Pith. sign in

REVIEW 3 major objections 52 references

Performance of universal machine-learned potentials with explicit long-range interactions in biomolecular simulations

T0 review · 3 major / 0 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 Universal machine-learned potentials, as currently trained and evaluated, do not reliably transfer to biomolecular simulations: benchmark gains from larger models do not carry over to simulated properties, and explicit long-range electrosta

desk verdict The abstract describes a useful benchmark of universal ML potentials with explicit long-range terms, but the supplied body is a different stereo-matching paper, so none of the claims can be checked. read the letter →

arxiv 2508.10841 v1 pith:KWN2YDTB submitted 2025-08-14 physics.chem-ph cond-mat.softcs.LGphysics.comp-ph

Viktor Zaverkin , Matheus Ferraz , Francesco Alesiani , Mathias Niepert This is my paper
classification physics.chem-phcond-mat.softcs.LGphysics.comp-ph
keywords machine-learnedpotentialsbiomolecularsimulationlong-rangeelectrostaticsequivariantmessage-passingSPICE-v2transferabilitymodelsizeevaluationpractices
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

This paper argues that universal machine-learned potentials, despite their promise of transferable accuracy, do not yet deliver reliable predictions for biomolecular simulations. The authors train equivariant message-passing models on the SPICE-v2 dataset, with and without explicit long-range dispersion and electrostatics, and compare benchmark errors against properties obtained from simulations of liquid water, aqueous NaCl, alanine tripeptide, Trp-cage, and Crambin. They find that larger models improve standard benchmark metrics without consistently improving simulated observables, that simulation predictions depend on training-data composition, and that explicit long-range electrostatics have no systematic benefit across systems—though they do increase conformational variability in Trp-cage. The paper concludes that imbalanced datasets and immature evaluation practices, not model capability alone, currently set the ceiling on the applicability of such potentials to biomolecular simulation.

What carries the argument

The central instrument is the evaluation protocol itself: a systematic comparison of the same equivariant message-passing architecture family, trained on the SPICE-v2 dataset, across three axes—model size, training-data composition, and inclusion of explicit long-range dispersion and electrostatics. These models are scored not only on in- and out-of-distribution benchmark datasets but also on observables derived from molecular simulations of bulk liquid water, aqueous NaCl, alanine tripeptide, Trp-cage, and Crambin, which allows the paper to separate what improves benchmark scores from what improves physically meaningful simulated behavior.

What would settle it

Measuring a simulation-derived observable (for instance, the oxygen–oxygen radial distribution function or density of liquid water) across model sizes on a broader set of systems and conditions; if larger models consistently improve these observables relative to experiment across systems, the paper's claim that benchmark gains do not extend to simulations would be falsified. Likewise, a system where explicitly adding long-range electrostatics consistently and markedly improves simulation accuracy across a diverse panel would undermine the 'no systematic impact' conclusion.

Watch

Extended reading notes

Core claim

The paper's central claim is that the transferability of universal machine-learned potentials to biomolecular simulation is currently limited less by architecture or model size than by training data and evaluation methodology. Specifically, it claims that while scaling up equivariant message-passing models trained on SPICE-v2 reduces errors on in- and out-of-distribution benchmark datasets, these gains do not consistently carry over to properties extracted from actual molecular simulations. It further claims that changing the composition of the training dataset shifts predicted simulation properties, and that adding explicit long-range terms for dispersion and electrostatics yields no system

Load-bearing premise

The load-bearing premise is that the chosen systems—liquid water, aqueous NaCl, alanine tripeptide, Trp-cage, and Crambin—are representative enough of biomolecular simulation that the observed trends, including the absence of a systematic long-range electrostatics effect, generalize to the broader class of biomolecular systems.

Editorial extensions

If this is right

  • Larger universal machine-learned potentials should not be assumed to produce better biomolecular simulation observables merely because their benchmark errors shrink.
  • Training-data composition is a primary lever on simulation-derived predictions, so dataset balancing has to be treated as part of the modeling problem.
  • Explicit long-range electrostatics are not a universal remedy for universal potentials; their value is system-dependent and may show up as enhanced conformational diversity rather than lower error.
  • Benchmark-only evaluation of universal ML potentials can be misleading; simulation-based observables are needed to judge practical applicability to biomolecules.

