Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Craniomaxillofacial deformities often necessitate orthognathic surgery to correct jaw positions and improve both function and aesthetics. The existing patient-specific optimal face prediction for soft-tissue-driven planning struggles to accurately capture fine facial details and maintain harmonious alignment among key facial features. In this paper, we propose a novel Conditional Autoregressive Modeling for Orthognathic Surgery (CAMOS) framework that directly predicts patients’ optimal 3D face from their preoperative appearance. Our approach employs a hierarchical, coarse-to-fine next-scale prediction strategy, beginning with large-scale pretraining on 44,602 control faces to construct a robust generative model that captures diverse demographic features. Subsequently, the model is fine-tuned on an in-house dataset of 86 orthognathic surgery patients, establishing a conditional path that integrates patient-specific information to form a conditional generative model. Evaluation on both public and in-house datasets demonstrates that CAMOS successfully generates patient-specific optimal face with high quality, effectively addressing the limitations of prior single-step approaches. Source code is available at https://github.com/RPIDIAL/CAMOS.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1908_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/RPIDIAL/CAMOS

Link to the Dataset(s)

N/A

BibTex

@InProceedings{LeeJun_Facial_MICCAI2025,
        author = { Lee, Jungwook and Xu, Xuanang and Kim, Daeseung and Kuang, Tianshu and Deng, Hannah H. and Song, Xinrui and Soubra, Yasmine and Dharia, Rohan and Liebschner, Michael A.K. and Gateno, Jaime and Yan, Pingkun},
        title = { { Facial Appearance Prediction with Conditional Multi-scale Autoregressive Modeling for Orthognathic Surgical Planning } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15969},
        month = {September},
        page = {212 -- 222}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    The main contributions of the paper are the curation of a large dataset and its use for the training and application of a Visual AutoRegressive model for the prediction of a normal-looking face based on landmarks extracted from a preoperative deformed face.

  • 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.

    • Convincing results: especially the qualitative comparison with post-op cases is impressive, but improvements in terms of Chamfer Distance are consistent, too.

    • Production and use of in-house dataset

  • 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.

    • unclear clinical application: Does the proposed soft-tissue driven approach also provide a bone plan that can be used in surgery, or does it only provide a visualization of an optimal face? Without a bone-plan, how exactly would the proposed approach be used in surgery-planning?

    • unclear with regards to architecture novelty: Are any parts of the architecture shown in Fig. 2 novel or was this architecture as a whole adopted from reference 18? Can the authors please clarify this point?

    • problematic methodology? Table 3 shows a hyper-parameter optimization for code-book length and scale. Typically, this should not be performed on the testing set to maintain the independence of testing data, but rather on a separate validation set. However, Table 2 shows exactly the same results, which suggests that they were obtained on the same set. Can the authors please clarify this point?

  • 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.

  • 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/2025/en/REVIEWER-GUIDELINES.html

    Fig 1: This is a nice figure, but the legend looks like part of the architecture at first glance. I suggest simply including the legend text in the architecture and removing the legend itself.

    Chapter 5.1, line 4: there is a formatting error that places the comma at the beginning of the line

  • 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.

    (3) Weak Reject — could be rejected, dependent on rebuttal

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

    I recommend that the authors address the mentioned weaknesses with regards to the clinical application and the potential hyperparameter tuning on the testing set.

  • Reviewer confidence

    Somewhat confident (2)

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

    Accept

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

    The authors addressed all issues appropriately.



