Pith. sign in

REVIEW 5 minor 67 references

WAFT-Stereo: Warping-Alone Field Transforms for Stereo Matching

T0 review · 0 major / 5 minor · reviewed 2026-07-13 · grok-4.5

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

Pith's one-line read Cost volumes are unnecessary for state-of-the-art stereo matching; high-resolution warping plus one classification step and iterative regression is enough, and faster.

desk verdict Clean empirical win: pure warping plus one soft classification step beats cost-volume stereo on public boards while running faster; the stereo-specific head is the real addition over their own WAFT flow work. read the letter →

arxiv 2603.24836 v3 pith:N5C4T2BK submitted 2026-03-25 cs.CV

Yihan Wang , Jia Deng This is my paper
classification cs.CV
keywords stereomatchingwarpingcostvolumedisparityestimationiterativerefinementsim-to-realgeneralizationDepthAnythingV2
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

Stereo matching estimates how much each pixel shifts horizontally between a pair of rectified camera images, which then gives depth. Leading methods build large cost volumes that compare every possible disparity at every pixel; those volumes are memory-hungry and force the network to work at low resolution. This paper shows that cost volumes can be dropped entirely. Instead, features from the right image are simply warped into alignment with the left using the current disparity guess, then a standard network refines the guess. Because pure regression struggles with the hundreds-of-pixels shifts common in stereo, the authors insert a single soft classification step over a fixed set of disparity bins before the iterative updates begin. The resulting system, WAFT-Stereo, tops the public ETH3D, Middlebury and KITTI leaderboards while running 1.8–6.7 times faster than the previous strongest competitors, and it generalizes well even when trained only on synthetic data.

What carries the argument

Feature-space warping: at each iteration the right-view feature map is backward-warped by the current disparity estimate and concatenated with the left-view features; a shared Vision-Transformer updater then either classifies soft bin probabilities (first step) or regresses a Mixture-of-Laplace residual (subsequent steps).

What would settle it

Replace the first classification step with an ordinary regression step (or with a cost-volume classifier) while keeping every other hyper-parameter fixed; if the hybrid model no longer converges in four iterations or loses its leaderboard rank, the claim that classification-without-cost-volumes is the key fails.

Watch

Extended reading notes

Core claim

Cost volumes are not required for strong stereo matching. A pure warping architecture that first classifies disparity into a modest number of bins and then iteratively regresses residual updates reaches first place on ETH3D (BP-0.5), Middlebury (RMSE) and all KITTI metrics, while cutting latency by 1.8–6.7 imes relative to leading cost-volume methods and reducing zero-shot error on ETH3D by up to 81 percent.

Load-bearing premise

A soft classification over only forty fixed disparity bins supplies an initial estimate accurate enough for later regression to finish the job in four or five iterations on real high-resolution pairs.

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

0 major / 5 minor

Summary. WAFT-Stereo proposes a fully warping-based stereo matcher that dispenses with cost volumes. It extracts features with a LoRA-adapted DepthAnythingV2 backbone, obtains an initial disparity via soft classification over a fixed set of bins (soft-CE loss), then iteratively refines the field by backward-warping the right-view features and regressing residual updates with a Mixture-of-Laplace loss. The architecture is a simplified adaptation of the optical-flow method WAFT, with the classification head, LoRA fine-tuning, and high-resolution ResNet blocks as the main stereo-specific changes. On public leaderboards the method ranks first on ETH3D BP-0.5, Middlebury RMSE and all KITTI metrics, reports large zero-shot gains (especially on ETH3D), and claims 1.8–6.7 imes lower latency than competitive cost-volume systems while releasing code and weights.

