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

Prototypical contrastive learning (PCL) has been widely used to learnclass-wise domain-invariant features recently. These methods are based on theassumption that the prototypes, which are represented as the central value ofthe same class in a certain domain, are domain-invariant. Since the prototypesof different domains have discrepancies as well, the class-wisedomain-invariant features learned from the source domain by PCL need to bealigned with the prototypes of other domains simultaneously. However, theprototypes of the same class in different domains may be different while theprototypes of different classes may be similar, which may affect the learningof class-wise domain-invariant features. Based on these observations, acalibration-based dual prototypical contrastive learning (CDPCL) approach isproposed to reduce the domain discrepancy between the learned class-wisefeatures and the prototypes of different domains for domain generalizationsemantic segmentation. It contains an uncertainty-guided PCL (UPCL) and ahard-weighted PCL (HPCL). Since the domain discrepancies of the prototypes ofdifferent classes may be different, we propose an uncertainty probabilitymatrix to represent the domain discrepancies of the prototypes of all theclasses. The UPCL estimates the uncertainty probability matrix to calibrate theweights of the prototypes during the PCL. Moreover, considering that theprototypes of different classes may be similar in some circumstances, whichmeans these prototypes are hard-aligned, the HPCL is proposed to generate ahard-weighted matrix to calibrate the weights of the hard-aligned prototypesduring the PCL. Extensive experiments demonstrate that our approach achievessuperior performance over current approaches on domain generalization semanticsegmentation tasks.

Calibration-based Dual Prototypical Contrastive Learning Approach for Domain Generalization Semantic Segmentation