Pith. sign in

REVIEW 2 major objections 5 minor 1 cited by

BridgeDepth: Bridging Monocular and Stereo Reasoning with Latent Alignment

T0 review · 2 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper claims that iteratively aligning monocular and stereo latent representations reduces zero-shot disparity error by more than 40 percent.

desk verdict Solid, novel fusion mechanism with strong zero-shot numbers; the top-2 DPN pruning ceiling is unmeasured but addressable, not fatal. read the letter →

arxiv 2508.04611 v2 pith:UCJ4GQAA submitted 2025-08-06 cs.CV cs.RO

Tongfan Guan , Jiaxin Guo , Chen Wang , Yun-Hui Liu This is my paper
classification cs.CVcs.RO
keywords stereodepthestimationmonocularlatentalignmentcross-attentionzero-shotgeneralizationdisparityproposalnetworktransparentandreflectivesurfacesfoundationmodels
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 proposes a single network that does stereo matching and monocular depth estimation together, instead of treating them as separate tasks. Its claim is that monocular context and stereo geometry can be fused at the representation level, so that each stream is refined by the other during cost aggregation. Trained only on synthetic Scene Flow data, the model reports zero-shot bad-pixel error reductions of 42.7% on Middlebury and 65.8% on ETH3D relative to the NMRF stereo baseline, while also keeping coherent depth on reflective and transparent surfaces. The dual outputs—metric disparity from the stereo branch and relative depth from the monocular branch—are produced by one aligned latent space, which is what the paper argues makes the improvement structural rather than a post-hoc fusion.

What carries the argument

The load-bearing mechanism is the bidirectional latent-alignment module: window-based cross-attention with a content-adaptive positional bias, interleaved with the Neural Message Passing (NMP) cost aggregation. The Disparity Proposal Network (DPN) prunes the disparity search space from 40 to 2 hypotheses per pixel, making the cross-modal attention tractable; the alignment module exchanges information between the monocular contextual features and those hypotheses before and after aggregation. A coarse-to-fine cascade at quarter resolution, with halved attention windows, recovers fine detail. This module is what lets monocular structure priors disambiguate stereo matching and lets stereo geome

What would settle it

Compute the oracle recall of the DPN's top-2 hypotheses on Middlebury and ETH3D high-resolution images: if ground-truth disparities fall outside the retained pair for a substantial fraction of error pixels, the alignment cannot be the source of those failures. Equivalently, evaluating at 2K+ resolution should show an error spike, as the paper's own conclusion concedes.

Watch

Extended reading notes

Core claim

The central claim is that monocular and stereo reasoning can be synchronized by iteratively aligning their latent representations inside the stereo cost-aggregation loop. A cross-attentive transformer performs two complementary operations per iteration: a monocular-readout phase in which disparity hypothesis embeddings attend to frozen monocular contextual features before aggregation, and a monocular-update phase in which the monocular features attend to the aggregated stereo embeddings. The paper reports that this bidirectional exchange, with a Disparity Proposal Network pruning the 40 candidates per pixel to 2, lowers zero-shot error by 42.7% on Middlebury and 65.8% on ETH3D versus NMRF an

Load-bearing premise

The argument stands on the DPN's two retained disparity hypotheses: if the true disparity has been pruned before the alignment runs, the cross-attention sees only wrong candidates and the monocular injection cannot recover the correct match.

Editorial extensions

If this is right

  • If the claim holds, stereo models can inherit zero-shot generalization from frozen monocular foundation models without fine-tuning the backbone.
  • Reflective and transparent surfaces—long-standing failure cases for correspondence matching—become tractable because monocular structure priors are injected into the matching hypotheses.
  • A single network can output both metric disparity and relative depth that are mutually consistent, so downstream 3D perception gets two cross-validated depth products.
  • The >40% zero-shot error reductions on Middlebury and ETH3D imply that the approach can be applied to unseen indoor and outdoor scenes without per-domain adaptation.
  • Because the DPN restricts alignment to two hypotheses per pixel, the method stays fast enough (0.13 s per frame on KITTI-resolution input) for near-real-time use.