Significance. If the empirical claims hold, the paper supplies a clean existence proof that cost volumes—long treated as indispensable—are not required for state-of-the-art stereo accuracy or efficiency. The design is deliberately simple (standard ViT/DPT blocks + warping), scales with backbone size, and generalizes strongly from synthetic data alone. Public leaderboard rankings, matched-data ablations (Table 4), controlled architecture ablations (Table 5), and open code make the result immediately usable and falsifiable. The work therefore has clear practical value for real-time and high-resolution stereo and conceptual value for the broader correspondence literature.

minor comments (5)
  1. Middlebury BP-2 is not first and is dominated by the single hard scene Classroom2E (illumination change). A short quantitative note of the gap after excluding that scene would make the Pareto claim more transparent.
  2. Latency and MAC numbers (Fig. 2, Table 6) are single-GPU (L40, BF16). A brief statement of measurement protocol (batch size, warm-up, whether amortized over multiple runs) would improve reproducibility.
  3. Table 5 shows that B=5 bins hurts while B=20/40/80 are comparable; a one-sentence discussion of how B should scale with Dmax or image resolution would help practitioners.
  4. The soft-target definition in §3.2 uses an absolute-distance kernel before Softmax; a short justification or reference for this particular soft labeling would be useful.
  5. Minor presentation: “diparity” typo in §3.1; inconsistent hyphenation of “zero-shot” / “zero shot”; Fig. 5 caption could more clearly label the three curves.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical architecture claims are validated on independent public benchmarks with standard training data.

full rationale

WAFT-Stereo is an empirical computer-vision paper whose central claims (cost volumes unnecessary; pure high-resolution warping plus one soft-classification step over fixed bins followed by MoL regression iterations yields SOTA accuracy and 1.8–6.7× speed-ups) are supported by training on public synthetic/real corpora and evaluation on external leaderboards (ETH3D, KITTI-2012/2015, Middlebury) whose ground truth is independent of the authors. The architecture is an adaptation of the authors’ prior optical-flow work WAFT, but that citation supplies only the starting design; the stereo-specific modifications (classification head, LoRA fine-tuning, ResNet high-resolution blocks) and all reported metrics are new experiments, not algebraic identities or fitted parameters re-labeled as predictions. No quantity is defined in terms of a later-reported result, no uniqueness theorem is imported to forbid alternatives, and no ansatz is smuggled that forces the leaderboard numbers by construction. The paper is therefore self-contained against external benchmarks; the ordinary self-citation of prior work does not constitute circularity under the stated criteria.

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

As an empirical deep-learning paper the load-bearing premises are standard CV assumptions plus a handful of hand-chosen hyper-parameters that control the classification discretization and iteration schedule. No new physical entities are postulated; the method re-uses off-the-shelf backbones and losses.

free parameters (5)
  • number of disparity bins B = 40
    Chosen by ablation (Table 5); B=40 is the default that balances accuracy and memory; performance drops sharply at B=5.
  • maximum disparity Dmax = 800 (main) / 320 (ablation)
    Hard-coded upper bound for the classification bins; set to 800 for full-resolution models and 320 for ablations.
  • LoRA rank = 8
    Rank of the low-rank adapters used to fine-tune the frozen DepthAnything backbone; fixed at 8 without extensive search.
  • number of high-resolution ResNet blocks = 4
    Replaces the skip connection of the original WAFT; 4 blocks chosen after ablation showing clear accuracy gain.
  • iteration count T and discount γ = T=4 or 5
    Number of refinement steps (including the classification step) and the exponential loss discount; set to 4–5 iterations for reported models.
assumptions (4)
  • domain assumption Feature-space bilinear warping of the right-view features using the current disparity estimate is a sufficient substitute for explicit cost-volume construction.
    Stated in §3.1 and inherited from WAFT; the entire architecture rests on this equivalence holding at half-resolution.
  • ad hoc to paper A soft-cross-entropy classification over a fixed set of disparity bins yields a coarse but stable initial disparity that subsequent regression can refine.
    Introduced in §3.2; ablation (Table 5, Fig. 5) shows it is necessary for large-displacement stereo, yet the binning scheme itself is a design choice.
  • domain assumption DepthAnythingV2 features (frozen + LoRA) already contain the geometric cues needed for accurate stereo matching.
    Used for all leaderboard submissions (§3.3); performance scales with backbone size (Table 5).
  • domain assumption Mixture-of-Laplace loss is superior to L1 for iterative disparity regression.
    Adopted from SEA-RAFT; confirmed by ablation in Table 5.

