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

The diversity of retinal imaging devices poses a significant challenge:domain shift, which leads to performance degradation when applying the deeplearning models trained on one domain to new testing domains. In this paper, wepropose a multi-scale input along with multiple domain adaptors appliedhierarchically in both feature and output spaces. The proposed trainingstrategy and novel unsupervised domain adaptation framework, calledCollaborative Adversarial Domain Adaptation (CADA), can effectively overcomethe challenge. Multi-scale inputs can reduce the information loss due to thepooling layers used in the network for feature extraction, while our proposedCADA is an interactive paradigm that presents an exquisite collaborativeadaptation through both adversarial learning and ensembling weights atdifferent network layers. In particular, to produce a better prediction for theunlabeled target domain data, we simultaneously achieve domain invariance andmodel generalizability via adversarial learning at multi-scale outputs fromdifferent levels of network layers and maintaining an exponential movingaverage (EMA) of the historical weights during training. Without annotating anysample from the target domain, multiple adversarial losses in encoder anddecoder layers guide the extraction of domain-invariant features to confuse thedomain classifier. Meanwhile, the ensembling of weights via EMA reduces theuncertainty of adapting multiple discriminator learning. Comprehensiveexperimental results demonstrate that our CADA model incorporating multi-scaleinput training can overcome performance degradation and outperformstate-of-the-art domain adaptation methods in segmenting retinal optic disc andcup from fundus images stemming from the REFUGE, Drishti-GS, and Rim-One-r3datasets.

CADA: Multi-scale Collaborative Adversarial Domain Adaptation for Unsupervised Optic Disc and Cup Segmentation