Reading between the lines

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

  • Editorial inference: the alignment pattern is not tied to stereo; the paper hints at optical flow, and the same query-update loop could couple motion hypotheses with monocular structure in a flow network.
  • Editorial inference: the strongest determinant of the ceiling is the DPN's two-hypothesis pruning, not the alignment module; improving proposal recall at high resolutions may matter more than stronger cross-attention.
  • Editorial inference: the monocular-update phase appears to buy little for stereo accuracy in the ablations, so its main payoff may be the improved relative-depth output; a version that omits the update phase might be nearly as strong on disparity.
  • Editorial inference: replacing the monocular backbone changes zero-shot results in the supplementary, so gains are likely to grow as monocular foundation models improve.
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

2 major / 5 minor

Summary. The paper proposes BridgeDepth, a stereo matching network that couples a frozen monocular DepthAnythingV2 (DAv2) branch with the NMRF stereo pipeline. A window-based cross-attention module iteratively aligns monocular contextual features with stereo hypothesis embeddings before and after neural message passing cost aggregation, followed by a coarse-to-fine cascade for fine detail. The network is trained on Scene Flow with dual stereo and monocular losses, and outputs both metric disparity and relative depth. The central empirical claim is that zero-shot generalization error drops by 42.7% on Middlebury (BP-2: 7.5→4.3) and 65.8% on ETH3D (BP-1: 3.8→1.3) relative to NMRF, with additional gains on KITTI and Scene Flow. The paper also reports improved robustness on reflective/transparent surfaces and faster inference than several recent hybrid methods.

Significance. The claimed improvements are externally measured benchmark differences, not internal derivatives, and the tables are internally consistent: Table 3 reproduces exactly the 42.7% and 65.8% reductions, and Table 4 shows cumulative benefits from each added component. The use of a frozen DAv2 branch and a cross-attention module embedded in cost aggregation is a genuine architectural contribution rather than post-hoc fusion; the supplementary comparison against post-hoc fusion (Table 7) supports this. Code is promised, and the dual outputs (metric disparity plus refined relative depth) give a falsifiable set of predictions. The main risk is not circularity—the monocular branch is externally pretrained and frozen—but whether the DPN top-2 pruning ceiling undermines the headline zero-shot generalization claim. Because the paper explicitly concedes high-resolution failures and never measures DPN oracle recall, the central claim needs additional evidence before full acceptance.

major comments (2)
  1. [§3.2 and Conclusion] The central zero-shot claim is bounded by the DPN's top-2 recall, which is never reported. In §3.2, the stereo branch prunes the disparity space from D=40 to K=2 before any monocular interaction. Equations (2)–(7) show that the cross-attention module reweights existing stereo hypotheses via the monocular-readout and updates monocular features via the monocular-update; neither operation proposes a new disparity candidate into the stereo hypothesis set, and the NMP aggregation in §3.3 operates on the same K=2 embeddings. The Conclusion concedes that 'large disparity ranges strain the top-k hypothesis selection' at ≥2K resolution. Because the headline result is the 42.7% Middlebury / 65.8% ETH3D zero-shot reduction, and those benchmarks include high-resolution pairs, the claim is conditional on the DPN retaining the correct disparity on the exact evaluation inputs. Please report per-dataset
  2. [§4.1 and Table 4] The architecture's defining hyperparameters are unspecified. The method depends on window size N, the number of alignment steps l and l', and the iteration weights w_i in Eqs. (8)–(9); Section 4.1 reports only the optimizer, crop size, batch size, and learning rate. Table 4's efficiency claims (e.g., 0.147 s) cannot be attributed to a concrete configuration, and the ablation cannot be reproduced. Please add a complete configuration list (N, l, l', w_i, and the coarse-to-fine candidate count) to the implementation details.
minor comments (5)
  1. [§1, Abstract, Table 3] The abstract and Table 3 give 42.7% for Middlebury, while §1 states 41.3%. Please harmonize the numbers.
  2. [Supplementary Sec. A and Table 7] The method variant is called 'BridgeDepth-L' in the main paper but 'OmiDepth-L' in the supplementary text; Table 7 also has a missing reference ('summarized in ,').
  3. [Supplementary Table 8] The text claims SOTA on ETH3D, but FoundationStereo has a lower BP-1 (0.48 vs. 0.50). Please qualify the claim by metric (e.g., best on BP-2/AvgErr/RMS).
  4. [Table 1 and §4.4] The text says the method 'outperforms' prior hybrid/geometry-only methods, but IGEV++ and MonSter have lower error on several KITTI 2012 and KITTI 2012 Reflective Noc metrics (e.g., 5.49 vs. 5.80 on Reflective Out-2 Noc). Please state the specific metrics for which the claim holds.
  5. [Table 9 and §3.4] Clarify the evaluation protocol for the monocular outputs: whether stereo pairs are used at inference, and how the affine alignment is computed. Without this, the comparison against DAv2-L and UDv2 is not on equal terms.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the headline result is an externally measured benchmark improvement, not a derivation that reduces to its inputs.