how reviews work

0 comments
Cite this review

Pith. "Pith review of WAFT-Stereo: Warping-Alone Field Transforms for Stereo Matching." pith.science (2026). https://pith.science/paper/N5C4T2BK

@misc{pith2026260324836,
  author       = {Pith},
  title        = {Pith review of: WAFT-Stereo: Warping-Alone Field Transforms for Stereo Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N5C4T2BK}},
  note         = {Machine review of arXiv:2603.24836}
}
read the original abstract

We introduce WAFT-Stereo, a simple and effective warping-based method for stereo matching. WAFT-Stereo demonstrates that cost volumes, a common design used in many leading methods, are not necessary for strong performance and can be replaced by warping with improved efficiency. WAFT-Stereo ranks first on ETH3D (BP-0.5), Middlebury (RMSE), and KITTI (all metrics), reducing the zero-shot error by 81% on ETH3D, while being 1.8-6.7x faster than competitive methods. Code and model weights are available at https://github.com/princeton-vl/WAFT-Stereo.

Figures

Figures reproduced from arXiv: 2603.24836 by the authors.

Figure 1
Figure 1. WAFT-Stereo achieves strong sim-to-real generalization [1, 29, 39, 41]. art. WAFT-Stereo demonstrates that strong performance does not require cost￾volume-specific designs. Instead, high-resolution warping paired with iterative updates [23, 45, 52] are sufficient. WAFT-stereo is based on WAFT, but involves non-trivial modifications. WAFT can be directly applied to stereo matching—just changing the predic￾tion to 1D … view at source ↗
Figure 2
Figure 2. WAFT-Stereo achieves state-of-the-art performance on Middlebury [39], KITTI-2015 [29], and ETH3D [41] public benchmarks. Our best-performing model reduces the zero-shot error on ETH3D by at least 61%, while being 1.8 − 6.7× faster than leading methods [30,55]. Our real-time model can process 540p stereo pairs at 21 FPS, while maintaining competitive performance. ‘ZS’ denotes zero-shot submissions. 2 by 13% on KITTI-… view at source ↗
Figure 3
Figure 3. WAFT-Stereo consists of three parts: (1) an input encoder that extracts fea￾tures from images; (2) a classification step that estimates probabilities over preset disparity bins, supervised by a soft-cross-entropy loss; and (3) a recurrent updater that takes backward-warped right view features as input and regresses disparity updates, supervised by a Mixture-of-Laplace loss [53] for T − 1 steps. work has used cost vo… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Full cost volumes compute matching costs for all disparity candidates; partial cost volumes compute costs only in a small window around the current disparity es￾timate; warping aligns the target feature using the current estimate and concatenates aligned and reference …
Figure 5
Figure 5. Figure 5: Left: combining classification and regression achieves better performance than using either one alone. Right: the classification step provides a rough estimate, which is later refined by regressions. where p = (ph, pw) is a pixel in the left view. In practice, Warp(dcu…
Figure 6
Figure 6. Figure 6: The latency grows almost linearly with the number of itera￾tive updates, reflecting reduced par￾allelism as the iteration count in￾creases. Method Latency #MACs #Params MonSter++ [7] 212ms 4.92T 0.39B MatchStereo [61] 135ms 0.39T 0.08B BridgeDepth [11] 51ms 1.34T 0.35B…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

67 extracted references · 6 linked inside Pith

  1. [1]

    Science China Information Sciences 63, 1–11 (2020) 2, 4, 9, 11, 12

    Bao, W., Wang, W., Xu, Y., Guo, Y., Hong, S., Zhang, X.: InStereo2k: a large real dataset for stereo matching in indoor scenes. Science China Information Sciences 63, 1–11 (2020) 2, 4, 9, 11, 12