Reading between the lines

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

  • If dataset imbalance is the bottleneck, reweighting or augmenting the training set toward biomolecular conformations may be a more cost-effective route to transferability than scaling the model.
  • The null systematic effect of explicit long-range electrostatics could reflect that the underlying short-range model absorbs some electrostatic influence, or that the training protocol gives long-range terms little room to matter; a follow-up measuring per-interaction energy decomposition would separate these explanations.
  • Increased conformational variability in Trp-cage with long-range electrostatics hints that such terms may release degrees of freedom rather than merely correct energies, which could matter for simulating folding or binding events.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 0 minor

Summary. The abstract of arXiv:2508.10841 claims a systematic evaluation of equivariant message-passing machine-learned potentials trained on SPICE-v2, with and without explicit long-range dispersion/electrostatics, tested on benchmark datasets and molecular simulations of bulk water, aqueous NaCl, alanine tripeptide, Trp-cage, and Crambin. The reported trends include that larger models improve benchmark accuracy but not consistently simulation properties, and that long-range electrostatics show no systematic impact. The concluding claim is that imbalanced datasets and immature evaluation practices limit the applicability of universal machine-learned potentials to biomolecular simulations. However, the supplied full text is an entirely different manuscript, 'Unsupervised Stereo via Multi-Baseline Geometry-Consistent Self-Training' (arXiv:2508.10838v2, cs.CV). The body contains no methods, data, tables, or results concerning machine-learned potentials, SPICE-v2, long-range interactions, or biomolecular systems. None of the abstract's claims can be verified from the submitted material.

Significance. If the reported evaluation were actually present and correct, the paper could be a timely and useful contribution to the machine-learned potential community, because transferring accuracy from benchmark datasets to simulation observables is a central open problem. A careful comparison of model size and explicit long-range terms across several physically diverse systems would be valuable. However, the submitted manuscript provides no evidence for these claims. There are no methods, simulation protocols, error bars, benchmark tables, or per-system results to assess. The potential significance is therefore entirely unsubstantiated by the submitted text, and no credit can be given for reproducible methodology, because none is present.

major comments (3)
  1. [Full text (title, abstract, and all sections)] The submitted manuscript body is arXiv:2508.10838v2, a stereo-matching paper (title: 'Unsupervised Stereo via Multi-Baseline Geometry-Consistent Self-Training'; footer on p.1: 'arXiv:2508.10838v2 [cs.CV] 5 Jan 2026'). It contains no methods, tables, or results concerning machine-learned potentials, SPICE-v2, long-range electrostatics/dispersion, or biomolecular simulations. Every claim in the provided abstract—e.g., 'larger models improve accuracy on benchmark datasets,' 'long-range electrostatics show no systematic impact across systems'—is therefore unsupported by the submitted full text.
  2. [Abstract, final sentence] The central conclusion that 'imbalanced datasets and immature evaluation practices currently challenge the applicability of universal machine-learned potentials to biomolecular simulations' purportedly follows from benchmark and simulation measurements on bulk water, NaCl, alanine tripeptide, Trp-cage, and Crambin. None of these measurements, the model configurations, the training/test splits, or the simulation protocols appear anywhere in the manuscript. The conclusion is thus not verifiable from the submitted material.
  3. [Whole manuscript (absence of evaluable content)] Because the body is a different paper, standard review checks cannot be performed: reproducibility of the MD simulations, definitions and statistical uncertainties of reported properties, composition analysis of the SPICE-v2 training subsets, and the claimed trend of model size versus simulation performance. This is a load-bearing deficiency, not a local presentation issue; the correct manuscript must be submitted for review.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found; the submitted full text is an unrelated stereo-matching manuscript, so the abstract's empirical claims are unverified but not circular.