full rationale

The paper's central claim is a zero-shot generalization improvement (+42.7% on Middlebury, +65.8% on ETH3D) measured on standard external benchmarks against the NMRF baseline. This is an empirical, falsifiable comparison, not a quantity defined in terms of the fitted parameters or the method's own outputs. The monocular branch uses a frozen external DAv2 model, and the proposed cross-attentive alignment module is a genuinely additional component; the stereo branch reuses NMRF's DPN and NMP, but this is architectural inheritance from a published CVPR paper by the same first author, not an unverified self-citation that forces the result. The paper explicitly concedes a limitation at high resolutions (>=2K) where the DPN's top-k pruning may discard the true disparity; this is a correctness/precondition risk about oracle recall, not circularity. The loss functions (Eqs. 8 and 9) supervise the stereo and monocular outputs against ground truth, so neither output is defined in terms of the other. No equation or claim reduces to its own input by construction, and no renamed known result is presented as a new derivation. The self-citation is present but does not make the central derivation circular.

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

The method introduces no physical entities. Its free parameters are standard architectural hyperparameters; the main load-bearing choices are K=2 hypothesis pruning and the unspecified alignment iterations and window size. All heavy pretrained components (DAv2, RAFT features, DPN/NMP from NMRF) are carried from prior literature.

free parameters (4)
  • Disparity hypothesis count K = 2
    Chosen by hand to trade fidelity for tractability of cross-attention; DPN prunes from D=40 to K=2 (Section 3.2). If K is too small, the correct match can be excluded.
  • Monocular loss exponential weight gamma = 0.8
    Set in Eq. (9), Section 3.5; controls weighting of intermediate monocular predictions.
  • Stereo iteration loss weights w_i = unspecified
    Balances contributions across all stereo iterations in Eq. (8); exact values are not reported.
  • Window size N and alignment step counts l, l' = unspecified
    Cross-attention window size and number of bidirectional alignment iterations are not given, yet they control accuracy and runtime.
assumptions (4)
  • domain assumption Frozen DAv2 encoder transfers monocular contextual priors
    The monocular branch is a frozen DAv2-ViT-L model, so all monocular gains depend on its pretrained generalization (Section 3.1).
  • domain assumption DPN top-2 hypothesis pruning retains the correct disparity
    The cross-attention only sees K=2 candidates after pruning; the authors acknowledge high-resolution failure in the Conclusion.
  • domain assumption Scene Flow-only training transfers to real benchmarks
    Zero-shot evaluation trains only on synthetic Scene Flow and expects transfer to KITTI, Middlebury, and ETH3D (Section 4.1).
  • domain assumption Affine-invariant loss is sufficient monocular supervision
    The monocular head is supervised with the affine-invariant MiDaS loss rather than metric depth (Section 3.5).

how reviews work

0 comments
Cite this review

Pith. "Pith review of BridgeDepth: Bridging Monocular and Stereo Reasoning with Latent Alignment." pith.science (2026). https://pith.science/paper/UCJ4GQAA