    2020

  2. [2]

    Computer vision and image understanding63(1), 75–104 (1996) 4

    Black, M.J., Anandan, P.: The robust estimation of multiple motions: Parametric and piecewise-smooth flow fields. Computer vision and image understanding63(1), 75–104 (1996) 4

    1996

  3. [3]

    In: European conference on computer vision

    Brox, T., Bruhn, A., Papenberg, N., Weickert, J.: High accuracy optical flow esti- mation based on a theory for warping. In: European conference on computer vision. pp. 25–36. Springer (2004) 4

    2004

  4. [4]

    Butler, D.J., Wulff, J., Stanley, G.B., Black, M.J.: A naturalistic open source movie for optical flow evaluation. pp. 611–625 (2012) 4, 9, 12

    2012

  5. [5]

    arXiv preprint arXiv:2001.10773 (2020) 4, 9, 12

    Cabon, Y., Murray, N., Humenberger, M.: Virtual KITTI 2. arXiv preprint arXiv:2001.10773 (2020) 4, 9, 12

    arXiv 2001

  6. [6]

    Chen, Z., Long, W., Yao, H., Zhang, Y., Wang, B., Qin, Y., Wu, J.: Mocha-stereo: Motif channel attention network for stereo matching. pp. 27768–27777 (2024) 3, 11, 12

    2024

  7. [7]

    arXiv preprint arXiv:2501.08643 (2025) 1, 2, 3, 4, 5, 6, 7, 9, 10, 11, 12, 13, 14

    Cheng, J., Liao, W., Cai, Z., Liu, L., Xu, G., Wang, X., Wang, Y., Yuan, Z., Deng, Y., Zang, J., et al.: Monster++: Unified stereo matching, multi-view stereo, and real-time stereo with monodepth priors. arXiv preprint arXiv:2501.08643 (2025) 1, 2, 3, 4, 5, 6, 7, 9, 10, 11, 12, 13, 14

    arXiv 2025

  8. [8]

    In: International Con- ference on Learning Representations (ICLR) (2021) 7, 8

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al.: An image is worth 16x16 words: Transformers for image recognition at scale. In: International Con- ference on Learning Representations (ICLR) (2021) 7, 8

    2021

Show all 67 references
  1. [9]

    arXiv preprint arXiv:2308.13561 (2023) 1

    Engel, J., Somasundaram, K., Goesele, M., Sun, A., Gamino, A., Turner, A., Ta- lattof, A., Yuan, A., Souti, B., Meredith, B., et al.: Project aria: A new tool for egocentric multi-modal ai research. arXiv preprint arXiv:2308.13561 (2023) 1

    2023 arXiv

  2. [10]

    Geiger, A., Lenz, P., Urtasun, R.: Are we ready for autonomous driving? the KITTI vision benchmark suite. pp. 3354–3361 (2012) 1, 3, 4, 9, 12

    2012

  3. [11]

    In: Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision

    Guan, T., Guo, J., Wang, C., Liu, Y.H.: Bridgedepth: Bridging monocular and stereo reasoning with latent alignment. In: Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision. pp. 27681–27691 (2025) 1, 3, 4, 7, 9, 11, 12, 13, 14, 15 16 Wang and Deng

    2025

  4. [12]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Guan, T., Wang, C., Liu, Y.H.: Neural markov random field for stereo matching. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 5459–5469 (2024) 3

    2024

  5. [13]

    arXiv preprint arXiv:2411.14053 (2024) 3

    Guo, X., Zhang, C., Zhang, Y., Wang, R., Nie, D., Zheng, W., Poggi, M., Zhao, H., Ye, M., Zou, Q., et al.: Stereo anything: Unifying zero-shot stereo matching with large-scale mixed data. arXiv preprint arXiv:2411.14053 (2024) 3

    2024

  6. [14]

