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

Supervised training of deep neural networks (DNNs) by noisy labels has beenstudied extensively in image classification but much less in imagesegmentation. Our understanding of the learning behavior of DNNs trained bynoisy segmentation labels remains limited. We address this deficiency in bothbinary segmentation of biological microscopy images and multi-classsegmentation of natural images. We classify segmentation labels according totheir noise transition matrices (NTMs) and compare performance of DNNs trainedby different types of labels. When we randomly sample a small fraction (e.g.,10%) or flip a large fraction (e.g., 90%) of the ground-truth labels to trainDNNs, their segmentation performance remains largely unchanged. This indicatesthat DNNs learn structures hidden in labels rather than pixel-level labels perse in their supervised training for semantic segmentation. We call these hiddenstructures meta-structures. When labels with different perturbations to themeta-structures are used to train DNNs, their performance in feature extractionand segmentation degrades consistently. In contrast, addition of meta-structureinformation substantially improves performance of an unsupervised model inbinary semantic segmentation. We formulate meta-structures mathematically asspatial density distributions. We show theoretically and experimentally howthis formulation explains key observed learning behavior of DNNs.

Elucidating Meta-Structures of Noisy Labels in Semantic Segmentation by Deep Neural Networks