@misc{pith2026250804611,
  author       = {Pith},
  title        = {Pith review of: BridgeDepth: Bridging Monocular and Stereo Reasoning with Latent Alignment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UCJ4GQAA}},
  note         = {Machine review of arXiv:2508.04611}
}
abstract

Monocular and stereo depth estimation offer complementary strengths: monocular methods capture rich contextual priors but lack geometric precision, while stereo approaches leverage epipolar geometry yet struggle with ambiguities such as reflective or textureless surfaces. Despite post-hoc synergies, these paradigms remain largely disjoint in practice. We introduce a unified framework that bridges both through iterative bidirectional alignment of their latent representations. At its core, a novel cross-attentive alignment mechanism dynamically synchronizes monocular contextual cues with stereo hypothesis representations during stereo reasoning. This mutual alignment resolves stereo ambiguities (e.g., specular surfaces) by injecting monocular structure priors while refining monocular depth with stereo geometry within a single network. Extensive experiments demonstrate state-of-the-art results: \textbf{it reduces zero-shot generalization error by $\!>\!40\%$ on Middlebury and ETH3D}, while addressing longstanding failures on transparent and reflective surfaces. By harmonizing multi-view geometry with monocular context, our approach enables robust 3D perception that transcends modality-specific limitations. Codes available at https://github.com/aeolusguan/BridgeDepth.

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.

  1. STEREOFLOW: Progressive Stereo Matching with StereoDiT and Transition Flow Matching

    cs.CV 2026-07 conditional novelty 5.0 of 10

    A hybrid stereo-matching model uses a cascade matching network to propose disparities and a diffusion transformer to refine ambiguous regions; it claims state-of-the-art benchmark results.

Reference graph

Works this paper leans on

61 extracted references · 44 canonical work pages · cited by 1 Pith paper

  1. [1]

    Multi- view depth estimation by fusing single-view depth prob- ability with multi-view geometry

    Gwangbin Bae, Ignas Budvytis, and Roberto Cipolla. Multi- view depth estimation by fusing single-view depth prob- ability with multi-view geometry. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2842–2851, 2022. 2, 3

    work page 2022

  2. [2]

    Stereo anywhere: Robust zero-shot deep stereo matching even where either stereo or mono fail

    Luca Bartolomei, Fabio Tosi, Matteo Poggi, and Stefano Mattoccia. Stereo anywhere: Robust zero-shot deep stereo matching even where either stereo or mono fail. arXiv preprint arXiv:2412.04472, 2024. 3

    arXiv 2024

  3. [3]

    Depth pro: Sharp monocular metric depth in less than a second

    Aleksei Bochkovskii, Ama ˜AG ¸ l Delaunoy, Hugo Germain, Marcel Santos, Yichao Zhou, Stephan R Richter, and Vladlen Koltun. Depth pro: Sharp monocular metric depth in less than a second. arXiv preprint arXiv:2410.02073, 2024. 2

    arXiv 2024

  4. [4]

    Pyramid stereo matching network

    Jia-Ren Chang and Yong-Sheng Chen. Pyramid stereo matching network. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 5410–5418,

  5. [5]

    Motif Channel Opened in a White-Box: Stereo Matching via Motif Correlation Graph

    Ziyang Chen, Yongjun Zhang, Wenting Li, Bingshu Wang, Yong Zhao, and CL Chen. Motif channel opened in a white- box: Stereo matching via motif correlation graph. arXiv preprint arXiv:2411.12426, 2024. 2, 6, 7, 8

    work page Pith review arXiv 2024

  6. [6]

    Monster: Marry monodepth to stereo unleashes power

    Junda Cheng, Longliang Liu, Gangwei Xu, Xianqi Wang, Zhaoxing Zhang, Yong Deng, Jinliang Zang, Yurui Chen, Zhipeng Cai, and Xin Yang. Monster: Marry monodepth to stereo unleashes power. arXiv preprint arXiv:2501.08643 ,

    arXiv

  7. [7]

    Learning depth with convolutional spatial propagation network

    Xinjing Cheng, Peng Wang, and Ruigang Yang. Learning depth with convolutional spatial propagation network. IEEE transactions on pattern analysis and machine intelligence ,

    work page

  8. [8]

