Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Mind the Gap: Aligning Vision Foundation Models to Image Feature Matching

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

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

Pith's one-line read This paper claims that the gap between single-image foundation models and cross-image matching is the reason matchers fail on multi-instance scenes, and that a frozen diffusion model with cross-image prompting closes it.

desk verdict Solid incremental matching method with real gains on standard benchmarks, but the headline IMIM improvement is built on a precision-only metric that may not actually measure multi-instance matching. read the letter →

arxiv 2507.10318 v1 pith:PUBT6F6R submitted 2025-07-14 cs.CV

Yuhan Liu , Jingwen Fu , Yang Wu , Kangyi Wu , Pengna Li , Jiayi Wu , Sanping Zhou , Jingmin Xin This is my paper
classification cs.CV
keywords imagefeaturematchingvisionfoundationmodelsdiffusionmulti-instancecross-imageinteractionpromptconditioningIMIMbenchmarkrelativeposeestimation
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 argues that adapting vision foundation models to image feature matching has overlooked a misalignment: models like CLIP and DINOv2 are trained for single-image understanding and emphasize global semantics, whereas matching requires instance-level detail and cross-image relations. To close this gap, the authors propose IMD, which extracts features from a frozen text-to-image diffusion UNet and conditions each image's extraction on the other image through a Cross-image Interaction Prompt Module. They also build IMIM, a 100-pair multi-instance benchmark from BURST footage, and report 88.7% accuracy there versus 79.2% for CasMTR, alongside leading results on MegaDepth, ScanNet, HPatches, and InLoc. The paper's point is that generative foundation models, used this way, alleviate the multi-instance failure mode.

What carries the argument

The load-bearing object is the frozen text-to-image diffusion UNet used as a feature extractor, with the paper exploiting the generative model's prompt condition as a two-way tunnel between images. For each image, CIPM computes a personalized prompt by cross-attending that image's CLIP features to the other image's CLIP features, then feeds the prompt into the UNet; the resulting coarse features (1/8 resolution) go through interleaved self- and cross-attention, coarse matching with mutual-nearest-neighbour selection, and a fine encoder with subpixel refinement. The diffusion backbone supplies instance-level detail that contrastive backbones discard, while the prompt interaction gives the pair a shared reference frame without sacrificing discriminability.

What would settle it

Construct a larger multi-instance matching benchmark with more pairs, more object categories, and varied instance counts, and run IMD against CasMTR under identical training conditions; if the accuracy margin drops well below 12% or reverses on the larger set, the central claim of misalignment mitigation is not supported.

Watch

Extended reading notes

Core claim

The central claim is that the misalignment between single-image foundation-model features and the cross-image requirements of feature matching is real, diagnosable, and fixable. IMD replaces contrastive backbones with the frozen UNet of Stable Diffusion 2-1, feeding clean images at timestep t=0 and taking intermediate features, so that every instance receives an explicit representation rather than a global semantic summary. The Cross-image Interaction Prompt Module builds the prompt for image A from image B's features via cross-attention, and vice versa, making the extracted features correlated without sharing a single prompt that would blur instance identity. The authors claim this design yields state-of-the-art pose estimation and homography results and, most importantly, a 12% accuracy improvement over CasMTR on their new IMIM benchmark, which measures whether matches land on the correct instance when several instances of the same class appear.

Load-bearing premise

The paper's headline result depends on IMIM being a fair and representative measure of multi-instance matching; if its 100 hand-filtered BURST pairs or its mask-cropping protocol favor diffusion features, the 12% gain would not generalize.

Editorial extensions

If this is right

  • If IMD is right, future matchers should treat foundation-model choice as a matching-specific design decision, not a plug-in: generative features should be preferred over contrastive ones when instances repeat.
  • Because the UNet is frozen and only the prompt module and matching heads are trained, the approach offers a lightweight way to repurpose large text-to-image models for geometric tasks.
  • The IMIM benchmark provides a way to measure misalignment directly; adopting it as a standard evaluation would expose failures that single-object benchmarks hide.
  • The personalized prompt mechanism suggests that bidirectional conditioning, rather than a shared prompt, is the right way to get correlated yet discriminative features.

