Pith. sign in

REVIEW 3 major objections 5 minor 53 references

Uncertainty-guided residual search keeps local cost volumes on the true disparity, fixing unrecoverable failures in real-time coarse-to-fine stereo matching.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-10 21:28 UTC pith:MLNE5NW6

load-bearing objection Useful engineering fix for a real coarse-to-fine failure mode, but the manuscript is unfinished and the uncertainty gate itself is never ablated. the 3 major comments →

arxiv 2607.06779 v1 pith:MLNE5NW6 submitted 2026-07-07 cs.CV

URS-Stereo: Uncertainty-Guided Residual Search for Real-Time Stereo Matching

Pouya Sohrabipour , Chaitanya kumar reddy Pallerla , Dongyi Wang This is my paper
classification cs.CV
keywords stereo matchingreal-time stereocoarse-to-fineuncertainty estimationresidual searchcost volumedisparity refinementzero-shot generalization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Coarse-to-fine stereo matching builds small local cost volumes around a disparity estimate carried up from a coarser scale. When that estimate is wrong, the true match can fall outside the search window and later stages cannot recover. This paper claims that predicting how reliable the propagated disparity is, then using that reliability to shift the search-window center by a residual offset, keeps the true correspondence inside the window far more often. The resulting URS-Stereo network improves zero-shot disparity accuracy on driving and indoor benchmarks while still running in real time. A reader who needs dense depth on robots or cars cares because the method attacks the accuracy–speed trade-off at the exact point where hierarchical stereo usually breaks.

Core claim

URS-Stereo shows that an Uncertainty-Guided Residual Search Module can predict both a reliability score for the propagated disparity and a residual offset, form an adjusted search center as the sum of the predicted disparity and the uncertainty-modulated offset, and thereby keep ground-truth correspondences inside local cost volumes at the 1/8 and 1/4 stages. This yields consistently lower zero-shot error on KITTI 2012, KITTI 2015, Middlebury and ETH3D while preserving real-time inference.

What carries the argument

The Uncertainty-Guided Residual Search Module (UGRSM): from image features and the propagated disparity it predicts uncertainty U and an initial residual offset, multiplies them so that only uncertain pixels move the window, and recenters the local cost volume before refinement.

Load-bearing premise

The method assumes that training the uncertainty head on a simple exponential of absolute disparity error produces a reliability score that correctly decides, on unseen real images, when and how far to move the local search window.

What would settle it

If an ablation that always applies the full residual offset (no uncertainty gate) matches or beats full UGRSM on the same SceneFlow-to-KITTI/Middlebury zero-shot metrics, or if measured rates show ground-truth disparities still fall outside the adjusted local windows as often as the unadjusted baseline, the central claim that uncertainty guidance is what restores robustness would be falsified.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Coarse-to-fine stereo networks can recover from coarse-stage errors that previously made later stages fail permanently.
  • Real-time stereo systems gain accuracy on boundaries, thin structures and large disparity jumps without building full high-resolution cost volumes.
  • Uncertainty-gated residual offsets improve zero-shot transfer from synthetic training to real driving and indoor scenes.
  • Adaptive search centers shrink the accuracy–speed trade-off that has limited stereo depth on robots and embedded platforms.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same uncertainty-gated offset idea could transfer to other hierarchical correspondence tasks (optical flow, multi-view stereo) that also search locally around a propagated estimate.
  • If the absolute-error pseudo-label for uncertainty degrades under domain shift, photometric or learned reliability cues could replace it as the gate.
  • The monocular branch already present in the architecture could pre-bias residual offsets in textureless regions where stereo uncertainty is highest.
  • Many “unrecoverable” failures in cascaded vision pipelines may be search-center misalignments fixable by a cheap confidence-modulated shift rather than heavier regularization.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes URS-Stereo, a real-time coarse-to-fine stereo matcher whose central novelty is the Uncertainty-Guided Residual Search Module (UGRSM). At the 1/8 and 1/4 stages, UGRSM predicts an uncertainty map U from left features and the upsampled disparity, predicts an unconstrained residual offset, multiplies them (Eqs. 9–13) to obtain an adjusted local-cost-volume center, and thereby tries to keep the ground-truth correspondence inside a small search window. The rest of the pipeline follows recent hybrid mono–stereo designs (full GWC cost volume + ConvGRU at 1/16, mono-stereo mutual refinement and DEM at 1/4). Training is fully supervised on SceneFlow with an L1 stereo loss plus a pseudo-uncertainty loss whose target is a simple function of absolute disparity error (Eq. 10). Zero-shot numbers on KITTI 2012/2015 and Middlebury (Table 1) are competitive with recent efficient baselines.

