acceptodds
Under review as a conference paper at ICLR 2027

SurgiMist: Controlled Smoke Synthesis for Surgical Image Restoration and Perception

Abstract

Surgical smoke changes image appearance and obscures cues used by both restoration models and downstream perception systems. Paired clear and smoky observations are difficult to acquire in surgery, while restoration scores alone do not establish whether removing smoke helps a task. We introduce SurgiMist, a controlled synthesis framework that combines relative depth, heterogeneous optical fields, layered accumulation, texture evolution, and scattering-inspired image formation. Its nested configurations support restoration benchmarking, and a geometry-preserving variant transfers existing semantic labels to smoke-augmented images. We evaluate the framework through a fixed restoration benchmark, three restoration backbones on external data, and a video-disjoint surgical segmentation study with three training seeds. On the fixed four-condition restoration benchmark, the complete configuration achieves the best mean PSNR, SSIM, and LPIPS among the four synthesis configurations. For segmentation, its geometry-preserving variant reaches 50.18 ± 0.12% mean intersection-over-union across 15 balanced synthetic conditions, a 15.71 percentage-point gain over original-image training. Under severe smoke, direct augmentation reaches 50.28%, compared with 39.45% for a learned restoration front end. In a blinded review of 500 paired judgments by five doctors on real-smoke frames, B4 receives slightly more votes than LSD3K for visibility, structure preservation, and viewing preference, and fewer artifact flags. Matched component controls separate the benefit of smoke augmentation from Gaussian smoothing alone. Together, these results position controlled synthesis as a practical interface for measuring restoration quality and training smoke-robust perception, with the downstream objective determining the useful processing route.

Primary area:
datasets and benchmarks
Keywords:
Surgical Image Restoration, low-level, defog, dehaze, desmoke

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_KshHOpenReview ↗
Rating
1
Confidence
4
Informativeness
1
Presentation
3
Soundness
3
Read review
Paper summary

The paper introduces SurgiMist, a controllable surgical smoke synthesis framework for restoration benchmarking and smoke-robust perception training.

Critical strengths

The paper addresses a well-established and clearly defined problem: surgical smoke degrades image quality and interferes with downstream visual perception. The motivation is straightforward and practically relevant, making the intended use of the proposed framework easy to understand.

Critical weaknesses

1.Limited novelty and incomplete related work. Controlled degradation synthesis and the mismatch between restoration quality and downstream performance are well-established topics. The paper does not clearly demonstrate a substantial advance. Its references rely heavily on older literature and omit recent, closely related studies, weakening its positioning.

2.Underdeveloped presentation. Typography and visual styles are inconsistent across figures. Figure 1 appears AI-generated on initial visual inspection, although its provenance is unverified. Figure 2 has excessively small labels, and Figure 3 largely repeats Table 3 without adding meaningful analysis. These issues impair readability and suggest insufficient preparation of the manuscript.

Questions for authors

What specific capability distinguishes SurgiMist from recent surgical smoke synthesis and downstream evaluation studies, particularly Rethinking Surgical Smoke and Seeing Through Smoke?

Reviewer_Q2HhOpenReview ↗
Rating
2
Confidence
4
Informativeness
2
Presentation
3
Soundness
2
Read review
Paper summary

This paper introduces SurgiMist, a controllable surgical smoke synthesis framework for restoration and segmentation. It combines depth-dependent attenuation, heterogeneous fields, layered accumulation, procedural texture evolution, and image-space optical effects. Experiments examine restoration across several backbones and compare smoke augmentation with restoration followed by segmentation.

Critical strengths
  1. The framework offers controllable degradation and a geometry-preserving variant that reuses existing segmentation labels.
  2. The component controls, video-disjoint splits, and three training seeds provide useful experimental structure.
  3. The paper appropriately distinguishes image quality from downstream accuracy and acknowledges mixed external results.
Critical weaknesses
  1. The baseline comparisons remain inadequate. Benchmarking against real-data self-supervised approaches, such as SelfSVD [1], is necessary to demonstrate the benefits of synthetic-data training.
  2. Downstream validation relies on synthetic smoke across only three videos, which limits clinical relevance. Evaluating segmentation on authentic smoky videos spanning diverse surgical procedures would strengthen the claims of real-world robustness.
  3. The augmentation-versus-restoration comparison lacks fair control, given that only the augmented predictor receives downstream adaptation. To fairly justify the preference for augmentation, the authors should include predictors fine-tuned on restored frames as well as jointly trained restoration–segmentation pipelines under a matched training budget.

[1] Wu, R., Zhang, Z., Zhang, S., Gou, L., Chen, H., Zhang, L., Chen, H., and Zuo, W. Self-Supervised Video Desmoking for Laparoscopic Surgery. ECCV, 2024.

Questions for authors

Please see Weaknesses.

Reviewer_2QZGOpenReview ↗
Rating
1
Confidence
4
Informativeness
2
Presentation
2
Soundness
1
Read review
Paper summary

This paper proposes SurgiMist, a depth-based controllable surgical smoke synthesis framework with four nested configurations (B1–B4) and a label-preserving variant. It models smoke using relative depth, heterogeneous optical fields, layered accumulation, texture evolution, and scattering-inspired effects, with a geometry-preserving variant for label-preserving segmentation augmentation. Experiments show that SurgiMist provides promising results.

Critical strengths

The paper includes nested synthesis configurations, multiple restoration backbones, cross-source evaluation, component controls, three-seed segmentation experiments, runtime analysis, and blinded doctor assessment. This breadth makes the conclusions about smoke augmentation and restoration routes substantially more convincing.

Critical weaknesses

1. The proposed formulation appears closer to generic atmospheric haze synthesis than to surgical smoke formation. Surgical endoscopes typically use near-field, co-axial, non-parallel illumination (unlike general sense), where smoke–light interaction can produce strong local backscatter and highly nonuniform visibility. In contrast, SurgiMist mainly models attenuation through relative scene depth and a constant airlight term, without explicitly accounting for the smoke source location or the endoscopic illumination geometry. This may lead to a different spatial obscuration pattern from real surgical smoke. Since the synthesized smoke is placed independently of any smoke source, to what extent does the B1–B4 benchmark reflect real surgical smoke? The authors should better justify why this haze-inspired formulation is appropriate for surgical smoke and clarify its domain of validity.

2. The comparison between the B4 and LSD3K datasets in the main text involves datasets of vastly different scales; therefore, it is impossible to determine whether the better fit is due to the higher data quality of B4 or simply the larger volume of data. In Table 39 and 42 in the appendix, the authors created datasets with the same number of pairs, but the metrics—including PSNR, SSIM, and LPIPS—were all lower than those of LSD3K. However, NIQE/BRISQUE/FADE are built on natural-scene statistics fitted to general scenes; it is unclear whether they are valid for laparoscopic frames.

Questions for authors

· I hope the authors can provide further analysis and explanation based on the modeling theory.

· Explanation of the experimental results for LSD3K and B4.

· There are a few articles that are not cited in the main text but appear in the references section: MaSTr1325, Foggy Cityscapes...

· Appendix F demonstrates, “Human-study packets were prepared, but no ratings were returned; no human preference results are reported.” However, the preprint reports 500 judgments made by five physicians. Please provide a further explanation for this.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.