Reading between the lines

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

  • The paper's diagnosis implies a testable general claim: contrastive self-supervised features should be systematically worse than diffusion features at same-class, different-instance discrimination, independent of the matching head; one can probe this with instance-retrieval tasks on BURST-style videos.
  • The IMIM benchmark's 100 pairs and 10 categories is small, so the 12% margin should be re-measured on a larger sample before treating it as a stable effect size.
  • The prompt-as-tunnel idea could transfer to other pairwise tasks such as optical flow, stereo, and tracking, and to multi-frame settings where each frame conditions all others.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper identifies a 'misalignment' between single-image-oriented vision foundation models and the cross-image requirements of feature matching, and proposes IMD to close this gap. IMD extracts coarse features from a frozen Stable Diffusion UNet at timestep t=0, with a Cross-image Interaction Prompt Module (CIPM) that uses cross-attention between CLIP embeddings of the two images to generate personalized prompts for the UNet; the coarse features are then refined by self/cross-attention and a fine-level sub-pixel module. The authors also introduce IMIM, a 100-pair benchmark built from BURST videos containing multiple instances of the same object class. Experiments report state-of-the-art results on MegaDepth, ScanNet, HPatches, and InLoc, and an 88.7% score on IMIM versus 79.2% for CasMTR, which the abstract interprets as a 12% improvement and as evidence that IMD mitigates the misalignment.

Significance. If the IMIM evaluation is valid, this is a useful contribution: it demonstrates a concrete way to adapt generative foundation models to geometric matching, and the benchmark addresses a real gap, since existing matching benchmarks rarely require disambiguating multiple instances of the same category. The paper gives appropriate credit to prior diffusion-feature work and includes a parameter-controlled comparison in Fig. 4 that mitigates the obvious model-capacity confound. The gains on standard benchmarks are consistent and non-trivial, particularly the ScanNet AUC@5 improvement from 23.9 (PRISM) to 29.8 for a model trained only on MegaDepth, and the core pipeline is reproducible in principle given its use of standard coarse-to-fine matching components. The main weakness is that the headline claim of mitigating misalignment rests on a 100-pair benchmark with a precision-only, underspecified protocol; this needs strengthening before the claim can be accepted.

major comments (3)
  1. [Sec. 4.2, IMIM evaluation protocol] The IMIM metric is precision-only. The protocol counts N matches whose source point i lies in the source instance mask and reports M/N, where M is the subset whose target point j lies in the target instance mask. This is a precision score over the matches emitted from the source instance: it ignores false negatives entirely and is not normalized by the total number of ground-truth correspondences or by the number of matches the method outputs. A method that returns a single correct match scores 100%, while a method that finds the correct instance but also emits additional matches is penalized. The 88.7 vs 79.2 gap in Table 1 could therefore reflect a match-count or score-threshold trade-off rather than better instance discrimination. Please add a recall-oriented measure (e.g., fraction of ground-truth instance correspondences recovered), report the number of matches per method, or fix the match budget per pair.
  2. [Sec. 4.2, IMIM cropping and scoring] The protocol says ground-truth masks are used to crop instances from source and target images, but the exact use of the crops is not specified. If the images are cropped to the instance before matching, the multi-instance condition is removed and the benchmark would not test the claimed capability; if the masks are used only to score matches in full images, the coordinate system in which i and j are expressed after cropping must be defined. Please specify the preprocessing pipeline precisely and, ideally, provide the exact evaluation script with the benchmark release.
  3. [Sec. 4.2, Table 1] The statistical basis for the headline improvement is weak. IMIM contains only 100 pairs drawn from 50 videos across 10 categories, and no error bars, per-category breakdown, or significance test is reported. The '12% improvement' is a relative gain (88.7/79.2), not 12 percentage points; with 100 pairs this difference may be within sampling variability. Please report confidence intervals or bootstrap estimates, per-category results, and the number of matches emitted by each method.
minor comments (4)
  1. [Sec. 1] The sentence 'fail to generalize on unseen data fail to generalize on unseen data [21]' contains a duplicated phrase; please remove the repetition.
  2. [Table 4] Row 2 labels the backbone 'DINOv2 (B)' with reference [18], but [18] is the MAE paper; DINOv2 is reference [34]. Please correct the citation or clarify which baseline was actually used.
  3. [Table 3 and Sec. 4.2] There are small presentation issues: 'SP [12]+SG [42]CVPR'20' in Table 3 lacks a space before the year, and Sec. 4.2 contains the typos 'MegeDepth' and 'Segemntation'.
  4. [Sec. 4.2, Reproducibility] The paper does not state whether code or checkpoints will be released; given that IMIM is a new benchmark, releasing the evaluation script and pair/mask metadata would materially strengthen confidence in the reported numbers.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: IMD's external-benchmark results are self-contained; the IMIM-based misalignment claim rests on benchmark validity, not on a fitted or definitional reduction.