Significance. If the uncertainty gate is truly responsible for the reported gains, the work supplies a lightweight, practically useful fix for a well-known failure mode of coarse-to-fine stereo (search-center drift). The idea is simple, the zero-shot protocol is clean, and the numbers on three real benchmarks are strong relative to other SceneFlow-only efficient methods. The contribution is incremental rather than foundational, but it is of clear interest to the real-time stereo community provided the ablation evidence is completed.

major comments (3)
  1. The load-bearing claim that uncertainty guidance (rather than unconstrained residual offsets) is what improves robustness is never isolated. Immediately before Table 1 the authors themselves state that an “offset-only” variant already helps but can harm reliable regions, and that uncertainty modulation yields the further gain. Yet Table 1 and the entire experimental section report only the full model. A quantitative comparison of (i) no residual, (ii) offset-only (U≡1), and (iii) full UGRSM on the same zero-shot benchmarks is required to substantiate the central claim of §3.4 and the abstract.
  2. The abstract and introduction promise “extensive experiments on SceneFlow, KITTI 2012, KITTI 2015, Middlebury, and ETH3D,” and Fig. 1 caption mentions MDEM and DEM. The manuscript body contains only Table 1 (three real benchmarks) and no SceneFlow or ETH3D numbers, no runtime/FPS table, and no architectural or ablation description of MDEM/DEM. These omissions leave the empirical support incomplete relative to the paper’s own claims and make it impossible to verify real-time performance or the contribution of the mono-stereo components.
  3. Eq. (10) defines the sole training signal for U as a monotonic function of absolute disparity error of the *propagated* map. No analysis or qualitative evidence is given that the resulting U, at test time on real images, actually correlates with search-center reliability rather than with residual matching difficulty. Because the gate is the only mechanism that distinguishes UGRSM from a free residual offset, this assumption needs at least a brief validation (e.g., correlation of U with final error, or visualizations of where the offset is applied).
minor comments (5)
  1. Typos and grammar: “in a end-to-end maner” (§3.4), “Mono-Disparity Estimation Module (MDEM)” appears only in the Fig. 1 caption and is never defined in the text, inconsistent capitalization of “Search” vs “search”.
  2. Eq. (2) writes the softmax with a missing closing parenthesis and an awkward fraction layout; the probability should be normalized over the disparity dimension more clearly.
  3. Hyper-parameters β, λ_unc, search radius r, δ^(s), number of ConvGRU iterations N_s and the decay γ are never given numerical values or sensitivity analysis.
  4. Related-work discussion of RT-MonSter / MonSter++ is repeated almost verbatim in two consecutive paragraphs (§2.3).
  5. Table 1 caption says “Best results are shown in red, and second-best \ldots in orange,” but the submitted text contains no color markup, making the ranking hard to read.

Circularity Check

0 steps flagged

No circularity: standard supervised coarse-to-fine architecture with pseudo-labels from GT error and zero-shot external benchmarks; nothing reduces by construction to its inputs.

full rationale

URS-Stereo proposes an architectural module (UGRSM) whose equations (9–14) define a learned uncertainty map U and residual offset that are trained end-to-end with ordinary L1 losses against SceneFlow ground truth (Eqs. 17–19) plus a pseudo-target U_gt derived from absolute disparity error (Eq. 10). At inference the network is applied without access to GT, and all reported numbers are zero-shot evaluations on independent real-world benchmarks (KITTI 2012/2015, Middlebury, ETH3D). There is no self-definitional loop, no fitted constant re-labeled as a prediction, no load-bearing self-citation of a uniqueness theorem, and no ansatz smuggled from prior author work that forces the result. The usual supervised-learning dependence on training labels is present but does not constitute circular derivation under the stated criteria; the paper is self-contained against external data.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 2 invented entities