Review #2

  • Please describe the contribution of the paper

    Within the scope of orthognathic surgical planning, the authors present a method for predicting a patient’s ideal, computer-generated 3D facial surface mesh to assist surgeons in repositioning the jaw, with the dual goal of improving both function and overall facial aesthetics. The proposed framework, CAMOS, takes a preoperative, deformed 3D facial surface mesh as input and uses a generative model to predict an optimal 3D facial structure that represents the desired surgical outcome. According to the authors, unlike most prior work that only considers limited facial regions or landmarks, CAMOS captures the full 3D facial geometry, allowing for the generation of more realistic and holistic results. In particular, it enables a harmonious alignment of key facial features such as the lips, jaw, and overall facial symmetry. A notable innovation of CAMOS is its hierarchical prediction strategy, in contrast to most existing methods that rely on a single-step face prediction. The method is evaluated experimentally using a large-scale dataset of normal subjects combined with in-house patient data. CAMOS outperforms three existing approaches in predicting several key facial features. An ablation study further analyzes the influence of critical hyperparameters on the multi-scale facial mesh prediction process, with a focus on individual facial components.

  • 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.

    Overall, the paper is well thought out, with a strong surgical motivation and a compelling illustration of the potential benefits for patients. Below are some of the major strengths of the paper: 1.) A detailed technical description of the proposed CAMOS framework, including several implementation details and two well-designed figures that illustrate the end-to-end workflow of the generative model.

    2.) A thorough quantitative evaluation of the CAMOS model against several state-of-the-art methods, complemented by a qualitative comparison of predicted face meshes with actual postoperative outcomes. The inclusion of an ablation study, which provides further insight into the model’s design choices and sensitivity to key components.

    3.) An intuitive presentation of the paper’s main contribution: a hierarchical, multi-scale 3D face mesh prediction approach, which is clearly motivated and visually supported in the paper.

  • 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.) Page 2: The authors state that existing methods can yield unrealistic results when predicting fine facial details. However, the paper does not elaborate on what is meant by “realistic” vs. “unrealistic” in this context. It would be helpful if the authors could define or quantify this distinction. For example, were these differences perceptually validated, or do they refer to geometric inconsistencies?

    2.) Experiments: It remains unclear whether clinical experts (e.g., surgeons) were involved in validating whether the predicted results are indeed more realistic compared to existing methods. I would encourage the authors to clarify whether expert feedback was collected and, if not, to discuss this as a potential limitation.

    3.) Page 2: The authors write: “When humans perceive or imagine a face, we tend to first grasp its overall structure and then refine the finer details.” This is an interesting claim, but a reference to relevant psychological or neuroscientific literature would strengthen it. I encourage the authors to support this point with a suitable citation.

    4.) Dataset contribution: The authors claim that one of the key contributions of this work lies in the curation of a large-scale dataset. However, Section 2 (“Datasets and Data Processing”) indicates that the model was primarily trained on over 44,000 facial surface meshes drawn from four publicly available datasets. In addition, 86 patient cases were collected in-house. It is unclear:

    • How many surface meshes were derived from these 86 patients
    • How these meshes relate quantitatively to the 44,000 public samples
    • Whether any of the patient data will be made publicly available

    I recommend that the authors clarify these points. If the patient data will not be released, they should provide a rationale (e.g., ethical or legal constraints). It may also be appropriate to mention whether ethics approval or informed consent was obtained for the use of patient data.

    5.) Clinical evaluation: The method is not evaluated in a clinical setting, and no surgical experts appear to have assessed its usefulness in practice. Including even a limited expert review or pilot study would strengthen the work and help demonstrate clinical utility.

    6.) Conclusion section: The “Conclusions” section is quite brief and does not discuss the limitations of the proposed method or potential directions for future work. Expanding this section would improve the completeness and critical reflection of the 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.

  • 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/2025/en/REVIEWER-GUIDELINES.html

    I would recommend including a statement on the availability of your code and dataset within the paper. This would enhance transparency and reproducibility of your work.

  • 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 — could be accepted, dependent on rebuttal

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

    Overall, I believe this paper falls between a “weak accept” and a “weak reject.” On one hand, the authors present an interesting and potentially valuable contribution that could be of interest to the research community. On the other hand, there are several weaknesses that should be addressed to improve the paper’s clarity, rigor, and transparency.

