Year: 2024
Type: conference-paper
Abstract: The smoke generated by surgical tools during laparoscopic surgeries can impair visibility, affecting surgeons' accuracy and safety during procedures. Therefore, it is highly desirable to remove smoke from laparoscopic images. Despite growing interest in laparoscopic image desmoking and the development of various algorithms, the main challenge in advancing this field remains the absence of publicly available, high-quality benchmark datasets. To further progress in this field, we have developed a new high-quality dataset for laparoscopic surgery image desmoking, called LSD3K, which comprises 3,000 pairs of synthetic non-homogeneous smoke images. In this paper, we introduce a dataset generation pipeline that includes modeling smoke shapes with Blender, gathering ground-truth images from the Cholec80 dataset, and randomly sampling smoke masks, along with other steps. We also performed a comprehensive evaluation of existing representative desmoking algorithms using the proposed benchmark. The proposed dataset will be made publicly available at https://drive.google.com/file/d/1U4fzZJM8ch9wlwwMwjX2ru5F0ObkeD/view?usp=sharing
Authors: Wenhui Chang, Yufeng Li, Zebang Zhu, Yuchen Yang
Institution: Shenyang Aerospace University
Language: English
Topic: Image Enhancement Techniques
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4404317678