full rationale

The claimed contribution (abstract of arXiv:2508.10841) is an evaluation of universal machine-learned potentials on biomolecular simulations; its conclusions are empirical generalizations from benchmark measurements and molecular dynamics simulations. Nothing in the abstract defines the outcome in terms of the input, fits a parameter and renames it a prediction, or imports a uniqueness theorem from prior work by the same authors. The submitted full text, however, is entirely a different manuscript, 'Unsupervised Stereo via Multi-Baseline Geometry-Consistent Self-Training' (arXiv:2508.10838v2 [cs.CV], 5 Jan 2026), with its own abstract, experiments on KITTI/Middlebury/ETH3D/DrivingStereo, and a CARLA-synthesized dataset. Consequently, the evaluation apparatus described in the abstract—SPICE-v2-trained equivariant message-passing models, explicit long-range electrostatics/dispersion, bulk water, NaCl, alanine tripeptide, Trp-cage, Crambin—is absent from the body, so the abstract's quantitative claims cannot be checked. That is a completeness/evidence problem, not circularity: no equation in the supplied text reduces a prediction to its input by construction, and no self-citation chain is load-bearing. The one limitation statement present in the supplied text ('its current fine-tuning relies on augmented KITTI data, where the generated images may contain unrealistic regions that limit learning') concerns the stereo framework and is not part of the claimed physics-chemistry derivation. Honest finding: no significant circularity, score 0.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

Only abstract-level assumptions can be listed. All quantitative details (cutoffs, long-range damping schemes, dataset splits, simulation lengths, error analysis) live in the missing body text, which is a different paper.

assumptions (2)
  • domain assumption SPICE-v2-trained models adequately represent 'universal' potential accuracy across the tested biomolecular systems.
    The abstract's transferability conclusions presuppose that the SPICE-v2 training set is representative of biomolecular chemical space; no supporting evidence is visible in the abstract.
  • domain assumption The evaluated simulation observables (liquid water properties, NaCl solvation, peptide, Trp-cage, and Crambin conformational behavior) are adequate proxies for biomolecular simulation fidelity.
    The abstract's conclusion that larger models do not consistently extend to simulated properties depends on the chosen systems and metrics, which are named but not justified in the abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Performance of universal machine-learned potentials with explicit long-range interactions in biomolecular simulations." pith.science (2026). https://pith.science/paper/KWN2YDTB

@misc{pith2026250810841,
  author       = {Pith},
  title        = {Pith review of: Performance of universal machine-learned potentials with explicit long-range interactions in biomolecular simulations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KWN2YDTB}},
  note         = {Machine review of arXiv:2508.10841}
}
read the original abstract

Universal machine-learned potentials promise transferable accuracy across compositional and vibrational degrees of freedom, yet their application to biomolecular simulations remains underexplored. This work systematically evaluates equivariant message-passing architectures trained on the SPICE-v2 dataset with and without explicit long-range dispersion and electrostatics. We assess the impact of model size, training data composition, and electrostatic treatment across in- and out-of-distribution benchmark datasets, as well as molecular simulations of bulk liquid water, aqueous NaCl solutions, and biomolecules, including alanine tripeptide, the mini-protein Trp-cage, and Crambin. While larger models improve accuracy on benchmark datasets, this trend does not consistently extend to properties obtained from simulations. Predicted properties also depend on the composition of the training dataset. Long-range electrostatics show no systematic impact across systems. However, for Trp-cage, their inclusion yields increased conformational variability. Our results suggest that imbalanced datasets and immature evaluation practices currently challenge the applicability of universal machine-learned potentials to biomolecular simulations.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

52 extracted references · 40 canonical work pages

  1. [1]

    Reversing the cycle: self-supervised deep stereo through enhanced monocular distillation

    Filippo Aleotti, Fabio Tosi, Li Zhang, Matteo Poggi, and Ste- fano Mattoccia. Reversing the cycle: self-supervised deep stereo through enhanced monocular distillation. InEuropean Conference on Computer Vision, pages 614–632. Springer,

    work page

  2. [2]

