Abstract

Quantitative eye movement analysis is important for neurological diagnostics, yet existing video-oculography (VOG) systems typically require calibration, device-specific settings, or accurate gaze labels. We present VOGeo-Gaze, a real-time, calibration-free, geometry-aware neural network that estimates gaze by reconstructing anatomically meaningful eyeball parameters from image features. Trained only on the public TEyeD dataset with weak gaze supervision, the method was evaluated on 116 clinical recordings from 17 patients and 19 healthy subjects using EyeSeeCam, a clinical gold-standard VOG system. VOGeo-Gaze achieved median absolute angular errors of 0.33° horizontally and 0.35° vertically, with nearly 92% of recordings below 1° error while operating at >300 FPS. These results demonstrate sub-degree clinical gaze estimation without subject-specific calibration, camera intrinsics, accurate gaze labels providing a scalable and interpretable alternative to conventional VOG pipelines. Code is available at https://github.com/DSGZ-MotionLab/VOGeo-Gaze

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3731_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: https://papers.miccai.org/miccai-2026/supp/3731_supp.zip

Link to the Code Repository

https://github.com/DSGZ-MotionLab/VOGeo-Gaze

Link to the Dataset(s)

N/A

BibTex

@InProceedings{ZhaJin_VOGeoGaze_MICCAI2026,
        author = { Zhao, Jingkang AND Ahmadi, Seyed-Ahmad AND Decker, Julian AND zu Eulenburg, Peter AND Zwergal, Andreas AND Flanagin, Virginia L. AND Wuehr, Max},
        title = { { VOGeo-Gaze: Calibration-Free, Geometry-Aware Deep Learning for Real-Time Gaze Tracking in Clinical Video-Oculography } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16896},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The novelty in this work lies in a novel method for estimating eye gaze from VOG video, without the requirement for patient-specific calibration or the provision of camera properties. This is achieved through an end-to-end model that minimises a custom loss function during training. The model learns eyeball parameters, corrects for refraction and provides temporal stabilisation. The estimation of refraction correction and temporal stabilisation are shown to be significant factors in the competitive level of accuracy obtained by the system.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    The approach developed achieves comparable results to the best-performing benchmark system (3DeepVOG), but without the requirement for patient calibration or the provision of additional parameters, such as camera properties. The level of accuracy ultimately obtained for gaze estimation is < 1 degree for horizontal and vertical gaze. The system developed is shown to be robust to adverse conditions, such as low-lighting. The work presents a novel approach involving the estimation of eyeball parameters rather than the estimation of gaze directly. The system is shown to be computationally efficient enough to allow for realtime processing of VOG data, which is an additional advantage over some poorer performing techniques and suggests that this system could have clinical applicability.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    The approach is shown to work not quite as well as the best-performing system tested (3DeepVOG).

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission has provided an anonymized link to the source code, dataset, or any other dependencies.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (6) Strong Accept — must be accepted due to excellence

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The paper presents the methods developed very thoroughly and clearly, and the system developed is evaluated comprehensively. The system is shown to work very well and appears to overcome a severe limitation of conventional systems, i.e. patient-specific calibration. In my opinion, this work is novel, technically interesting, practically useful and it is well described and presented.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Review #2

  • Please describe the contribution of the paper

    The authors propose VOGeo-Gaze, a calibration-free geometry-constrained framework. The contributions of this paper as far as I can tell:

    1) awareness of corneal refraction, 2) sub-degree clinical gaze, 3) model based recovery of gaze

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    Well written paper with highly promising results.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    I did not have any major issues with this paper.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    What is the advantage of a differentiable reformulation ?

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (6) Strong Accept — must be accepted due to excellence

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Very good and convincing presentation Good results comparable to gold standard

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Review #3

  • Please describe the contribution of the paper

    This work reformulates VOG gaze estimation from angular regression to anatomical parameter reconstruction, enabling physically interpretable optimization via a differentiable refractive eye model.

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    Compared with existing methods, VOGeo-Gaze requires no calibration, no camera intrinsics, and no precise gaze labels, replacing direct angle regression with geometry-aware anatomical parameter reconstruction.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    1.Quantitative and qualitative comparisons with state-of-the-art methods are insufficiently comprehensive. The current evaluation only includes a limited set of baselines on a single clinical dataset, lacking comparisons with recent calibration-free gaze estimation, geometry-aware eye modeling, and real-time VOG tracking approaches published in the past 2–3 years. Meanwhile, qualitative results only demonstrate one failure case of calibration-based methods, without systematic visualization on diverse clinical scenarios (e.g., occlusion, low-amplitude eye movement, headset misalignment), making it difficult to fully verify the robustness and generalization of the proposed method.