The central claim rests on standard stereo-matching assumptions plus a handful of free hyperparameters that control the pseudo-uncertainty target and the local search geometry. No new physical entities are postulated; the invented modules are engineering constructs whose only evidence is the reported tables. The pseudo-uncertainty definition is an ad-hoc modeling choice that is never independently validated.

free parameters (5)
  • β (uncertainty target sensitivity)
    Controls how absolute disparity error is mapped to the pseudo-label U_gt = 1-exp(-β|error|). Value never stated; directly shapes the training signal for the uncertainty head.
  • λ_unc (uncertainty loss weight)
    Balances L_stereo against L_unc in the joint objective; value unreported.
  • search radius r and interval δ^(s)
    Define the local disparity candidate set around the adjusted center; chosen by designers and never ablated in the text.
  • γ (exponential loss decay)
    Weights later GRU iterations more heavily; standard but free.
  • number of ConvGRU iterations N_s per scale
    Controls refinement budget and accuracy/speed trade-off; unreported.
axioms (4)
  • domain assumption Group-wise correlation cost volumes plus 3D conv regularization produce usable disparity probability distributions at 1/16 resolution.
    Standard stereo practice invoked in §3.3; not re-derived.
  • domain assumption ConvGRU iterative updates (RAFT-style) can refine an initial disparity given geometry features from a local cost volume.
    Taken from RAFT-Stereo/IGEV citations; used without re-proof.
  • ad hoc to paper Absolute disparity error is a sufficient proxy for the reliability of the *propagated search center* (Eq. 10).
    Core modeling choice of UGRSM; no independent calibration or human-annotated uncertainty is provided.
  • domain assumption SceneFlow-trained models transfer zero-shot to real KITTI/Middlebury imagery under the reported metrics.
    Common evaluation protocol in the efficient-stereo literature; assumed valid for claiming generalization.
invented entities (2)
  • Uncertainty-Guided Residual Search Module (UGRSM) no independent evidence
    purpose: Predict U and ΔO_init, form ΔO_final = U·ΔO_init, and relocate local cost-volume centers before refinement.
    The paper's main architectural contribution; evidence is solely the end-to-end disparity tables, with no external falsifiable prediction (e.g., calibrated uncertainty scores on held-out sensors).
  • Mono-Disparity Estimation Module (MDEM) / Disparity Estimation Module (DEM) no independent evidence
    purpose: Provide monocular guidance and full-resolution reconstruction via learned disparity derivatives (mentioned in Fig. 1).
    Named in the overview figure and abstract-adjacent text but given almost no equations or ablation; treated as black-box components whose necessity is unproven.

pith-pipeline@v1.1.0-grok45 · 15572 in / 3520 out tokens · 37625 ms · 2026-07-10T21:28:30.037929+00:00 · methodology

0 comments
read the original abstract

Real-time stereo matching is crucial for robotics, autonomous systems, and embedded vision applications, where both computational efficiency and disparity accuracy are required. Recent coarse-to-fine stereo matching methods improve efficiency by progressively refining disparity estimates using local cost volumes at higher resolutions. However, these methods rely heavily on the accuracy of propagated disparity estimates from previous stages. When the propagated disparity is inaccurate, the ground-truth correspondence may fall outside the predefined local search range, leading to unrecoverable matching failures during subsequent refinement. In this paper, we propose URS-Stereo, a real-time coarse-to-fine stereo matching framework that addresses this limitation through uncertainty-guided search adaptation. Specifically, we introduce an Uncertainty-Guided Residual Search Module (UGRSM), which predicts the reliability of propagated disparities together with residual search offsets to adaptively relocate the centers of local cost volumes before disparity refinement. By dynamically adjusting the search region according to the confidence of the propagated disparity, the proposed method significantly improves the robustness of local correspondence estimation while preserving the computational efficiency of coarse-to-fine stereo matching. Extensive experiments on SceneFlow, KITTI 2012, KITTI 2015, Middlebury, and ETH3D demonstrate that URS-Stereo consistently improves disparity estimation while maintaining real-time inference speed, validating the effectiveness of the proposed uncertainty-guided search strategy

Figures

Figures reproduced from arXiv: 2607.06779 by Chaitanya kumar reddy Pallerla, Dongyi Wang, Pouya Sohrabipour.