    Guo, X., Yang, K., Yang, W., Wang, X., Li, H.: Group-wise correlation stereo network. pp. 3273–3282 (2019) 11

    2019

  7. [15]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 770–778 (2016) 14

    2016

  8. [16]

    IEEE transactions on pattern anal- ysis and machine intelligence35(2), 504–511 (2012) 4

    Hosni, A., Rhemann, C., Bleyer, M., Rother, C., Gelautz, M.: Fast cost-volume filtering for visual correspondence and beyond. IEEE transactions on pattern anal- ysis and machine intelligence35(2), 504–511 (2012) 4

    2012

  9. [17]

    Iclr1(2), 3 (2022) 2, 7, 9

    Hu, E.J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., Chen, W., et al.: Lora: Low-rank adaptation of large language models. Iclr1(2), 3 (2022) 2, 7, 9

    2022

  10. [18]

    In: European conference on computer vision

    Huang, Z., Shi, X., Zhang, C., Wang, Q., Cheung, K.C., Qin, H., Dai, J., Li, H.: Flowformer: A transformer architecture for optical flow. In: European conference on computer vision. pp. 668–685. Springer (2022) 6

    2022

  11. [19]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

    Jahedi, A., Luz, M., Rivinius, M., Bruhn, A.: Ccmr: High resolution optical flow estimation via coarse-to-fine context-guided motion reasoning. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. pp. 6899– 6908 (2024) 6

    2024

  12. [20]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Jiang, H., Lou, Z., Ding, L., Xu, R., Tan, M., Jiang, W., Huang, R.: Defom-stereo: Depth foundation model based stereo matching. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 21857–21867 (2025) 1, 2, 3, 4, 7, 9, 11, 12, 14

    2025

  13. [21]

    Karaev, N., Rocco, I., Graham, B., Neverova, N., Vedaldi, A., Rupprecht, C.: DynamicStereo: Consistent dynamic depth from stereo videos. pp. 13229–13239 (2023) 3, 12

    2023

  14. [22]

    Li, J., Wang, P., Xiong, P., Cai, T., Yan, Z., Yang, L., Liu, J., Fan, H., Liu, S.: Practicalstereomatchingviacascadedrecurrentnetworkwithadaptivecorrelation. pp. 16263–16272 (2022) 4, 9, 11, 12, 13

    2022

  15. [23]

    In: International Conference on 3D Vision (3DV)

    Lipson, L., Teed, Z., Deng, J.: RAFT-Stereo: Multilevel recurrent field transforms for stereo matching. In: International Conference on 3D Vision (3DV). pp. 218–227 (2021) 1, 2, 3, 4, 11, 12, 14, 15

    2021

  16. [24]

    ICLR (2019) 10

    Loshchilov, I.: Decoupled weight decay regularization. ICLR (2019) 10

    2019

  17. [25]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Mayer, N., Ilg, E., Hausser, P., Fischer, P., Cremers, D., Dosovitskiy, A., Brox, T.: 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 pattern recognition. pp. ...

    2016

  18. [26]

    Mayer, N., Ilg, E., Hausser, P., Fischer, P., Cremers, D., Dosovitskiy, A., Brox, T.: A large dataset to train convolutional networks for disparity, optical flow, and scene flow estimation. pp. 4040–4048 (2016) 4, 9, 12, 13

    2016

  19. [27]

    In: Proc

    Mehl, L., Schmalfuss, J., Jahedi, A., Nalivayko, Y., Bruhn, A.: Spring: A high- resolutionhigh-detaildatasetandbenchmarkforsceneflow,opticalflowandstereo. In: Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2023) 4, 9, 11, 12

    2023

  20. [28]

    In: ICCV

    Memin, E., Perez, P.: A multigrid approach for hierarchical motion estimation. In: ICCV. vol. 98, pp. 933–938 (1998) 4 WAFT-Stereo: Warping-Alone Field Transforms for Stereo Matching 17