    Hierarchical neural architecture search for deep stereo matching

    Xuelian Cheng, Yiran Zhong, Mehrtash Harandi, Yuchao Dai, Xiaojun Chang, Hongdong Li, Tom Drummond, and Zongyuan Ge. Hierarchical neural architecture search for deep stereo matching. Advances in Neural Information Pro- cessing Systems, 33, 2020. 6

    work page 2020

Show all 61 references
  1. [9]

    An image is worth 16x16 words: Trans- formers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, et al. An image is worth 16x16 words: Trans- formers for image recognition at scale. arXiv preprint a...

    2010 arXiv

  2. [10]

    Depth map prediction from a single image using a multi-scale deep net- work

    David Eigen, Christian Puhrsch, and Rob Fergus. Depth map prediction from a single image using a multi-scale deep net- work. In Advances in neural information processing systems, pages 2366–2374, 2014. 2

    2014

  3. [11]

    Geowiz- ard: Unleashing the diffusion priors for 3d geometry esti- mation from a single image

    Xiao Fu, Wei Yin, Mu Hu, Kaixuan Wang, Yuexin Ma, Ping Tan, Shaojie Shen, Dahua Lin, and Xiaoxiao Long. Geowiz- ard: Unleashing the diffusion priors for 3d geometry esti- mation from a single image. In European Conference on Computer Vision, pages 241–258. Springer, 2024. 2

    2024

  4. [12]

    Are we ready for autonomous driving? the kitti vision benchmark suite

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the kitti vision benchmark suite. In 2012 IEEE conference on computer vision and pat- tern recognition, pages 3354–3361. IEEE, 2012. 2, 7, 8

    2012

  5. [13]

    Neural markov random field for stereo matching

    Tongfan Guan, Chen Wang, and Yun-Hui Liu. Neural markov random field for stereo matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 5459–5469, 2024. 1, 2, 3, 4, 5, 6, 7, 8

    2024

  6. [14]

    Context-enhanced stereo transformer

    Weiyu Guo, Zhaoshuo Li, Yongkui Yang, Zheng Wang, Rus- sell H Taylor, Mathias Unberath, Alan Yuille, and Yingwei Li. Context-enhanced stereo transformer. In European Con- ference on Computer Vision, pages 263–279. Springer, 2022. 3

    2022

  7. [15]

    Group-wise correlation stereo network

    Xiaoyang Guo, Kai Yang, Wukui Yang, Xiaogang Wang, and Hongsheng Li. Group-wise correlation stereo network. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3273–3282, 2019. 2, 3

    2019

  8. [16]

    Lotus: Diffusion-based visual foundation model for high-quality dense prediction

    Jing He, Haodong Li, Wei Yin, Yixun Liang, Leheng Li, Kaiqiang Zhou, Hongbo Liu, Bingbing Liu, and Ying- Cong Chen. Lotus: Diffusion-based visual foundation model for high-quality dense prediction. arXiv preprint arXiv:2409.18124, 2024. 2

    2024 arXiv

  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. 1

    2016

  10. [18]

    Gaussian error linear units (gelus)

    Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (gelus). arXiv preprint arXiv:1606.08415, 2016. 4

    2016 arXiv

  11. [19]

    Stereo processing by semiglobal match- ing and mutual information

    Heiko Hirschmuller. Stereo processing by semiglobal match- ing and mutual information. IEEE Transactions on pattern analysis and machine intelligence, 30(2):328–341, 2007. 2

    2007

  12. [20]

    Learning accurate 3d shape based on stereo polarimetric imaging

    Tianyu Huang, Haoang Li, Kejing He, Congying Sui, Bin Li, and Yun-Hui Liu. Learning accurate 3d shape based on stereo polarimetric imaging. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 17287–17296, 2023. 2

    2023

  13. [21]

    Defom-stereo: Depth foundation model based stereo matching

    Hualie Jiang, Zhiqiang Lou, Laiyan Ding, Rui Xu, Minglang Tan, Wenjie Jiang, and Rui Huang. Defom-stereo: Depth foundation model based stereo matching. arXiv preprint arXiv:2501.09466, 2025. 3, 6, 8, 13

    2025 arXiv

  14. [22]

    Uncertainty guided adaptive warping for robust and efficient stereo matching

    Junpeng Jing, Jiankun Li, Pengfei Xiong, Jiangyu Liu, Shuaicheng Liu, Yichen Guo, Xin Deng, Mai Xu, Lai Jiang, and Leonid Sigal. Uncertainty guided adaptive warping for robust and efficient stereo matching. In Proceedings of the IEEE/CVF International Conference on Computer Vi...