full rationale

The paper's derivation chain is not circular in any of the enumerated senses. IMD is trained on MegaDepth with the standard coarse-to-fine losses L = Lc + αLf1 + βLf2 (Sec. 3.5) and is evaluated on held-out external benchmarks (MegaDepth, ScanNet, HPatches, InLoc) under established protocols (Secs. 4.3-4.5), so the reported state-of-the-art results do not depend on the authors' proposed IMIM benchmark. IMIM is constructed from BURST as a multi-instance evaluation set (Sec. 4.2), and no IMD parameter is fitted to IMIM; IMIM scores appear only in the final evaluation table, not in any loss or selection criterion. The one same-author citation, SRMatcher [28], is used as a related-work example and baseline (Sec. 2, Table 2) and is not load-bearing for IMD's design. The claim that a 12% IMIM gain 'indicates our method efficiently mitigates the misalignment' is an interpretive inference from an author-proposed benchmark rather than a definitional identity: the paper defines misalignment as a discrepancy between single-image foundation-model features and cross-image matching needs, and IMIM is offered as a measure of one consequence of that discrepancy. The M/N metric is precision-only, lacks recall or match-count normalization, and is computed on only 100 pairs, which raises a validity/generalization concern about the 12% gap, but this is a benchmark-design risk, not a circular reduction. Because the central external results are self-contained and no prediction is equivalent to an input by construction, the appropriate finding is no significant circularity.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard deep-learning hyperparameters and empirical domain assumptions about diffusion features and the new benchmark. No new physical or mathematical entities are postulated.

free parameters (5)
  • Diffusion timestep t = 0
    Selected by hand after ablation (Table 4 row 6 shows t=100 degrades performance). Critical to the feature extraction, but it is a hyperparameter choice, not a fitted constant in a derivation.
  • UNet upsampling block index n = 2
    Chosen by hand to give 1/8 resolution features with 640 dimensions.
  • Coarse matching threshold tau = 0.2
    Hand-set threshold for mutual nearest neighbor selection.
  • Loss weights alpha and beta = 1.0 and 0.25
    Hand-set weights for the coarse and fine supervision terms.
  • Number of attention iterations N = 2
    Interleaved self- and cross-attention iterations on coarse features.
assumptions (4)
  • domain assumption Stable Diffusion 2-1 UNet features at t=0 encode instance-level details suitable for local feature matching.
    The paper relies on this empirical claim to motivate using a generative model over contrastive ones; it is supported by prior work (DIFT) but not proven here.
  • domain assumption Cross-attention between CLIP features of the two images provides effective cross-image interaction for matching.
    CIPM uses CLIP features and cross-attention to build prompts; the paper argues this captures correspondence better than concatenation, but this is an empirical design choice.
  • domain assumption Training on MegaDepth generalizes to ScanNet, HPatches, InLoc, and IMIM.
    The model is trained only on MegaDepth and evaluated on all other datasets; this cross-dataset generalization assumption is standard in matching literature.
  • domain assumption The IMIM mask-based evaluation protocol accurately measures multi-instance matching performance.
    The benchmark is newly proposed; its validity as a measure of misalignment is not independently established.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mind the Gap: Aligning Vision Foundation Models to Image Feature Matching." pith.science (2026). https://pith.science/paper/PUBT6F6R