    1998

  21. [29]

    Menze, M., Geiger, A.: Object scene flow for autonomous vehicles. pp. 3061–3070 (2015) 1, 2, 3, 4, 9, 12, 13

    2015

  22. [30]

    In: Proceedings of the IEEE/CVF International Confer- ence on Computer Vision

    Min, J., Jeon, Y., Kim, J., Choi, M.: S2m2: Scalable stereo matching model for reliable depth estimation. In: Proceedings of the IEEE/CVF International Confer- ence on Computer Vision. pp. 26729–26739 (2025) 1, 3, 4, 5, 6, 7, 9, 10, 11, 12, 14

    2025

  23. [31]

    arXiv preprint arXiv:2511.16993 (2025) 3

    Min, J., Kim, J., Min, C.H., Kim, M., Jeon, Y., Choi, M.: Depthfocus: Controllable depth estimation for see-through scenes. arXiv preprint arXiv:2511.16993 (2025) 3

    2025

  24. [32]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Morimitsu, H., Zhu, X., Cesar, R.M., Ji, X., Yin, X.C.: Dpflow: Adaptive optical flow estimation with a dual-pyramid framework. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 17810–17820 (2025) 6

    2025

  25. [33]

    ACM Transactions on Graphics (ToG)38(6), 1–15 (2019) 12

    Niklaus, S., Mai, L., Yang, J., Liu, F.: 3d ken burns effect from a single image. ACM Transactions on Graphics (ToG)38(6), 1–15 (2019) 12

    2019

  26. [34]

    In: 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

    Patel, M., Yang, F., Qiu, Y., Cadena, C., Scherer, S., Hutter, M., Wang, W.: Tartanground: A large-scale dataset for ground robot perception and navigation. In: 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). pp. 20524–20531. IEEE (2025) 4, 9

    2025

  27. [35]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Raistrick, A., Lipson, L., Ma, Z., Mei, L., Wang, M., Zuo, Y., Kayan, K., Wen, H., Han, B., Wang, Y., et al.: Infinite photorealistic worlds using procedural genera- tion. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 12630–12641 (2023) 4

    2023

  28. [36]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Raistrick, A., Mei, L., Kayan, K., Yan, D., Zuo, Y., Han, B., Wen, H., Parakh, M., Alexandropoulos, S., Lipson, L., et al.: Infinigen indoors: Photorealistic indoor scenes using procedural generation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Re...

    2024

  29. [37]

    IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI) (2023) 4, 9, 11, 12

    Ramirez, P.Z., Costanzino, A., Tosi, F., Poggi, M., Salti, S., Mattoccia, S., Di Ste- fano, L.: Booster: A benchmark for depth from images of specular and transparent surfaces. IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI) (2023) 4, 9, 11, 12

    2023

  30. [38]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Ranftl, R., Bochkovskiy, A., Koltun, V.: Vision transformers for dense prediction. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 12179–12188 (2021) 7, 8

    2021

  31. [39]

    In: Pattern Recognition: 36th German Conference, GCPR 2014, Münster, Germany, September 2-5, 2014, Proceedings 36

    Scharstein, D., Hirschmüller, H., Kitajima, Y., Krathwohl, G., Nešić, N., Wang, X., Westling, P.: High-resolution stereo datasets with subpixel-accurate ground truth. In: Pattern Recognition: 36th German Conference, GCPR 2014, Münster, Germany, September 2-5, 2014, Proceedings...