    However, given the method’s promising clinical relevance and potential for real-world impact, I would recommend leaning toward acceptance, provided that the authors adequately address the identified issues.

  • Reviewer confidence

    Very confident (4)

  • [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

    The manuscript proposes a novel soft-tissue-driven approach for predicting normal facial features in orthognathic surgery. Unlike traditional methods that first correct craniomaxillofacial deformities at the skeletal level and subsequently estimate soft tissue changes, the authors prioritize soft tissue outcomes from the outset. The key contributions include a multi-scale hierarchical prediction strategy inspired by human perception - first capturing the global facial structure and then refining local details - implemented by adapting a Visual AutoRegressive architecture. Additionally, the authors introduce a curated, large-scale dataset of 44,602 normal subjects, designed to represent a broad spectrum of facial appearances and support effective model pretraining.

  • 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.

    Overall, the manuscript is well-written, clear, and provides sufficient contextual background. The methods are thoroughly described, and the experimental results are comprehensive and compelling, effectively supporting the authors’ claims. The core idea - directly predicting deformed (or undeformed) facial appearances without relying on intermediate bone-level modeling - is both innovative and well-suited to a deep learning framework.

  • 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.

    While this soft-tissue-driven approach offers a novel perspective by bypassing traditional bone modeling, it naturally raises important questions about clinical translatability. Two key concerns arise: First, are the predicted facial reconstructions anatomically “realistic” in the sense that they correspond to viable underlying bone configurations? Second, the manuscript does not clearly address how directly predicting soft tissue outcomes informs or assists in planning the necessary skeletal adjustments. In current clinical workflows, surgeons typically modify bone structures and observe the resulting facial changes; however, in this reversed paradigm, it remains unclear how one derives the corresponding pre-operative bone movements. The authors also list the creation of a curated dataset as one of their primary contributions; however, the dataset appears to be largely composed of selected samples from existing sources, with only limited additions from in-house patient data. There is no mention of plans to release the dataset publicly upon acceptance, which may limit the broader impact of the work.

  • 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.

  • 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/2025/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 — could be accepted, dependent on rebuttal

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

    The manuscript presents a novel and technically sound approach that is well-motivated and clearly described, with strong experimental results. However, key concerns remain regarding the clinical translatability of the method - particularly the lack of discussion on how predicted soft tissue outcomes inform surgical planning and bone repositioning. Additionally, the dataset contribution is somewhat overstated, and the absence of plans for public release limits reproducibility. These issues could be addressed in the rebuttal and would strengthen the case for acceptance.

  • 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

    In this work, the authors propose a new method for predicting patients’ optimal 3D face from their preoperative appearance, called CAMOS. This approach combines visual autoregressive modeling—to predict facial landmark tokens from coarse to fine scales—with a vector-quantized variational autoencoder (VQ-VAE) to learn discrete tokens at multiple scales. To address the issue of limited patient data, the authors collected a large-scale dataset consisting only of healthy subjects and pretrained the multi-scale VQ-VAE on it, followed by fine-tuning on the patient group. The model’s generation and face prediction performance are compared with previous methods, showing promising results. Visualizations of the proposed model also appear very encouraging.

  • 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 presentation of the entire paper is highly satisfactory. From the textual descriptions to the plots, tables, and illustrative figures, the authors have done an excellent job of clearly articulating the problem definition, motivation, proposed methods, implementation details, and results. Notably, the work is accessible even to readers without specific expertise in 3D face generation or orthognathic surgery. While this may seem basic, such clarity is crucial. The logic from motivation to solution is coherent and well-structured, which is especially valuable in the medical domain. When introducing new techniques into this field, it is often the case that insufficient attention is paid to the specific needs of the application area. In contrast, this paper grounds its design choices in domain-relevant observations. For example, the motivation that “when humans perceive or imagine a face, we tend to first grasp its overall structure and then refine the finer details” directly informs the hierarchical, multi-scale token design. Similarly, the challenge of acquiring sufficient patient data due to privacy concerns leads to a sensible and domain-specific solution: pretraining on a large dataset of healthy subjects. In terms of novelty, the introduction of multi-scale facial landmark tokens is a fresh and promising idea for this domain. Furthermore, the application of state-of-the-art techniques such as visual autoregressive modeling (VAR) in the medical context is commendable. The experimental evaluation is nearly complete, including an ablation study. Visual comparisons between the proposed method and prior approaches also indicate strong potential.