    Emerg- ing properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Herv ´e J´egou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerg- ing properties in self-supervised vision transformers. InPro- ceedings of the IEEE/CVF international conference on com- puter vision, pages 9650–9660, 2021. 3

    2021

  3. [3]

    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,

    work page

  4. [4]

    Monster: Marry monodepth to stereo unleashes power.arXiv preprint arXiv:2501.08643,

    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

  5. [5]

    Carla: An open urban driv- ing simulator

    Alexey Dosovitskiy, German Ros, Felipe Codevilla, Anto- nio Lopez, and Vladlen Koltun. Carla: An open urban driv- ing simulator. InConference on robot learning, pages 1–16. PMLR, 2017. 2, 5, 1

    work page 2017

  6. [6]

    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. In2012 IEEE conference on computer vision and pat- tern recognition, pages 3354–3361. IEEE, 2012. 1, 2, 5, 6, 7, 3

    work page 2012

  7. [7]

    Vision meets robotics: The kitti dataset.Interna- tional Journal of Robotics Research (IJRR), 2013

    Andreas Geiger, Philip Lenz, Christoph Stiller, and Raquel Urtasun. Vision meets robotics: The kitti dataset.Interna- tional Journal of Robotics Research (IJRR), 2013. 5

    work page 2013

  8. [8]

    Digging into self-supervised monocular depth estimation

    Cl ´ement Godard, Oisin Mac Aodha, Michael Firman, and Gabriel J Brostow. Digging into self-supervised monocular depth estimation. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 3828–3838,

Show all 52 references
  1. [9]

    Bootstrap your own latent-a new approach to self-supervised learning.Advances in neural information processing systems, 33:21271–21284, 2020

    Jean-Bastien Grill, Florian Strub, Florent Altch ´e, Corentin Tallec, Pierre Richemond, Elena Buchatskaya, Carl Doersch, Bernardo Avila Pires, Zhaohan Guo, Mohammad Ghesh- laghi Azar, et al. Bootstrap your own latent-a new approach to self-supervised learning.Advances in neura...

    2020

  2. [10]

    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/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 3273–3282, 2019. 2, 5, 7, 8, 3

    2019

  3. [11]

    Defom-stereo: Depth foundation model based stereo matching.arXiv preprint arXiv:2501.09466, 2025

    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. 2, 7, 3

    2025 arXiv

  4. [12]

    Emr-msf: Self- supervised recurrent monocular scene flow exploiting ego- motion rigidity

    Zijie Jiang and Masatoshi Okutomi. Emr-msf: Self- supervised recurrent monocular scene flow exploiting ego- motion rigidity. InProceedings of the IEEE/CVF Interna- tional Conference on Computer Vision, pages 69–78, 2023. 6, 3

    2023

  5. [13]

    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. InProceedings of the IEEE international confer- ence on computer vision, pages 66–75, 2017. 2, 6

    2017

  6. [14]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization, 2017. 5

    2017

  7. [15]

    Occlusion aware stereo matching via cooperative unsupervised learning

    Ang Li and Zejian Yuan. Occlusion aware stereo matching via cooperative unsupervised learning. InAsian Conference on Computer Vision, pages 197–213. Springer, 2018. 2, 6, 3

    2018

  8. [16]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Re...

    2022

  9. [17]

    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. 2, 5, 7, 8, 3

    2021

  10. [18]

    Flow2stereo: Effective self-supervised learning of optical flow and stereo matching

    Pengpeng Liu, Irwin King, Michael R Lyu, and Jia Xu. Flow2stereo: Effective self-supervised learning of optical flow and stereo matching. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6648–6657, 2020. 6, 3

    2020

  11. [19]

    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. InProceedings of the IEEE conference on computer vision and p...

