English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Despite their impressive performance in various surgical scene understandingtasks, deep learning-based methods are frequently hindered from deploying toreal-world surgical applications for various causes. Particularly, datacollection, annotation, and domain shift in-between sites and patients are themost common obstacles. In this work, we mitigate data-related issues byefficiently leveraging minimal source images to generate synthetic surgicalinstrument segmentation datasets and achieve outstanding generalizationperformance on unseen real domains. Specifically, in our framework, only onebackground tissue image and at most three images of each foreground instrumentare taken as the seed images. These source images are extensively transformedand employed to build up the foreground and background image pools, from whichrandomly sampled tissue and instrument images are composed with multipleblending techniques to generate new surgical scene images. Besides, weintroduce hybrid training-time augmentations to diversify the training datafurther. Extensive evaluation on three real-world datasets, i.e., Endo2017,Endo2018, and RoboTool, demonstrates that our one-to-many synthetic surgicalinstruments datasets generation and segmentation framework can achieveencouraging performance compared with training with real data. Notably, on theRoboTool dataset, where a more significant domain gap exists, our frameworkshows its superiority of generalization by a considerable margin. We expectthat our inspiring results will attract research attention to improving modelgeneralization with data synthesizing.

Generalizing Surgical Instruments Segmentation to Unseen Domains with One-to-Many Synthesis