    2023

  15. [23]

    Repurpos- ing diffusion-based image generators for monocular depth estimation

    Bingxin Ke, Anton Obukhov, Shengyu Huang, Nando Met- zger, Rodrigo Caye Daudt, and Konrad Schindler. Repurpos- ing diffusion-based image generators for monocular depth estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. 1, 2

    2024

  16. [24]

    End-to-end learning of geometry and context for deep stereo regression

    Alex Kendall, Hayk Martirosyan, Saumitro Dasgupta, Peter Henry, Ryan Kennedy, Abraham Bachrach, and Adam Bry. End-to-end learning of geometry and context for deep stereo regression. In Proceedings of the IEEE international confer- ence on computer vision, pages 66–75, 2017. 1, 2, 3

    2017

  17. [25]

    From big to small: Multi-scale local planar guidance for monocular depth estimation

    Jin Han Lee, Myung-Kyu Han, Dong Wook Ko, and Il Hong Suh. From big to small: Multi-scale local planar guidance for monocular depth estimation. arXiv preprint arXiv:1907.10326, 2019. 2 9

    1907 arXiv

  18. [26]

    Practical stereo matching via cascaded recurrent net- work with adaptive correlation

    Jiankun Li, Peisen Wang, Pengfei Xiong, Tao Cai, Ziwei Yan, Lei Yang, Jiangyu Liu, Haoqiang Fan, and Shuaicheng Liu. Practical stereo matching via cascaded recurrent net- work with adaptive correlation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern R...

    2022

  19. [27]

    Los: Local structure-guided stereo matching

    Kunhong Li, Longguang Wang, Ye Zhang, Kaiwen Xue, Shunbo Zhou, and Yulan Guo. Los: Local structure-guided stereo matching. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition , pages 19746–19756, 2024. 2, 3, 6, 8, 13

    2024

  20. [28]

    Revisiting stereo depth estimation from a sequence- to-sequence perspective with transformers

    Zhaoshuo Li, Xingtong Liu, Nathan Drenkow, Andy Ding, Francis X Creighton, Russell H Taylor, and Mathias Un- berath. Revisiting stereo depth estimation from a sequence- to-sequence perspective with transformers. In Proceedings of the IEEE/CVF international conference on comput...

    2021

  21. [29]

    Raft-stereo: Multilevel recurrent field transforms for stereo matching

    Lahav Lipson, Zachary Teed, and Jia Deng. Raft-stereo: Multilevel recurrent field transforms for stereo matching. In 2021 International Conference on 3D Vision (3DV) , pages 218–227. IEEE, 2021. 1, 2, 6, 7

    2021

  22. [30]

    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. 5

    2021

  23. [31]

    A convnet for the 2020s

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feicht- enhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , pages 11976–11986,

  24. [32]

    A large dataset to train convolutional networks for disparity, optical flow, and scene flow estimation

    Nikolaus Mayer, Eddy Ilg, Philip Hausser, Philipp Fischer, Daniel Cremers, Alexey Dosovitskiy, and Thomas Brox. A large dataset to train convolutional networks for disparity, optical flow, and scene flow estimation. In Proceedings of the IEEE conference on computer vision and ...

