List of Papers Browse by Subject Areas Author List
Abstract
Minimally invasive and robot-assisted surgery relies heavily on endoscopic imaging, yet surgical smoke produced by electrocautery and vessel-sealing instruments can severely degrade visual perception and hinder vision-based functionalities. We present a transformer-based surgical desmoking model with a physics-inspired desmoking head that jointly predicts smoke-free image and corresponding smoke map. To address the scarcity of paired smoky-to-smoke-free training data, we develop a synthetic data generation pipeline that blends artificial smoke patterns with real laparoscopic images, yielding over 80,000 paired samples for supervised training. We further curate, to our knowledge, the largest paired surgical smoke dataset to date, comprising 5,817 image pairs captured with the da Vinci robotic surgical system, enabling benchmarking on high-resolution endoscopic images. Extensive experiments on both a public benchmark and our dataset demonstrate state-of-the-art performance in image reconstruction compared to existing dehazing and desmoking approaches. We also assess the impact of desmoking on downstream stereo depth estimation and instrument segmentation, highlighting both the potential benefits and current limitations of digital smoke removal methods.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/4094_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026/supp/4094_supp.zip
Link to the Code Repository
N/A
Link to the Dataset(s)
Smoke dataset: https://huggingface.co/datasets/artJiang20/SeeThroughSmoke
BibTex
@InProceedings{LuJin_Seeing_MICCAI2026,
author = { Lu, Jingpei AND Jiang, Fengyi AND Zhang, Xiaorui AND Jin, Lingbo AND Mohareri, Omid},
title = { { Seeing Through Smoke: Surgical Desmoking for Improved Visual Perception } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16892},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper presents a transformer for desmoking. One major contribution will be the 80,000 paired samples of smoky and smoke-free synthetic data and over 5000 pairs of real data.
- 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.
- Dataset will be a great contribution to everyone working de-smoking. It’d be good to release the splits used in this paper too for benchmarking
- A fairly straightforward network architecture seems to work well to solve the smoke map estimation
- Demonstrates the utility of smoke removal for downstream tasks
- 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.
- No evaluation on real smoke from cautery. Is a fogging unit similar enough in pattern compared to smoke from cautery?
- 47 mm is a pretty large depth error, and the standard deviation is larger. The results may show that desmoking is introducing image artifacts that affect downstream tasks
- 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
- Overall, a clearly written paper with interesting contributions
- It’d have been interesting to see ablation over some structural choices of the neural network architectures
- 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?
De-smoking is an important problem to address to enable downstream tasks in surgical guidance/automation. While the results presented in this work do not accomplish desmoking, it does show improvement upon previous results. The dataset this work provides will help spur more work in this area.
- Reviewer confidence
Very confident (4)
- [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.
I think the dataset is a valid contribution, and while the results aren’t amazing, the authors’ discussion is appropriate.
Review #2
- Please describe the contribution of the paper
The paper proposes a transformer-based surgical desmoking model that combines a ViT-Base backbone, a DPT decoder, and a physics-inspired head that jointly predicts a smoke-free image and a smoke map. And a dataset with real smoke.
- 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.This paper presents a meaningful engineering effort on surgical desmoking, including dataset synthesis and real-data collection. 2.The problem is clinically and technically relevant: surgical smoke degrades visibility and can interfere with downstream vision modules in minimally invasive and robot-assisted surgery.
- 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.The paper lacks critical ablations. There is no study isolating the benefit of the physics-inspired head, the auxiliary smoke-map prediction. 2.The downstream-task evaluation is only partially persuasive since the ground truth depth and segmentation labels are pseudo-labels generated by FoundationStereo and SAM3.
- 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.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Although the paper proposes a meaningful method and a corresponding dataset, the downstream-task evaluation is only partially convincing, and there is a lack of ablation for the key component.
- 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.
The author responds to the major concern of the pseudo label, and yet, although the ablation is not allowed, the paper’s contribution with a real dataset is meaningful.
Review #3
- Please describe the contribution of the paper
This work proposes a model to desmoke surgical video, integrating a standard atmospheric scattering formulation with a transformer-based architecture to recover a smoke-free image.
- 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.
- Paper is well-written
- Disentangling the smoke map from the image in both the architectural design and for data augmentation is coherent
- Resulting desmoked images outperform all related baselines and qualitative results support this.
- 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.
- Zero-shot foundation models for depth estimation and segmentation outperform the proposed method
- Unclear what how reconstruction is evaluated in Table 2 vs Table 3
- 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.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if 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-structured approach to surgical desmoking with clearly described methods. The design choice to disentangle the smoke map from the underlying image is coherent for both the model and data generation pipeline. Experimental results demonstrate improved image reconstruction over related baselines.
However, the central motivation of the paper is to improve downstream vision-based surgical tasks which is not convincingly supported by the results. It’s unclear how reconstruction in Table 3 differs from Table 2 as SSIM and PSNR are image reconstruction metrics, not 3D reconstruction as a downstream vision task. Also, since zero-shot foundation models outperform the proposed approach on depth estimation and segmentation, this raises a fundamental question regarding the utility of the proposed method, dataset, and training pipeline. It does not seem the claimed benefits translate into measurable improvements to motivate this work.