Figure 1
Figure 1. Figure 1: Overview of the proposed URS-Stereo framework. A full cost volume [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of the proposed Uncertainty-Guided Residual Search [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 53 canonical work pages · 9 internal anchors

  1. [1]

    Iterative geometry encoding volume for stereo matching,

    G. Xu, X. Wang, X. Ding, and X. Yang, “Iterative geometry encoding volume for stereo matching,” inCVPR, 2023

    work page 2023

  2. [2]

    Foun- dationstereo: Zero-shot stereo matching,

    B. Wen, M. Trepte, J. Aribido, J. Kautz, O. Gallo, and S. Birchfield, “Foun- dationstereo: Zero-shot stereo matching,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

    work page 2025

  3. [3]

    Global occlusion-aware transformer for robust stereo matching,

    Z. Liu, Y. Li, and M. Okutomi, “Global occlusion-aware transformer for robust stereo matching,” inProceedings of the IEEE/CVF Winter Confer- ence on Applications of Computer Vision (WACV), 2024

    work page 2024

  4. [4]

    A Wavelet-based Stereo Matching Framework for Solving Frequency Convergence Inconsistency

    X. Wei, J. Liu, D. Yang, J. Cheng, C. Shu, and W. Wang, “A wavelet- based stereo matching framework for solving frequency convergence incon- sistency,”arXiv preprint arXiv:2505.18024, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    GR-Gaussian: Graph-Based Radiative Gaussian Splatting for Sparse-View CT Reconstruction

    X. Wang, H. Yang, H. Wang, J. Cheng, G. Xu, M. Lin, and X. Yang, “Promptstereo: Zero-shot stereo matching via structure and motion prompts,”arXiv preprint arXiv:2508.02408, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  6. [6]

    Mocha- stereo: Motif channel attention network for stereo matching,

    Z. Chen, W. Long, H. Yao, Y. Zhang, B. Wang, Y. Qin, and J. Wu, “Mocha- stereo: Motif channel attention network for stereo matching,” inProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition (CVPR), 2024

    work page 2024

  7. [7]

    Monster++: Unified stereo matching, multi-view stereo, and real-time stereo with monodepth priors,

    J. Cheng, W. Liao, Z. Cai, L. Liu, G. Xu, X. Wang, Y. Wang, Z. Yuan, Y. Deng, J. Zang, Y. Shi, J. Tang, and X. Yang, “Monster++: Unified stereo matching, multi-view stereo, and real-time stereo with monodepth priors,” 2025

    work page 2025

  8. [8]

    Raft-stereo: Multilevel recurrent field transforms for stereo matching,

    L. Lipson, Z. Teed, and J. Deng, “Raft-stereo: Multilevel recurrent field transforms for stereo matching,” in3DV, 2021

    work page 2021

  9. [9]

    Anytime stereo image depth estimation on mobile devices,

    Y. Wang, Z. Lai, G. Huang, B. H. Wang, L. van der Maaten, M. Campbell, and K. Q. Weinberger, “Anytime stereo image depth estimation on mobile devices,” inIEEE International Conference on Robotics and Automation (ICRA), 2019

    work page 2019

  10. [10]

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

    A. Kendall, H. Martirosyan, S. Dasgupta, P. Henry, R. Kennedy, A. Bachrach, and A. Bry, “End-to-end learning of geometry and context for deep stereo regression,” inICCV, pp. 66–75, 2017

    work page 2017

  11. [11]

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

    F. Zhang, V. Prisacariu, R. Yang, and P. H. S. Torr, “Ga-net: Guided aggregation net for end-to-end stereo matching,” inCVPR, pp. 185–194, 2019

    work page 2019

  12. [12]

