuang Wang, Hujun Bao, and Xiaowei Zhou. Loftr: Detector-free local feature matching with transformers. In CVPR, pages 8922–8931, 2021. 1, 2, 4, 5, 6, 7

    2021

  25. [33]

    Inloc: Indoor visual localization with dense matching and view synthesis

    Hajime Taira, Masatoshi Okutomi, Torsten Sattler, Mircea Cimpoi, Marc Pollefeys, Josef Sivic, Tomas Pajdla, and Ak- ihiko Torii. Inloc: Indoor visual localization with dense matching and view synthesis. In CVPR, pages 7199–7209,

  26. [34]

    Quadtree attention for vision transformers

    Shitao Tang, Jiahui Zhang, Siyu Zhu, and Ping Tan. Quadtree attention for vision transformers. In ICLR, 2022. 1, 3, 4, 6

    2022

  27. [35]

    Long-term visual lo- calization revisited

    Carl Toft, Will Maddern, Akihiko Torii, Lars Hammarstrand, Erik Stenborg, Daniel Safari, Masatoshi Okutomi, Marc Polle- feys, Josef Sivic, Tomas Pajdla, et al. Long-term visual lo- calization revisited. IEEE TPAMI, 44(4):2074–2088, 2020. 1

    2020

  28. [36]

    Glu-net: Global-local universal network for dense flow and correspon- dences

    Prune Truong, Martin Danelljan, and Radu Timofte. Glu-net: Global-local universal network for dense flow and correspon- dences. In CVPR, pages 6258–6268, 2020. 3

    2020

  29. [37]

    Disk: Learning local features with policy gradient

    Michał Tyszkiewicz, Pascal Fua, and Eduard Trulls. Disk: Learning local features with policy gradient. In NeurIPS, pages 14254–14265, 2020. 1, 2

    2020

  30. [38]

    Matchformer: Interleaving attention in transformers for feature matching

    Qing Wang, Jiaming Zhang, Kailun Yang, Kunyu Peng, and Rainer Stiefelhagen. Matchformer: Interleaving attention in transformers for feature matching. In ACCV, pages 2746– 2762, 2022. 3, 4

    2022

  31. [39]

    Homomatcher: Dense feature matching results with semi-dense efficiency by homography estimation

    Xiaolong Wang, Lei Yu, Yingying Zhang, Jiangwei Lao, Lix- iang Ru, Liheng Zhong, Jingdong Chen, Yu Zhang, and Ming Yang. Homomatcher: Dense feature matching results with semi-dense efficiency by homography estimation. In AAAI, pages 7952–7960, 2025. 6

    2025

  32. [40]

    Efficient loftr: Semi-dense local feature matching with sparse-like speed

    Yifan Wang, Xingyi He, Sida Peng, Dongli Tan, and Xiaowei Zhou. Efficient loftr: Semi-dense local feature matching with sparse-like speed. In CVPR, pages 21666–21675, 2024. 2, 4, 5, 6, 7, 8

    2024

  33. [41]

    Alike: Accurate and lightweight keypoint detection and descriptor extraction

    Xiaoming Zhao, Xingming Wu, Jinyu Miao, Weihai Chen, Peter CY Chen, and Zhengguo Li. Alike: Accurate and lightweight keypoint detection and descriptor extraction. IEEE TMM, 25:3101–3112, 2022. 2

    2022

  34. [42]

    Aliked: A lighter keypoint and descriptor extraction network via deformable transforma- tion

    Xiaoming Zhao, Xingming Wu, Weihai Chen, Peter CY Chen, Qingsong Xu, and Zhengguo Li. Aliked: A lighter keypoint and descriptor extraction network via deformable transforma- tion. IEEE TIM, 72:1–16, 2023. 2 10

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.