While desmoking may still provide value for surgeon interpretability (real-time intraoperative visualization), this aspect is not evaluated or benchmarked against related methods. As a result, the paper does not establish a compelling use case where the proposed approach is advantageous.
Overall, despite strong image-level results and a coherent methodological design, the lack of demonstrated benefit significantly weakens the uti of the work and motivates my verdict.
- 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
Author Feedback
We thank the reviewers for the constructive feedback. We address the main concerns below.
Real smoke / clinical relevance (R1). We agree that our paired da Vinci benchmark uses a controlled smoke/fog source to obtain aligned smoky/clear pairs under repeatable densities. However, the submission is not evaluated only in this setting: Table 2 reports results on the public De-Smoking dataset of real in vivo laparoscopic videos from cholecystectomy and prostatectomy, where our method achieves the best SSIM/PSNR on both subsets. Thus, our benchmark provides high-resolution paired data, while the public in vivo dataset evaluates transfer to real surgical smoke. We will clarify this distinction.
Downstream evaluation and pseudo labels (R2,R3). The downstream experiments are not intended to claim that desmoking replaces task-specific models or universally improves every downstream algorithm. Their purpose is to provide a functional measure of how well desmoking reconstructs the original clear image. Existing work mainly uses pixel metrics such as SSIM/PSNR, which we also report. We further ask whether downstream models applied to desmoked images produce outputs close to those from the corresponding clear images. Under this interpretation, pseudo labels do not undermine the study: both the reference outputs and desmoked-image outputs are generated by the same downstream pipeline. The evaluation measures consistency with the clear-image behavior of that pipeline, rather than clinically validated ground truth for depth or segmentation.
This functional test assesses whether restored images preserve task-relevant structures. We explicitly report the negative depth finding that current desmoking methods do not consistently improve stereo depth under smoke. The large depth error suggests that stereo matching is difficult and that restoration can disrupt left-right consistency; we regard this as an important limitation. Conversely, for instrument segmentation, our method improves IoU over the smoky input, suggesting benefit for boundary/appearance-driven perception. We will revise the wording to emphasize this nuanced conclusion: desmoking improves image reconstruction and may help some perception tasks, but is not a drop-in solution for geometry.
Clarification on foundation models (R3). FoundationStereo and SAM3 are not competing desmoking methods; they are downstream modules applied after either no preprocessing or desmoking. Thus, the result does not mean that a zero-shot foundation model outperforms our desmoking model. It shows that FoundationStereo is robust enough that preprocessing can be unnecessary or harmful for depth, consistent with our discussion.
Reconstruction metrics in Tables 2 and 3 (R3). SSIM/PSNR in both tables evaluate 2D smoke-free image reconstruction, not 3D reconstruction. Table 2 evaluates the public in vivo De-Smoking dataset; Table 3 evaluates our held-out da Vinci benchmark and additionally reports depth/segmentation after desmoking. We will rename the Table 3 column to “Image reconstruction.”
Ablations of the physics-inspired head and smoke-map prediction (R1,R2). We acknowledge that ablations would strengthen the analysis. Eqs. (1)-(4) motivate the head: rather than predicting RGB only, the network predicts physically motivated K and B terms and derives a smoke map correlated with the airlight component. The smoke-map loss is enabled by our synthetic pipeline, which provides ground-truth smoke maps. Since rebuttal rules do not allow new experimental results, we cannot add ablations here. The submitted evidence shows that the complete design outperforms eight prior dehazing/desmoking methods.
Overall, the evidence supports our main conclusions: state-of-the-art image-level desmoking on both a public in vivo dataset and our paired da Vinci benchmark; a new paired dataset and split release; and downstream analysis identifying a positive case for segmentation and a limitation for stereo depth.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- 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.
A real-smoke dataset and transformer architecture are proposed for de-smoking endoscopic images. Issues remain regarding the performance comparison with zero-shot foundation models, as well as the lack of ablations. There was also no evaluation on real smoke from a cautery pen, the most likely source. Large depth errors may affect results.
- 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.
Reviewers appreciated the author’s comments in the rebuttal, which have addressed their concerns. The rebuttal clarifies that Table 2 reports results on the De-Smoking dataset from real in vivo laparoscopic videos, which alleviates concerns about real smoke evaluation. It also clarified the use of pseudo-labels in evaluation, the point of which is to evaluate consistency with clear-image behavior of models trained on clear images, which is a reasonable evaluation strategy given the lack of real smoke data with ground truth.
I particularly appreciated the point that the use of pseudo-labels here goes beyond comparable work that relies solely on image-level metrics like SSIM and PSNR.
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.
The rebuttal effectively addressed the reviewers’ concerns, providing sufficient clarification and additional evidence where requested. Following the rebuttal, all reviewers expressed support for acceptance, indicating a clear consensus that the paper meets the standard for acceptance at MICCAI.
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.
The reviewers’ concerns have been successfully addressed in the rebuttal, making the work a solid contribution.