    2016

  12. [20]

    Object scene flow for au- tonomous vehicles

    Moritz Menze and Andreas Geiger. Object scene flow for au- tonomous vehicles. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 3061– 3070, 2015. 1, 2, 5, 6, 7, 8, 3, 4

    2015

  13. [21]

    Zoom and learn: Generalizing deep stereo matching to novel domains

    Jiahao Pang, Wenxiu Sun, Chengxi Yang, Jimmy Ren, Ruichao Xiao, Jin Zeng, and Liang Lin. Zoom and learn: Generalizing deep stereo matching to novel domains. InPro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 2070–2079, 2018. 2, 3, 6

  14. [22]

    A taxonomy and evaluation of dense two-frame stereo correspondence algo- rithms.International journal of computer vision, 47(1):7– 42, 2002

    Daniel Scharstein and Richard Szeliski. A taxonomy and evaluation of dense two-frame stereo correspondence algo- rithms.International journal of computer vision, 47(1):7– 42, 2002. 2

    2002

  15. [23]

    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. InGerman conference on pattern recognition, pages 31–42. Springer, 2014. 5, 7, 3

    2014

  16. [24]

    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. InProceedings of the IEEE Conference on Computer Vision and Pat...

    2017

  17. [25]

    Sgm-nets: Semi-global matching with neural networks

    Akihito Seki and Marc Pollefeys. Sgm-nets: Semi-global matching with neural networks. InProceedings of the 9 IEEE conference on computer vision and pattern recogni- tion, pages 231–240, 2017. 6

    2017

  18. [26]

    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. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13906–13915, 2021. 2, 5, 7, 8, 3

    2021

  19. [27]

    Z. Shen, Y . Dai, et al. Pcw-net: Pyramid combination and warping cost volume for stereo matching. InECCV, 2022. 2

    2022

  20. [28]

    Leslie N. Smith. A disciplined approach to neural network hyper-parameters: Part 1 – learning rate, batch size, momen- tum, and weight decay, 2018. 5

    2018

  21. [29]

    Adastereo: a simple and efficient ap- proach for adaptive stereo matching

    Xiao Song, Guorun Yang, Xinge Zhu, Hui Zhou, Zhe Wang, and Jianping Shi. Adastereo: a simple and efficient ap- proach for adaptive stereo matching. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10328–10337, 2021. 6, 3

    2021

  22. [30]

    Real-time self-adaptive deep stereo

    Alessio Tonioni, Fabio Tosi, Matteo Poggi, Stefano Mat- toccia, and Luigi Di Stefano. Real-time self-adaptive deep stereo. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 195–204, 2019. 6, 3

    2019

  23. [31]

    Nerf-supervised deep stereo

    Fabio Tosi, Alessio Tonioni, Daniele De Gregorio, and Mat- teo Poggi. Nerf-supervised deep stereo. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 855–866, 2023. 3, 7

    2023

  24. [32]

    Weakly supervised learning of deep metrics for stereo recon- struction

    Stepan Tulyakov, Anton Ivanov, and Francois Fleuret. Weakly supervised learning of deep metrics for stereo recon- struction. InProceedings of the IEEE International Confer- ence on Computer Vision, pages 1339–1348, 2017. 6, 3

    2017

  25. [33]

    Faster self-adaptive deep stereo

    Haiyang Wang, Xinchao Wang, Jie Song, Jie Lei, and Mingli Song. Faster self-adaptive deep stereo. InProceedings of the Asian Conference on Computer Vision, 2020. 6

    2020

  26. [34]

    Parallax attention for unsupervised stereo correspondence learning

    Longguang Wang, Yulan Guo, Yingqian Wang, Zhengfa Liang, Zaiping Lin, Jungang Yang, and Wei An. Parallax attention for unsupervised stereo correspondence learning. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 44(4):2108–2125, 2022. 2, 6, 3

    2022

  27. [35]

    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. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 19701–19710, 2024. 2, 7, 8, 3

    2024

  28. [36]

    Zerostereo: Zero-shot stereo matching from single images

    Xianqi Wang, Hao Yang, Gangwei Xu, Junda Cheng, Min Lin, Yong Deng, Jinliang Zang, Yurui Chen, and Xin Yang. Zerostereo: Zero-shot stereo matching from single images. arXiv preprint arXiv:2501.08654, 2025. 7, 1, 3