    Cfnet: Cascade and fused cost volume for robust stereo matching,

    Z. Shen, Y. Dai, and Z. Rao, “Cfnet: Cascade and fused cost volume for robust stereo matching,” inCVPR, pp. 13906–13915, 2021. 12

    work page 2021

  13. [13]

    Uas- net: Uncertainty adaptive sampling network for deep stereo matching,

    Y. Mao, Z. Liu, W. Li, Y. Dai, Q. Wang, Y.-T. Kim, and H.-S. Lee, “Uas- net: Uncertainty adaptive sampling network for deep stereo matching,” in CVPR, pp. 6311–6319, 2021

    work page 2021

  14. [14]

    High- frequency stereo matching network,

    H. Zhao, H. Zhou, Y. Zhang, J. Chen, Y. Yang, and Y. Zhao, “High- frequency stereo matching network,” inProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pp. 1327–1336, 2023

    work page 2023

  15. [15]

    MC-Stereo: Multi-peak Lookup and Cascade Search Range for Stereo Matching

    M. Feng, J. Cheng, H. Jia, L. Liu, G. Xu, and X. Yang, “Mc-stereo: Multi- peak lookup and cascade search range for stereo matching,”arXiv preprint arXiv:2311.02340, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  16. [16]

    Unifying flow, stereo and depth estimation,

    H. Xu, J. Zhang, J. Cai, H. Rezatofighi, F. Yu, D. Tao, and A. Geiger, “Unifying flow, stereo and depth estimation,”IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 2023

    work page 2023

  17. [17]

    Computing the stereo matching cost with a convolutional neural network,

    J. Zbontar and Y. LeCun, “Computing the stereo matching cost with a convolutional neural network,” inCVPR, pp. 1592–1599, 2015

    work page 2015

  18. [18]

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

    N. Mayer, E. Ilg, P. H¨ ausser, P. Fischer, D. Cremers, A. Dosovitskiy, and T. Brox, “A large dataset to train convolutional networks for disparity, optical flow, and scene flow estimation,” inCVPR, pp. 4040–4048, 2016

    work page 2016

  19. [19]

    Group-wise correlation stereo network,

    X. Guo, K. Yang, W. Yang, X. Wang, and H. Li, “Group-wise correlation stereo network,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 3273–3282, 2019

    work page 2019

  20. [20]

    Correlate-and-excite: Real-time stereo matching via guided cost volume excitation,

    A. Bangunharcana, J. W. Cho, S. Lee, I. S. Kweon, K.-S. Kim, and S. Kim, “Correlate-and-excite: Real-time stereo matching via guided cost volume excitation,” inCVPR, pp. 1959–1968, 2021

    work page 1959

  21. [21]

    Hierarchical neural architecture search for deep stereo matching,

    X. Cheng, Y. Zhong, M. Harandi, Y. Dai, X. Chang, H. Li, T. Drum- mond, and Z. Ge, “Hierarchical neural architecture search for deep stereo matching,”Advances in Neural Information Processing Systems, vol. 33, 2020

    work page 2020

  22. [22]

    Deeppruner: Learning efficient stereo matching via differentiable patchmatch,

    S. Duggal, S. Wang, W.-C. Ma, R. Hu, and R. Urtasun, “Deeppruner: Learning efficient stereo matching via differentiable patchmatch,” inICCV, 2019

    work page 2019

  23. [23]

    Fast-FoundationStereo: Real-time zero-shot stereo matching,

    B. Wen, S. Dewan, and S. Birchfield, “Fast-FoundationStereo: Real-time zero-shot stereo matching,”CVPR, 2026

    work page 2026

  24. [24]

    Pyramid stereo matching network,

    J.-R. Chang and Y.-S. Chen, “Pyramid stereo matching network,” in CVPRW, 2018

    work page 2018

  25. [25]

    Cascade cost volume for high-resolution multi-view stereo and stereo matching,

    X. Gu, Z. Fan, S. Zhu, Z. Dai, and P. Tan, “Cascade cost volume for high-resolution multi-view stereo and stereo matching,” inCVPR, 2020. 13

    work page 2020

  26. [26]