    2.The justification of novelty and contribution lacks clear reasoning grounded in prior work and experimental evidence. The authors do not explicitly highlight the critical limitations of existing clinical VOG pipelines and learning-based gaze estimators. In addition, the connection between the experimental results and the claimed advantages (e.g., calibration-free, geometry-aware, interpretable) is not sufficiently elaborated.

  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    This assessment is mainly based on prior studies, the method proposed in this work, and its experimental results.

  • Reviewer confidence

    Somewhat confident (2)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Review #4

  • Please describe the contribution of the paper

    1.The paper presents VOGeo-Gaze, a calibration-free, geometry-aware neural network for gaze estimation in clinical video-oculography (VOG). The framework integrates a corneal refraction corrector (Fref), a prior gaze estimator (Fdir), and a temporal stabilization module (Ftem). 2.The core idea is to shift optimization from angular gaze space to anatomical parameter space: instead of directly regressing gaze angles (θH, θV), the model reconstructs latent eyeball parameters (eyeball center ceye, optic axis n, pupil radius r_pup) via a differentiable, refraction-aware reformulation of the two-sphere eye model. 3.Supervision relies on segmentation labels and weak directional constraints derived from the TEyeD dataset, requiring neither subject-specific calibration, camera intrinsics, nor accurate gaze annotations. 4.Clinical evaluation on 116 EyeSeeCam recordings from 17 patients and 19 healthy subjects reports 0.33°/0.35° median absolute angular error, with >91% of recordings below 1° error, at real-time throughput (>300 FPS).

  • Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    1.Novel paradigm with physically grounded design: By explicitly modeling corneal refraction in a differentiable manner, VOGeo-Gaze addresses a known limitation of prior geometry-free and refraction-neglecting methods. The three-contribution structure is clearly articulated. 2.Clinically meaningful and rigorous evaluation: The evaluation is conducted on real EyeSeeCam recordings spanning multiple oculomotor protocols across both patient and healthy subject populations. The use of median MAE with interquartile ranges, together with the <1° threshold percentage, provides a clinically interpretable performance characterization. The Wilcoxon paired test for geometric consistency adds statistical rigor. 3.Comprehensive ablation study: Table 4 systematically isolates the contribution of each module. The large performance gap when removing Fref provides strong evidence that explicit refraction modeling is the critical driver of accuracy. 4.Efficiency and real-time capability: At 2.3M parameters and 9.2 GFLOPs per forward pass, VOGeo-Gaze is computationally competitive with 3DeepVOG while achieving >300 FPS. This is practically relevant for clinical deployment.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    1.Fixed anatomical priors may limit clinical generalizability: The method relies on population-average anatomical constants. In a neuro-otology clinic, patients may present with conditions affecting ocular anatomy. The authors do not analyze sensitivity to anatomical prior mismatch or report per-pathology performance. 2.Missing relevant baselines: The paper cites De²Gaze and CVL but only includes CVL in Table 3 and does not include De²Gaze in the comparison. De²Gaze is a geometry-aware learning method and would represent a more directly comparable Type B baseline. The authors should either include De²Gaze in the comparison or provide an explicit justification for its exclusion. 3.The “device-independent” claim is not experimentally validated: The abstract and introduction claim device independence, but the model is trained on TEyeD and evaluated exclusively on EyeSeeCam. No cross-device experiment is presented. 4.Architecture description and Figure 2 readability: The architectural overview is extremely dense and contains overlapping notation (e.g., Epup, Ê_pup, E^pup_R, E^pup_PD appearing in close proximity with subtly different superscripts). A cleaner schematic and more self-contained method description would significantly improve readability and reproducibility.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    N/A

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.

    (4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    VOGeo-Gaze addresses a genuine clinical need, such as calibration-free, interpretable gaze estimation, with a technically sound and well-evaluated approach. The core idea of anatomical parameter space reconstruction with explicit corneal refraction is novel and the clinical results are compelling. The main factors preventing a stronger recommendation are: (1) the missing De²Gaze baseline; and (2) the device-independence claim lacking empirical support.

  • Reviewer confidence

    Somewhat confident (2)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Author Feedback

N/A




Meta-Review

Meta-review #1

  • Your recommendation

    Provisional Accept

  • Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.

    This paper provides an eye gaze tracking algorithm that exploits geometric constraints grounded on eye anatomy to improve accuracy without relying on camera calibration. The method’s purpose is to assist in video oculography measurements. The method is tested on an in-house dataset, showing the proposed algorithm achieves similar accuracy to prior work, with the advantage of not requiring calibration.

    The paper received a consensus of “accept/weak accept” reviews. There is broad agreement on the novelty of the approach and that the reported results are meaningful and relevant.

    Some reviewers would like to see the method compared against more baselines, although it is recognised that utilised baseline 3DeepVOG is also recognised as good representative of the state-of-the-art. I believe that the current experiments are justifiable in the context of the short MICCAI format and provide enough evidence that this is a strong enough paper for MICCAI.



back to top