    2014

  32. [40]

    International journal of computer vision47(1), 7–42 (2002) 4

    Scharstein, D., Szeliski, R.: A taxonomy and evaluation of dense two-frame stereo correspondence algorithms. International journal of computer vision47(1), 7–42 (2002) 4

    2002

  33. [41]

    Schops, T., Schonberger, J.L., Galliani, S., Sattler, T., Schindler, K., Pollefeys, M., Geiger, A.: A multi-view stereo benchmark with high-resolution images and multi-camera videos. pp. 3260–3269 (2017) 2, 3, 9, 12, 13

    2017

  34. [42]

    arXiv preprint arXiv:2508.10104 (2025) 7, 14

    Siméoni, O., Vo, H.V., Seitzer, M., Baldassarre, F., Oquab, M., Jose, C., Khali- dov, V., Szafraniec, M., Yi, S., Ramamonjisoa, M., et al.: Dinov3. arXiv preprint arXiv:2508.10104 (2025) 7, 14

    2025 arXiv

  35. [43]

    In: Artificial intelligence and machine learning for multi- domain operations applications

    Smith, L.N., Topin, N.: Super-convergence: Very fast training of neural networks using large learning rates. In: Artificial intelligence and machine learning for multi- domain operations applications. vol. 11006, pp. 369–386. SPIE (2019) 10 18 Wang and Deng

    2019

  36. [44]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Sun, D., Yang, X., Liu, M.Y., Kautz, J.: Pwc-net: Cnns for optical flow using pyramid, warping, and cost volume. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 8934–8943 (2018) 1, 4

    2018

  37. [45]

    Teed, Z., Deng, J.: RAFT: Recurrent all-pairs field transforms for optical flow. pp. 402–419 (2020) 2, 3, 4, 8, 9

    2020

  38. [46]

    In: Proceedings of the IEEE/CVF conference on computer vision and pat- tern recognition

    Tosi, F., Liao, Y., Schmitt, C., Geiger, A.: Smd-nets: Stereo mixture density net- works. In: Proceedings of the IEEE/CVF conference on computer vision and pat- tern recognition. pp. 8942–8952 (2021) 4, 9, 11, 12

    2021

  39. [47]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops

    Tremblay, J., To, T., Birchfield, S.: Falling things: A synthetic dataset for 3d ob- ject detection and pose estimation. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. pp. 2038–2041 (2018) 4, 9, 11, 12

  40. [48]

    In: IEEE International Conference on Multimedia and Expo (ICME) (2021) 12

    Wang,Q.,Zheng,S.,Yan,Q.,Deng,F.,Zhao,K.,Chu,X.:IRS:Alargenaturalistic indoor robotics stereo dataset to train deep models for disparity and surface normal estimation. In: IEEE International Conference on Multimedia and Expo (ICME) (2021) 12

    2021

  41. [49]

    In: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

    Wang, W., Zhu, D., Wang, X., Hu, Y., Qiu, Y., Wang, C., Hu, Y., Kapoor, A., Scherer, S.: TartanAir: A dataset to push the limits of visual slam. In: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). pp. 4909–4916 (2020) 4, 9, 12, 13

    2020

  42. [50]

    Wang, X., Xu, G., Jia, H., Yang, X.: Selective-Stereo: Adaptive frequency infor- mation selection for stereo matching. pp. 19701–19710 (2024) 3, 9, 10, 11, 12

    2024

  43. [51]

    arXiv preprint arXiv:2507.13347 (2025) 7

    Wang, Y., Zhou, J., Zhu, H., Chang, W., Zhou, Y., Li, Z., Chen, J., Pang, J., Shen, C., He, T.: Pi3: Permutation-equivariant visual geometry learning. arXiv preprint arXiv:2507.13347 (2025) 7

    2025 arXiv

  44. [52]

    arXiv preprint arXiv:2506.21526 (2025) 1, 2, 4, 5, 6, 7, 9

    Wang, Y., Deng, J.: Waft: Warping-alone field transforms for optical flow. arXiv preprint arXiv:2506.21526 (2025) 1, 2, 4, 5, 6, 7, 9

    2025

  45. [53]

    In: European Conference on Computer Vision

    Wang, Y., Lipson, L., Deng, J.: Sea-raft: Simple, efficient, accurate raft for optical flow. In: European Conference on Computer Vision. pp. 36–54. Springer (2024) 4, 6, 8, 14

    2024

  46. [54]