    MatchAttention: Embedding Explicit Matching Constraints into Attention for Efficient Stereo Matching

    T. Yan, T. Liu, X. Yang, Q. Zhao, and Z. Xia, “Matchattention: Match- ing the relative positions for high-resolution cross-view matching,”arXiv preprint arXiv:2510.14260, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  27. [27]

    Mafnet: Multi-frequency adaptive fusion network for real-time stereo matching,

    A. Xu, R. Zhao, X. Xu, B. Huang, Y. Jia, H. Long, F. Chen, Z. Cao, and F. Chen, “Mafnet: Multi-frequency adaptive fusion network for real-time stereo matching,”arXiv preprint arXiv:2512.04358, 2026

    work page arXiv 2026

  28. [28]

    Lite any stereo: Efficient zero-shot stereo matching,

    J. Jing, W. Luo, Y. Mao, and K. Mikolajczyk, “Lite any stereo: Efficient zero-shot stereo matching,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 21725–21735, June 2026

    work page 2026

  29. [29]

    Stereo matching in time: 100+ fps video stereo matching for extended reality,

    Z. Cheng, J. Yang, and H. Li, “Stereo matching in time: 100+ fps video stereo matching for extended reality,” inWACV, 2024

    work page 2024

  30. [30]

    Hitnet: Hierarchical iterative tile refinement network for real-time stereo matching,

    V. Tankovich, C. Hane, S. Fanello, C. Rhemann, S. Izadi, and A. Kowdle, “Hitnet: Hierarchical iterative tile refinement network for real-time stereo matching,” inCVPR, pp. 14362–14372, 2021

    work page 2021

  31. [31]

    Unsupervised monocular depth estimation with left-right consistency,

    C. Godard, O. M. Aodha, and G. J. Brostow, “Unsupervised monocular depth estimation with left-right consistency,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017

    work page 2017

  32. [32]

    Digging into self-supervised monocular depth prediction,

    C. Godard, O. Mac Aodha, M. Firman, and G. J. Brostow, “Digging into self-supervised monocular depth prediction,” October 2019

    work page 2019

  33. [33]

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

    R. Ranftl, K. Lasinger, D. Hafner, K. Schindler, and V. Koltun, “Towards robust monocular depth estimation: Mixing datasets for zero-shot cross- dataset transfer,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 3, 2022

    work page 2022

  34. [34]

    MiDaS v3.1 -- A Model Zoo for Robust Monocular Relative Depth Estimation

    R. Birkl, D. Wofk, and M. M¨ uller, “Midas v3.1 – a model zoo for robust monocular relative depth estimation,”arXiv preprint arXiv:2307.14460, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  35. [36]

    Depth map prediction from a single image using a multi-scale deep network,

    D. Eigen, C. Puhrsch, and R. Fergus, “Depth map prediction from a single image using a multi-scale deep network,” inNeurIPS, pp. 2366–2374, 2014

    work page 2014

  36. [37]

    Deeper depth prediction with fully convolutional residual networks,

    I. Laina, C. Rupprecht, V. Belagiannis, F. Tombari, and N. Navab, “Deeper depth prediction with fully convolutional residual networks,” in3D Vision (3DV), 2016 Fourth International Conference on, pp. 239–248, IEEE, 2016

    work page 2016

  37. [38]

    Unsupervised monocular depth estimation with left-right consistency,

    C. Godard, O. M. Aodha, and G. J. Brostow, “Unsupervised monocular depth estimation with left-right consistency,” inCVPR, pp. 270–279, 2017

    work page 2017

  38. [39]

    Depth Anything V2

    L. Yang, B. Kang, Z. Huang, Z. Zhao, X. Xu, J. Feng, and H. Zhao, “Depth anything v2,”arXiv:2406.09414, 2024. 14

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  39. [40]

    DINOv2: Learning Robust Visual Features without Supervision

    M. Oquabet al., “Dinov2: Learning robust visual features without super- vision,” 2023. arXiv:2304.07193

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  40. [41]