@misc{pith2026250710318,
  author       = {Pith},
  title        = {Pith review of: Mind the Gap: Aligning Vision Foundation Models to Image Feature Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PUBT6F6R}},
  note         = {Machine review of arXiv:2507.10318}
}
read the original abstract

Leveraging the vision foundation models has emerged as a mainstream paradigm that improves the performance of image feature matching. However, previous works have ignored the misalignment when introducing the foundation models into feature matching. The misalignment arises from the discrepancy between the foundation models focusing on single-image understanding and the cross-image understanding requirement of feature matching. Specifically, 1) the embeddings derived from commonly used foundation models exhibit discrepancies with the optimal embeddings required for feature matching; 2) lacking an effective mechanism to leverage the single-image understanding ability into cross-image understanding. A significant consequence of the misalignment is they struggle when addressing multi-instance feature matching problems. To address this, we introduce a simple but effective framework, called IMD (Image feature Matching with a pre-trained Diffusion model) with two parts: 1) Unlike the dominant solutions employing contrastive-learning based foundation models that emphasize global semantics, we integrate the generative-based diffusion models to effectively capture instance-level details. 2) We leverage the prompt mechanism in generative model as a natural tunnel, propose a novel cross-image interaction prompting module to facilitate bidirectional information interaction between image pairs. To more accurately measure the misalignment, we propose a new benchmark called IMIM, which focuses on multi-instance scenarios. Our proposed IMD establishes a new state-of-the-art in commonly evaluated benchmarks, and the superior improvement 12% in IMIM indicates our method efficiently mitigates the misalignment.

Figures

Figures reproduced from arXiv: 2507.10318 by the authors.

Figure 1
Figure 1. The main idea of our proposed IMD. Previous works [15, 28] failed in multi-instance scenarios due to the misalignment. Our IMD leverages the explicit representation of specific instances within the diffusion model with an interaction module, accurately matching the specific instance. 1. Introduction Image feature matching serves as the foundation for many computer vision tasks, finds extensive applications in ar￾eas… view at source ↗
Figure 2
Figure 2. Pipeline Overview. Given an image pair, we first encode them with the image encoder and employ the cross-attention mechanism to proceed with explicit interaction between images. With the images and its prompt containing interactive information as input, we extract their coarse features CA, CB from a frozen text-to-image diffusion UNet. Then, we refine the coarse features for more discriminative by iteratively applyi… view at source ↗
Figure 3
Figure 3. Qualitative comparison of outdoor MegaDepth and IMIM datasets compared with DINOv2 [34], LoFTR [49], CasMTR [8], RoMa [15] and ours IMD. Green indicates the correctly matching results. More visualizations are provided in the Suppl [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Performance comparison and the circle size indicating parameter count. NOv2(G) (1,100M) of similar scale; (2) SD-tiny and SD￾small with CIPM, having 75% and 55% fewer parameters respectively. We further compare with RoMa [15] (dense) and SRMatcher [28] (semi-dense) usi…

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Probing and Leveraging Video Diffusion Transformer Features for Robust Point Tracking

    cs.CV 2025-12 conditional novelty 5.0 of 10

    Video diffusion transformer features, adapted with LoRA and fused with ResNet costs, produce a point tracker that matches or beats a real-data-trained CoTracker3 on hard benchmarks.