    Weinzaepfel, P., Lucas, T., Leroy, V., Cabon, Y., Arora, V., Brégier, R., Csurka, G., Antsfeld, L., Chidlovskii, B., Revaud, J.: CroCo v2: Improved cross-view com- pletion pre-training for stereo matching and optical flow. pp. 17969–17980 (2023) 3, 9, 11, 12

    2023

  47. [55]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Wen, B., Trepte, M., Aribido, J., Kautz, J., Gallo, O., Birchfield, S.: Foundation- stereo: Zero-shot stereo matching. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 5249–5260 (2025) 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15

    2025

  48. [56]

    Wu, C.Y., Wang, J., Hall, M., Neumann, U., Su, S.: Toward practical monocular indoordepthestimation.In:ProceedingsoftheIEEE/CVFconferenceoncomputer vision and pattern recognition. pp. 3814–3824 (2022) 12

    2022

  49. [57]

    In: Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition

    Xu, G., Wang, X., Ding, X., Yang, X.: Iterative geometry encoding volume for stereo matching. In: Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition. pp. 21919–21928 (2023) 1, 3, 6, 8, 12, 15

    2023

  50. [58]

    arXiv preprint arXiv:2409.00638 (2024) 3, 6

    Xu, G., Wang, X., Zhang, Z., Cheng, J., Liao, C., Yang, X.: IGEV++: Itera- tive multi-range geometry encoding volumes for stereo matching. arXiv preprint arXiv:2409.00638 (2024) 3, 6

    2024 arXiv

  51. [59]

    IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI) (2023) 3, 12 WAFT-Stereo: Warping-Alone Field Transforms for Stereo Matching 19

    Xu, H., Zhang, J., Cai, J., Rezatofighi, H., Yu, F., Tao, D., Geiger, A.: Unifying flow, stereo and depth estimation. IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI) (2023) 3, 12 WAFT-Stereo: Warping-Alone Field Transforms for Stereo Matching 19

    2023

  52. [60]

    Yan, D., Raistrick, A., Deng, J.: What makes good synthetic training data for zero-shot stereo matching? arXiv preprint arXiv:2504.16930 (2025) 4, 9

    2025

  53. [61]

    arXiv preprint arXiv:2510.14260 (2025) 9, 11, 12, 14

    Yan, T., Liu, T., Yang, X., Zhao, Q., Xia, Z.: Matchattention: Matching the relative positions for high-resolution cross-view matching. arXiv preprint arXiv:2510.14260 (2025) 9, 11, 12, 14

    2025 arXiv

  54. [62]

    Yang, G., Manela, J., Happold, M., Ramanan, D.: Hierarchical deep stereo match- ing on high-resolution images. pp. 5515–5524 (2019) 9, 11, 12

    2019

  55. [63]

    Yang, G., Song, X., Huang, C., Deng, Z., Shi, J., Zhou, B.: DrivingStereo: A large- scale dataset for stereo matching in autonomous driving scenarios. pp. 899–908 (2019) 4

    2019

  56. [64]

    Yang, L., Kang, B., Huang, Z., Xu, X., Feng, J., Zhao, H.: Depth anything: Un- leashing the power of large-scale unlabeled data. pp. 10371–10381 (2024) 2, 3, 7, 10, 11, 12, 14

    2024

  57. [65]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Zbontar, J., LeCun, Y.: Computing the stereo matching cost with a convolutional neural network. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 1592–1599 (2015) 1, 3, 4

    2015

  58. [66]

    Zhao, H., Zhou, H., Zhang, Y., Chen, J., Yang, Y., Zhao, Y.: High-frequency stereo matching network. pp. 1327–1336 (2023) 4, 7, 11, 13

    2023

  59. [67]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Zhou, S., Zhang, J., Zuo, W., Xie, H., Pan, J., Ren, J.S.: Davanet: Stereo deblurring with view aggregation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 10996–11005 (2019) 12

    2019

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.