    2016

  25. [33]

    Object scene flow for au- tonomous vehicles

    Moritz Menze and Andreas Geiger. Object scene flow for au- tonomous vehicles. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 3061– 3070, 2015. 2, 7, 8

    2015

  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. 2, 3

    2023 arXiv

  27. [35]

    Unidepth: Universal monocular metric depth estimation

    Luigi Piccinelli, Yung-Hsu Yang, Christos Sakaridis, Mattia Segu, Siyuan Li, Luc Van Gool, and Fisher Yu. Unidepth: Universal monocular metric depth estimation. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10106–10116, 2024. 1, 2

    2024

  28. [36]

    UniDepthV2: Universal monocular metric depth estimation made simpler, 2025

    Luigi Piccinelli, Christos Sakaridis, Yung-Hsu Yang, Mat- tia Segu, Siyuan Li, Wim Abbeloos, and Luc Van Gool. UniDepthV2: Universal monocular metric depth estimation made simpler, 2025. 12

    2025

  29. [37]

    Towards robust monocular depth estimation: Mixing datasets for zero-shot cross-dataset transfer

    Ren ´e Ranftl, Katrin Lasinger, David Hafner, Konrad Schindler, and Vladlen Koltun. Towards robust monocular depth estimation: Mixing datasets for zero-shot cross-dataset transfer. IEEE transactions on pattern analysis and machine intelligence, 44(3):1623–1637, 2020. 1, 2, 6

    2020

  30. [38]

    Vi- sion transformers for dense prediction

    Ren ´e Ranftl, Alexey Bochkovskiy, and Vladlen Koltun. Vi- sion transformers for dense prediction. In Proceedings of the IEEE/CVF international conference on computer vision, pages 12179–12188, 2021. 3

    2021

  31. [39]

    Masked representation learn- ing for domain generalized stereo matching

    Zhibo Rao, Bangshu Xiong, Mingyi He, Yuchao Dai, Renjie He, Zhelun Shen, and Xing Li. Masked representation learn- ing for domain generalized stereo matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 5435–5444, 2023. 7

    2023

  32. [40]

    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. 2

    2022

  33. [41]

    High-resolution stereo datasets with subpixel-accurate ground truth

    Daniel Scharstein, Heiko Hirschm ¨uller, York Kitajima, Greg Krathwohl, Nera Ne ˇsi´c, Xi Wang, and Porter West- ling. High-resolution stereo datasets with subpixel-accurate ground truth. In Pattern Recognition: 36th German Confer- ence, GCPR 2014, M¨unster, Germany, September...

    2014

  34. [42]

    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 Proceed- ings of the IEEE conference on computer vision and ...

    2017

  35. [43]

    Cfnet: Cascade and fused cost volume for robust stereo matching

    Zhelun Shen, Yuchao Dai, and Zhibo Rao. Cfnet: Cascade and fused cost volume for robust stereo matching. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 13906–13915, 2021. 2

    2021

  36. [44]

    Pcw-net: Pyramid combination and warping cost volume for stereo matching

    Zhelun Shen, Yuchao Dai, Xibin Song, Zhibo Rao, Dingfu Zhou, and Liangjun Zhang. Pcw-net: Pyramid combination and warping cost volume for stereo matching. In European Conference on Computer Vision , pages 280–297. Springer,

  37. [45]

    Raft: Recurrent all-pairs field transforms for optical flow

    Zachary Teed and Jia Deng. Raft: Recurrent all-pairs field transforms for optical flow. In European conference on com- puter vision, pages 402–419. Springer, 2020. 7

    2020

  38. [46]

    Moge: Unlocking accurate monocular geometry estimation for open-domain images with optimal training supervision

    Ruicheng Wang, Sicheng Xu, Cassie Dai, Jianfeng Xiang, Yu Deng, Xin Tong, and Jiaolong Yang. Moge: Unlocking accurate monocular geometry estimation for open-domain images with optimal training supervision. In Proceedings of the Computer Vision and Pattern Recognition Conferenc...