Reference graph

Works this paper leans on

63 extracted references · 51 canonical work pages · cited by 1 Pith paper

  1. [1]

    Burst: A benchmark for unifying object recognition, segmentation and tracking in video

    Ali Athar, Jonathon Luiten, Paul V oigtlaender, Tarasha Khu- rana, Achal Dave, Bastian Leibe, and Deva Ramanan. Burst: A benchmark for unifying object recognition, segmentation and tracking in video. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 1674–1683, 2023. 2, 6

    work page 2023

  2. [2]

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

    Vassileios Balntas, Karel Lenc, Andrea Vedaldi, and Krys- tian Mikolajczyk. Hpatches: A benchmark and evaluation of handcrafted and learned local descriptors. In Proceedings of the IEEE conference on computer vision and pattern recog- nition, pages 5173–5182, 2017. 7

    work page 2017

  3. [3]

    Axel Barroso-Laguna, Edgar Riba, Daniel Ponsa, and Krys- tian Mikolajczyk. Key. net: Keypoint detection by hand- crafted and learned cnn filters. In Proceedings of the IEEE/CVF international conference on computer vision , pages 5836–5844, 2019. 2

    work page 2019

  4. [4]

    Surf: Speeded up robust features

    Herbert Bay, Tinne Tuytelaars, and Luc Van Gool. Surf: Speeded up robust features. In Computer Vision–ECCV 2006: 9th European Conference on Computer Vision, Graz, Austria, May 7-13, 2006. Proceedings, Part I 9, pages 404–

    2006

  5. [5]

    Speeded-up robust features (surf)

    Herbert Bay, Andreas Ess, Tinne Tuytelaars, and Luc Van Gool. Speeded-up robust features (surf). Computer vi- sion and image understanding, 110(3):346–359, 2008. 2

    work page 2008

  6. [6]

    VOLoc: Visual Place Recognition by Querying Compressed Lidar Map

    Xudong Cai, Yongcai Wang, Zhe Huang, Yu Shao, and Dey- ing Li. V oloc: Visual place recognition by querying com- pressed lidar map. arXiv preprint arXiv:2402.15961, 2024. 1

    work page Pith review arXiv 2024

  7. [7]

    Prism: Pro- gressive dependency maximization for scale-invariant image matching

    Xudong Cai, Yongcai Wang, Lun Luo, Minhang Wang, Dey- ing Li, Jintao Xu, Weihao Gu, and Rui Ai. Prism: Pro- gressive dependency maximization for scale-invariant image matching. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 5250–5259, 2024. 6, 7

    work page 2024

  8. [8]

    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 Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 12129–12139, 2023. 1, 2, 5, 6, 7, 8

    work page 2023

Show all 63 references
  1. [9]

    Aspanformer: Detector-free image matching with adaptive span transformer

    Hongkai Chen, Zixin Luo, Lei Zhou, Yurun Tian, Ming- min Zhen, Tian Fang, David Mckinnon, Yanghai Tsin, and Long Quan. Aspanformer: Detector-free image matching with adaptive span transformer. In European Conference on Computer Vision, pages 20–36. Springer, 2022. 2, 5, 7

    2022

  2. [10]

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

    Peiqi Chen, Lei Yu, Yi Wan, Yongjun Zhang, Jian Wang, Li- heng Zhong, Jingdong Chen, and Ming Yang. Ecomatcher: Efficient clustering oriented matcher for detector-free image matching. In European Conference on Computer Vision , pages 344–360. Springer, 2024. 6

    2024

  3. [11]

    Scannet: Richly-annotated 3d reconstructions of indoor scenes

    Angela Dai, Angel X Chang, Manolis Savva, Maciej Hal- ber, Thomas Funkhouser, and Matthias Nießner. Scannet: Richly-annotated 3d reconstructions of indoor scenes. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5828–5839, 2017. 1, 6, 7

    2017

  4. [12]

    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 Proceedings of the IEEE conference on computer vision and pattern recognition workshops , pages 224–236, 2018. 1, 2, 6, 7

    2018

  5. [13]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. Advances in neural informa- tion processing systems, 34:8780–8794, 2021. 4

    2021

  6. [14]

    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 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17765–17775, 2023. 1, 2

    2023

  7. [15]

    Roma: Robust dense fea- ture matching

    Johan Edstedt, Qiyu Sun, Georg B ¨okman, M ˚arten Wadenb¨ack, and Michael Felsberg. Roma: Robust dense fea- ture matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19790– 19800, 2024. 1, 2, 3, 8