  • 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.

    There are no major weaknesses in this work, but I have a few questions and concerns that I hope the authors can address in the rebuttal or a future revision of the manuscript:

    1. Assumption in the 1st para in page 2: the statement “In addition, those methods often fail to achieve a harmonious alignment among the lips, jaw, and overall facial structure” raises a question. Is it realistic to assume that orthognathic surgery can consistently achieve such harmonious alignment? I imagine that in many cases, perfect alignment may not be achievable, and thus the notion of an “optimal” face might not necessarily imply such harmony. It would be helpful if the authors could clarify this assumption or nuance the claim to reflect clinical variability.

    2. Loss Term in Fine-Tuning: The loss function for the fine-tuning phase seems missing. I tend to think that you are not using eq 1 for fine-tuning, as the first term eq 1 defines a reconstruction loss, which does not seems to be correct to keep minimizing this part of loss in fine-tuning phase, as the inputs x and reconstructions x_hat should differ (as the patient’s optimal face is not the same as their input), it is unclear whether the same reconstruction objective still applies. Please clarify this.

    3. asterisk in Tables 2 and 3: you mention that “p-values less than 0.05 are marked with an asterisk”, but none of the results for CAMOS in Tables 2 and 3 include such an asterisk. Was this an oversight, or is it indeed the case that CAMOS did not achieve statistical significance in these comparisons? If the latter, an explanation or discussion would be valuable to understand the results better.

    Minor comments (suggestions):

    • When describing the collected dataset, I suggest replacing the term “normal subjects” with “healthy subjects” or “control subjects,” as the word “normal” can inadvertently imply that the comparison group is “abnormal,” which may be considered insensitive or inappropriate in clinical or ethical contexts.
    • Figure1: CAMOS also consists of the MQ and even the encoder and decoder, right?
    • Figure 2 lacks some notation definitions, which are also not clearly explained in the main text. For instance: What do C and K refer to? Does S_k × C denote the shape of the token sequence at the k-th scale? Since S_i(x) was used earlier to denote a token variable at scale i, it’s important to disambiguate between variable notation and dimensional description.
  • 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.

  • 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/2025/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 — could be accepted, dependent on rebuttal

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

    this paper presents a well-motivated and novel approach to 3D face prediction in the context of orthognathic surgery, with clear relevance to clinical applications. The paper is well-written and well-presented, with novel method and intuitive visualization of results, making it accessible even to readers without deep expertise in 3D face generation or medical imaging. however, a few aspects would benefit from clarification or deeper analysis. Please refer to the weakness part for details While none of these points critically undermine the contribution, they collectively warrant a cautious recommendation. Addressing them—either in the rebuttal or in a future revision—would significantly strengthen the paper.

  • Reviewer confidence

    Confident but not absolutely certain (3)

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

    Accept

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

    My recommendation for this paper is acceptance, as it’s clearly presented, logically structured, grounded in domain knowledge, methodologically novel, and supported by solid experiments. It is a great attempt to use advanced techniques in the medical surgery domain with novelty and nice presentation. I believe it could motivate more work in the specific area.



Author Feedback

We thank reviewers for their feedback and recognition of our method’s novelty (R3, R4), convincing experimental results (R1, R3), clear presentation (R2, R4), and effective use of large-scale public datasets (R1, R4). Below, we address key concerns grouped into categories for clarity and brevity.