    2025

  39. [47]

    Selective-stereo: Adaptive frequency information selection for stereo matching

    Xianqi Wang, Gangwei Xu, Hao Jia, and Xin Yang. Selective-stereo: Adaptive frequency information selection for stereo matching. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 19701–19710, 2024. 2, 6, 7, 8, 13

    2024

  40. [48]

    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

  41. [49]

    Foundationstereo: Zero- shot stereo matching

    Bowen Wen, Matthew Trepte, Joseph Aribido, Jan Kautz, Orazio Gallo, and Stan Birchfield. Foundationstereo: Zero- shot stereo matching. arXiv preprint arXiv:2501.09898 ,

    arXiv

  42. [50]

    Atten- tion concatenation volume for accurate and efficient stereo matching

    Gangwei Xu, Junda Cheng, Peng Guo, and Xin Yang. Atten- tion concatenation volume for accurate and efficient stereo matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 12981– 12990, 2022. 2, 6

    2022

  43. [51]

    Iterative geometry encoding volume for stereo matching

    Gangwei Xu, Xianqi Wang, Xiaohuan Ding, and Xin Yang. Iterative geometry encoding volume for stereo matching. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 21919–21928, 2023. 2, 6, 7

    2023

  44. [52]

    Igev++: iterative multi-range geometry encoding volumes for stereo matching

    Gangwei Xu, Xianqi Wang, Zhaoxing Zhang, Junda Cheng, Chunyuan Liao, and Xin Yang. Igev++: iterative multi-range geometry encoding volumes for stereo matching. arXiv preprint arXiv:2409.00638, 2024. 6, 7, 8, 13

    2024 arXiv

  45. [53]

    Depth anything: Unleashing the power of large-scale unlabeled data

    Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything: Unleashing the power of large-scale unlabeled data. In CVPR, 2024. 1, 2

    2024

  46. [54]

    Depth any- thing v2

    Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiao- gang Xu, Jiashi Feng, and Hengshuang Zhao. Depth any- thing v2. Advances in Neural Information Processing Sys- tems, 37:21875–21911, 2024. 1, 2, 3, 7, 12

    2024

  47. [55]

    Metric3d: Towards zero-shot metric 3d prediction from a single image

    Wei Yin, Chi Zhang, Hao Chen, Zhipeng Cai, Gang Yu, Kaixuan Wang, Xiaozhi Chen, and Chunhua Shen. Metric3d: Towards zero-shot metric 3d prediction from a single image. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9043–9053, 2023. 2

    2023

  48. [56]

    Open chal- lenges in deep stereo: the booster dataset

    Pierluigi Zama Ramirez, Fabio Tosi, Matteo Poggi, Samuele Salti, Luigi Di Stefano, and Stefano Mattoccia. Open chal- lenges in deep stereo: the booster dataset. In Proceedings of the IEEE conference on computer vision and pattern recog- nition, 2022. CVPR. 7

    2022

  49. [57]

    Stereo matching by training a convolutional neural network to compare image patches

    Jure ˇZbontar and Yann LeCun. Stereo matching by training a convolutional neural network to compare image patches. Journal of Machine Learning Research, 17(65):1–32, 2016. 1, 2

    2016

  50. [58]

    Ga-net: Guided aggregation net for end-to- end stereo matching

    Feihu Zhang, Victor Prisacariu, Ruigang Yang, and Philip HS Torr. Ga-net: Guided aggregation net for end-to- end stereo matching. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 185–194, 2019. 2, 6

    2019

  51. [59]

    Domain-invariant stereo matching networks

    Feihu Zhang, Xiaojuan Qi, Ruigang Yang, Victor Prisacariu, Benjamin Wah, and Philip Torr. Domain-invariant stereo matching networks. In Europe Conference on Computer Vi- sion (ECCV), 2020. 7

    2020

  52. [60]

    Learning representations from foundation models for domain generalized stereo matching

    Yongjian Zhang, Longguang Wang, Kunhong Li, Yun Wang, and Yulan Guo. Learning representations from foundation models for domain generalized stereo matching. InEuropean Conference on Computer Vision , pages 146–162. Springer,

  53. [61]

    High-frequency stereo match- ing network

    Haoliang Zhao, Huizhou Zhou, Yongjun Zhang, Jie Chen, Yitong Yang, and Yong Zhao. High-frequency stereo match- ing network. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 1327– 1336, 2023. 2, 7 11 BridgeDepth: Bridging Monocular an...

    2023

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.