    ZoeDepth: Zero-shot Transfer by Combining Relative and Metric Depth

    S. F. Bhat, R. Birkl, D. Wofk, P. Wonka, and M. M¨ uller, “Zoedepth: Zero-shot transfer by combining relative and metric depth,”arXiv preprint arXiv:2302.12288, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  41. [42]

    Towards zero-shot scale-aware monocular depth estimation,

    V. Guizilini, I. Vasiljevic, D. Chen, R. Ambrus, and A. Gaidon, “Towards zero-shot scale-aware monocular depth estimation,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Octo- ber 2023

    work page 2023

  42. [43]

    RSA: Resolving Scale Ambiguities in Monocular Depth Estimators through Language Descriptions

    Z. Zeng, Y. Wu, H. Park, D. Wang, F. Yang, S. Soatto, D. Lao, B.-W. Hong, and A. Wong, “Rsa: Resolving scale ambiguities in monocular depth esti- mators through language descriptions,”arXiv preprint arXiv:2410.02924, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  43. [44]

    Monster: Marry monodepth to stereo unleashes power,

    J. Chenget al., “Monster: Marry monodepth to stereo unleashes power,” inCVPR, 2025

    work page 2025

  44. [45]

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

    A. Geiger, P. Lenz, and R. Urtasun, “Are we ready for autonomous driving? the kitti vision benchmark suite,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 3354–3361, 2012

    work page 2012

  45. [46]

    Object scene flow for autonomous vehicles,

    M. Menze and A. Geiger, “Object scene flow for autonomous vehicles,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 3061–3070, 2015

    work page 2015

  46. [47]

    A taxonomy and evaluation of dense two- frame stereo correspondence algorithms,

    D. Scharstein and R. Szeliski, “A taxonomy and evaluation of dense two- frame stereo correspondence algorithms,”International Journal of Com- puter Vision, vol. 47, no. 1–3, pp. 7–42, 2002

    work page 2002

  47. [48]

    Mobilestereonet: To- wards lightweight deep networks for stereo matching,

    F. Shamsafar, S. Woerz, R. Rahim, and A. Zell, “Mobilestereonet: To- wards lightweight deep networks for stereo matching,” inProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pp. 2417–2426, 2022

    work page 2022

  48. [49]

    Attention concatenation vol- ume for accurate and efficient stereo matching,

    G. Xu, J. Cheng, P. Guo, and X. Yang, “Attention concatenation vol- ume for accurate and efficient stereo matching,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 12981–12990, 2022

    work page 2022

  49. [50]

    Accurate and efficient stereo matching via attention concatenation volume,

    G. Xu, Y. Wang, J. Cheng, J. Tang, and X. Yang, “Accurate and efficient stereo matching via attention concatenation volume,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 4, pp. 2461–2474, 2024

    work page 2024

  50. [51]

    Uncertainty guided adaptive warping for robust and efficient stereo matching,

    J. Jing, J. Li, P. Xiong, J. Liu, S. Liu, Y. Guo, X. Deng, M. Xu, L. Jiang, and L. Sigal, “Uncertainty guided adaptive warping for robust and efficient stereo matching,” inProceedings of the IEEE/CVF International Confer- ence on Computer Vision (ICCV), pp. 3318–3327, 2023. 15

    work page 2023

  51. [52]

    Lightstereo: Channel boost is all your need for efficient 2d cost aggrega- tion,

    X. Guo, C. Zhang, Y. Zhang, W. Zheng, D. Nie, M. Poggi, and L. Chen, “Lightstereo: Channel boost is all your need for efficient 2d cost aggrega- tion,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025

    work page 2025

  52. [53]

    Banet: Bilateral aggregation network for mobile stereo matching,

    G. Xu, J. Liu, X. Wang, J. Cheng, Y. Deng, J. Zang, Y. Chen, and X. Yang, “Banet: Bilateral aggregation network for mobile stereo matching,” inPro- ceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2025

    work page 2025

  53. [54]

    Lite any stereo: Efficient zero-shot stereo matching,

    J. Jing, W. Luo, Y. Mao, and K. Mikolajczyk, “Lite any stereo: Efficient zero-shot stereo matching,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026. 16

    work page 2026