  • Clinical Application & Impact (R1-4) A complete soft-tissue-driven planning framework involves: (1) Predicting a patient’s optimal face (2) Estimating the corresponding bone movements required to achieve this optimal face (3) Verifying surgical feasibility by re-simulating soft-tissue outcomes based on the predicted bone movements. This manuscript focuses on step 1. Estimating and incorporating bone movement is out of the scope of this paper. In addition, we would like to clarify that “harmonious alignment” or “optimal” refers to the best achievable facial outcome within surgical constraints, rather than absolute perfection. We will clarify this scope in the camera-ready version.

  • Architecture Novelty & Training Clarity (R1, R4) Our architecture differs from VAR [18] in two aspects: (1) CAMOS uses point set down/up sampling strategies suitable for 3D landmarks, unlike VAR’s fixed 2D image interpolation, (2) CAMOS introduces a conditional path for preoperative appearance-driven generation while preserving large-scale pretraining benefits, unlike VAR’s unconditional model. The loss functions are different in the three training steps: initial discrete tokenization of landmarks (Eq.1), unconditional next-token prediction (cross-entropy (CE) loss), and conditional generative fine-tuning (also CE loss). The loss function of VQ-VAE in Eq.1 applies only to the initial tokenization, not subsequent generative steps.

  • Dataset Composition & Availability (R2, R3) Our dataset consists of 44,602 publicly available facial meshes and an additional 86 patient facial meshes (CT + stereophotogrammetry). The public dataset supports generalizability across diverse demographics, whereas the patient data provides paired pre-/post-operative faces for conditional modeling. Due to privacy concerns surrounding identifiable facial data (under IRB approval), patient data cannot be publicly released. Code and model weights will be made publicly available upon acceptance.

  • Clinical Validation (R2) We agree that expert validation is crucial for confirming practical utility. The surgeon on our team assessed the surgical plausibility of the outputs. However, a formal clinical evaluation was not included. We will perform systematic validation in our future work.

  • Realism Definition & Motivational Basis (R2) In our context, “realism” refers to both visual coherence and anatomical plausibility. Visually, as shown in Fig.3, CAMOS produces smoother and more anatomically coherent surfaces than existing methods. Quantitatively, realism could be assessed via surface smoothness (e.g., mean curvature) or anatomical consistency (e.g., cephalometric measurements) to be explored in future work. Regarding the coarse-to-fine perceptual hypothesis, we appreciate the suggestion and will include supporting references (Sugase et al., 1999; Goffaux et al., 2011).

  • Hyperparameter Tuning (R1) We thank the reviewer for pointing this out. Table 3 was initially based on test set results, which was inappropriate for hyperparameter tuning. Re-evaluation on the validation set showed the same trend, with the chosen configuration (codebook:256, scale:8) still performing the best. We will update Table 3 with validation results in the camera-ready version.

  • Additional Clarifications & Corrections (R1, R4)
  • Figure legends, labels, and notations as well as typos will be fixed in the final version.
  • In Tables 2-3, bold values indicate the best performance (CAMOS). Asterisks mark statistical significance for comparisons against CAMOS, hence not applicable to CAMOS results.
  • “Normal subject” will be renamed to “Control subject”.



Meta-Review

Meta-review #1

  • Your recommendation

    Invite for Rebuttal

  • If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.

    N/A

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    N/A



Meta-review #2

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    CAMOS presents a hierarchical, multi‐scale autoregressive model that predicts a patient’s optimal 3D facial surface for orthognathic surgical planning by pretraining on large-scale normal faces and fine‐tuning on paired pre‐/postoperative cases. Reviewers praised its novelty in directly modeling soft‐tissue outcomes and its clear, comprehensive evaluation showing superior reconstruction accuracy. Initial concerns about clinical translatability, hyperparameter tuning, and dataset release were addressed in the rebuttal, leading all reviewers to upgrade to “Accept.” Given its technical contribution, thorough rebuttal, and unanimous post‐rebuttal comments, I recommend accepting this paper.



Meta-review #3

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    N/A



back to top