    2024

  8. [16]

    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 Proceedings of the AAAI conference on artificial in- telligence, pages 2447–2455, 2023. 6

    2023

  9. [17]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 6

    2016

  10. [18]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Doll´ar, and Ross Girshick. Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 16000– 16009, 2022. 8

    2022

  11. [19]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 4

    2020

  12. [20]

    Dynamicid: Zero-shot multi-id image personalization with flexible facial editabil- ity

    Xirui Hu, Jiahao Wang, Hao Chen, Weizhan Zhang, Benqi Wang, Yikun Li, and Haishun Nan. Dynamicid: Zero-shot multi-id image personalization with flexible facial editabil- ity. 2025. 4

    2025

  13. [21]

    Omniglue: Generalizable feature match- ing with foundation model guidance

    Hanwen Jiang, Arjun Karpur, Bingyi Cao, Qixing Huang, and Andr´e Araujo. Omniglue: Generalizable feature match- ing with foundation model guidance. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 19865–19875, 2024. 1, 2, 3, 6, 7

    2024

  14. [22]

    Segment any- thing

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C Berg, Wan-Yen Lo, et al. Segment any- thing. In Proceedings of the IEEE/CVF international confer- ence on computer vision, pages 4015–4026, 2023. 2

    2023

  15. [23]

    Sd4match: Learning to prompt stable diffu- sion model for semantic matching

    Xinghui Li, Jingyi Lu, Kai Han, and Victor Adrian Prisacariu. Sd4match: Learning to prompt stable diffu- sion model for semantic matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 27558–27568, 2024. 3, 4, 5 9

    2024

  16. [24]

    Megadepth: Learning single- view depth prediction from internet photos

    Zhengqi Li and Noah Snavely. Megadepth: Learning single- view depth prediction from internet photos. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 2041–2050, 2018. 1, 6, 7

    2018

  17. [25]

    Recrecnet: Rectangling rectified wide-angle images by thin-plate spline model and dof-based curriculum learn- ing

    Kang Liao, Lang Nie, Chunyu Lin, Zishuo Zheng, and Yao Zhao. Recrecnet: Rectangling rectified wide-angle images by thin-plate spline model and dof-based curriculum learn- ing. In Proceedings of the IEEE/CVF International Confer- ence on Computer Vision, pages 10800–10809, 2023. 1

    2023

  18. [26]

    Feature pyra- mid networks for object detection

    Tsung-Yi Lin, Piotr Doll ´ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyra- mid networks for object detection. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 2117–2125, 2017. 5

    2017

  19. [27]

    Lightglue: Local feature matching at light speed

    Philipp Lindenberger, Paul-Edouard Sarlin, and Marc Polle- feys. Lightglue: Local feature matching at light speed. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 17627–17638, 2023. 1, 6

    2023

  20. [28]

    Semantic- aware representation learning for homography estimation

    Yuhan Liu, Qianxin Huang, Siqi Hui, Jingwen Fu, Sanping Zhou, Kangyi Wu, Pengna Li, and Jinjun Wang. Semantic- aware representation learning for homography estimation. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 2506–2514, 2024. 1, 2, 7, 8

    2024

  21. [29]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision, pages 10012–10022, 2021. 8

    2021

  22. [30]

    Distinctive image features from scale- invariant keypoints

    David G Lowe. Distinctive image features from scale- invariant keypoints. International journal of computer vi- sion, 60:91–110, 2004. 2

    2004

  23. [31]

    Raising the ceiling: Conflict- free local feature matching with dynamic view switching

    Xiaoyong Lu and Songlin Du. Raising the ceiling: Conflict- free local feature matching with dynamic view switching. In European Conference on Computer Vision, pages 256–273. Springer, 2024. 6

    2024

  24. [32]

    Diffusion model for dense matching.arXiv preprint arXiv:2305.19094,

    Jisu Nam, Gyuseong Lee, Sunwoo Kim, Hyeonsu Kim, Hy- oungwon Cho, Seyeon Kim, and Seungryong Kim. Diffusion model for dense matching.arXiv preprint arXiv:2305.19094,

    arXiv

  25. [33]