    2025 arXiv

  29. [37]

    Dualnet: Ro- bust self-supervised stereo matching with pseudo-label su- pervision

    Yun Wang, Jiahao Zheng, Chenghao Zhang, Zhanjie Zhang, Kunhong Li, Yongjian Zhang, and Junjie Hu. Dualnet: Ro- bust self-supervised stereo matching with pseudo-label su- pervision. InProceedings of the AAAI Conference on Artifi- cial Intelligence, pages 8178–8186, 2025. 2, 3, 6

    2025

  30. [38]

    Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004. 3

    2004

  31. [39]

    Foundationstereo: Zero- shot stereo matching.arXiv preprint arXiv:2501.09898,

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

    arXiv

  32. [40]

    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, 5, 7, 8, 3

    2023

  33. [41]

    Aanet: Adaptive aggrega- tion network for efficient stereo matching

    Haofei Xu and Juyong Zhang. Aanet: Adaptive aggrega- tion network for efficient stereo matching. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1959–1968, 2020. 6

    1959

  34. [42]

    Drivingstereo: A large-scale dataset for stereo matching in autonomous driving scenarios

    Guorun Yang, Xiao Song, Chaoqin Huang, Zhidong Deng, Jianping Shi, and Bolei Zhou. Drivingstereo: A large-scale dataset for stereo matching in autonomous driving scenarios. InProceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 899–908, 2019. 5, 7

    2019

  35. [43]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10371–10381, 2024. 3

    2024

  36. [44]

    Depth any- thing v2.Advances in Neural Information Processing Sys- tems, 37:21875–21911, 2024

    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. 2

    2024

  37. [45]

    Unsupervised hierarchical iterative tile refinement network with 3d planar segmentation loss.IEEE Robotics and Au- tomation Letters, 9(3):2678–2685, 2024

    Ruizhi Yang, Xingqiang Li, Rigang Cong, and Jinsong Du. Unsupervised hierarchical iterative tile refinement network with 3d planar segmentation loss.IEEE Robotics and Au- tomation Letters, 9(3):2678–2685, 2024. 3, 6

    2024

  38. [46]

    Semi- stereo: A universal stereo matching framework for imper- fect data via semi-supervised learning

    Xin Yue, Zongqing Lu, Xiangru Lin, Wenjia Ren, Zhijing Shao, Haonan Hu, Yu Zhang, and Qingmin Liao. Semi- stereo: A universal stereo matching framework for imper- fect data via semi-supervised learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Re...

    2024

  39. [47]

    Eai-stereo: Error aware iterative network for stereo matching

    Haoliang Zhao, Huizhou Zhou, Yongjun Zhang, Yong Zhao, Yitong Yang, and Ting Ouyang. Eai-stereo: Error aware iterative network for stereo matching. InProceedings of the Asian Conference on Computer Vision, pages 315–332,

  40. [48]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1327– 1336, 2023. 2

    2023

  41. [49]

    Self- supervised learning for stereo matching with self-improving ability.arXiv preprint arXiv:1709.00930, 2017

    Yiran Zhong, Yuchao Dai, and Hongdong Li. Self- supervised learning for stereo matching with self-improving ability.arXiv preprint arXiv:1709.00930, 2017. 1, 2, 3, 5, 6

    2017 arXiv

  42. [50]

    Consistency-aware self-training for iterative-based stereo matching

    Jingyi Zhou, Peng Ye, Haoyu Zhang, Jiakang Yuan, Rao Qiang, Liu YangChenXu, Wu Cailin, Feng Xu, and Tao Chen. Consistency-aware self-training for iterative-based stereo matching. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 16641–16650, 2025. ...

    2025

  43. [51]

    Randomly select one image as the reference view

  44. [52]

    #"$"%&"'()*+

    Independently sample two target views (with replace- ment) from the remaining(N−1)images to serve as the student and teacher target inputs. This sampling protocol yields a total ofN(N−1) 2 dis- tinct triplets per timestamp. Under our configuration with N= 5cameras, this produc...

    2015

Pith tools

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