    Unsupervised deep image stitching: Reconstructing stitched features to images

    Lang Nie, Chunyu Lin, Kang Liao, Shuaicheng Liu, and Yao Zhao. Unsupervised deep image stitching: Reconstructing stitched features to images. IEEE Transactions on Image Processing, 30:6184–6197, 2021. 1

    2021

  26. [34]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 3, 6, 8

    2023 arXiv

  27. [35]

    Enhancing deformable local features by jointly learning to detect and describe keypoints

    Guilherme Potje, Felipe Cadar, Andr ´e Araujo, Renato Mar- tins, and Erickson R Nascimento. Enhancing deformable local features by jointly learning to detect and describe keypoints. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 13...

    2023

  28. [36]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  29. [37]

    R2d2: Reliable and repeatable detec- tor and descriptor

    Jerome Revaud, Cesar De Souza, Martin Humenberger, and Philippe Weinzaepfel. R2d2: Reliable and repeatable detec- tor and descriptor. Advances in neural information process- ing systems, 32, 2019. 1, 7

    2019

  30. [38]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 4, 5, 8

    2022

  31. [39]

    Orb: An efficient alternative to sift or surf

    Ethan Rublee, Vincent Rabaud, Kurt Konolige, and Gary Bradski. Orb: An efficient alternative to sift or surf. In 2011 International conference on computer vision , pages 2564–

    2011

  32. [40]

    Where’s waldo: Diffusion features for person- alized segmentation and retrieval

    Dvir Samuel, Rami Ben-Ari, Matan Levy, Nir Darshan, and Gal Chechik. Where’s waldo: Diffusion features for person- alized segmentation and retrieval. In The Thirty-eighth An- nual Conference on Neural Information Processing Systems,

  33. [41]

    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 Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 12716–12725, 2019. 1, 7

    2019

  34. [42]

    Superglue: Learning feature matching with graph neural networks

    Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Superglue: Learning feature matching with graph neural networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4938–4947, 2020. 2, 7

    2020

  35. [43]

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

    Paul-Edouard Sarlin, Ajaykumar Unagar, Mans Larsson, Hugo Germain, Carl Toft, Viktor Larsson, Marc Pollefeys, Vincent Lepetit, Lars Hammarstrand, Fredrik Kahl, et al. Back to the feature: Learning robust camera localization from pixels to pose. In Proceedings of the IEEE/CVF c...

    2021

  36. [44]

    Benchmarking 6dof outdoor visual localization in changing conditions

    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 Proceedings of the IEEE conference on compu...

    2018

  37. [45]

    Structure- from-motion revisited

    Johannes L Schonberger and Jan-Michael Frahm. Structure- from-motion revisited. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 4104–4113, 2016. 1, 7

    2016

  38. [46]

    Pixelwise view selection for unstructured multi-view stereo

    Johannes L Sch ¨onberger, Enliang Zheng, Jan-Michael Frahm, and Marc Pollefeys. Pixelwise view selection for unstructured multi-view stereo. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Nether- lands, October 11-14, 2016, Proceedings, Part III 14, pag...

    2016

  39. [47]

    Laion-5b: An open large-scale dataset for training 10 next generation image-text models

    Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Worts- man, et al. Laion-5b: An open large-scale dataset for training 10 next generation image-text models. Advances in Neural In...

    2022

  40. [48]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020. 4

    2010 arXiv

  41. [49]

    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 Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 8922–8931, 2021. 1, 2, 5, 6, 7, 8

    2021

  42. [50]

    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 Proceedings of the IEEE Conference on Computer Vision and Pattern Recog...

    2018

  43. [51]

    Emergent correspondence from image diffusion

    Luming Tang, Menglin Jia, Qianqian Wang, Cheng Perng Phoo, and Bharath Hariharan. Emergent correspondence from image diffusion. Advances in Neural Information Pro- cessing Systems, 36:1363–1389, 2023. 2, 3, 4, 5, 6

    2023

  44. [52]

    Quadtree attention for vision transformers

    Shitao Tang, Jiahui Zhang, Siyu Zhu, and Ping Tan. Quadtree attention for vision transformers. arXiv preprint arXiv:2201.02767, 2022. 1, 2

    2022 arXiv

  45. [53]

    Homomatcher: Dense feature matching results with semi-dense efficiency by homography estimation.arXiv preprint arXiv:2411.06700, 2024

    Xiaolong Wang, Lei Yu, Yingying Zhang, Jiangwei Lao, Lixiang Ru, Liheng Zhong, Jingdong Chen, Yu Zhang, and Ming Yang. Homomatcher: Dense feature matching results with semi-dense efficiency by homography estimation.arXiv preprint arXiv:2411.06700, 2024. 6, 7

    2024 arXiv

  46. [54]

    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 Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 21666–21675, 2024. 5, 6, 7

    2024

  47. [55]

    Croco: Self-supervised pre-training for 3d vision tasks by cross-view completion

    Philippe Weinzaepfel, Vincent Leroy, Thomas Lucas, Ro- main Br´egier, Yohann Cabon, Vaibhav Arora, Leonid Ants- feld, Boris Chidlovskii, Gabriela Csurka, and J ´erˆome Re- vaud. Croco: Self-supervised pre-training for 3d vision tasks by cross-view completion. Advances in Neura...

    2022

  48. [56]

    Croco v2: Improved cross-view completion pre- training for stereo matching and optical flow

    Philippe Weinzaepfel, Thomas Lucas, Vincent Leroy, Yohann Cabon, Vaibhav Arora, Romain Br ´egier, Gabriela Csurka, Leonid Antsfeld, Boris Chidlovskii, and J ´erˆome Revaud. Croco v2: Improved cross-view completion pre- training for stereo matching and optical flow. InProceedin...

    2023

  49. [57]

    Open-vocabulary panop- tic segmentation with text-to-image diffusion models

    Jiarui Xu, Sifei Liu, Arash Vahdat, Wonmin Byeon, Xiao- long Wang, and Shalini De Mello. Open-vocabulary panop- tic segmentation with text-to-image diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition , pages 2955–2966, 2023. 3, 4

    2023

  50. [58]

    Telling left from right: Identifying geometry-aware semantic corre- spondence

    Junyi Zhang, Charles Herrmann, Junhwa Hur, Eric Chen, Varun Jampani, Deqing Sun, and Ming-Hsuan Yang. Telling left from right: Identifying geometry-aware semantic corre- spondence. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 307...

    2024

  51. [59]

    A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence

    Junyi Zhang, Charles Herrmann, Junhwa Hur, Luisa Pola- nia Cabrera, Varun Jampani, Deqing Sun, and Ming-Hsuan Yang. A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence. Advances in Neural Information Processing Systems, 36, 2024. 2, 3, 4

    2024

  52. [60]

    Diffglue: Diffusion-aided image feature matching

    Shihua Zhang and Jiayi Ma. Diffglue: Diffusion-aided image feature matching. In Proceedings of the 32nd ACM Interna- tional Conference on Multimedia , pages 8451–8460, 2024. 2

    2024

  53. [61]

    Mesa: Matching everything by segmenting anything

    Yesheng Zhang and Xu Zhao. Mesa: Matching everything by segmenting anything. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 20217–20226, 2024. 2, 6

    2024

  54. [62]

    Unleashing text-to-image diffu- sion models for visual perception

    Wenliang Zhao, Yongming Rao, Zuyan Liu, Benlin Liu, Jie Zhou, and Jiwen Lu. Unleashing text-to-image diffu- sion models for visual perception. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 5729–5739, 2023. 3

    2023

  55. [63]

    Pmatch: Paired masked image modeling for dense geometric matching

    Shengjie Zhu and Xiaoming Liu. Pmatch: Paired masked image modeling for dense geometric matching. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21909–21918, 2023. 2 11

    2023

Pith tools

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