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

While image data starts to enjoy the simple-but-effective self-supervisedlearning scheme built upon masking and self-reconstruction objective thanks tothe introduction of tokenization procedure and vision transformer backbone,convolutional neural networks as another important and widely-adoptedarchitecture for image data, though having contrastive-learning techniques todrive the self-supervised learning, still face the difficulty of leveragingsuch straightforward and general masking operation to benefit their learningprocess significantly. In this work, we aim to alleviate the burden ofincluding masking operation into the contrastive-learning framework forconvolutional neural networks as an extra augmentation method. In addition tothe additive but unwanted edges (between masked and unmasked regions) as wellas other adverse effects caused by the masking operations for ConvNets, whichhave been discussed by prior works, we particularly identify the potentialproblem where for one view in a contrastive sample-pair the randomly-sampledmasking regions could be overly concentrated on important/salient objects thusresulting in misleading contrastiveness to the other view. To this end, wepropose to explicitly take the saliency constraint into consideration in whichthe masked regions are more evenly distributed among the foreground andbackground for realizing the masking-based augmentation. Moreover, we introducehard negative samples by masking larger regions of salient patches in an inputimage. Extensive experiments conducted on various datasets, contrastivelearning mechanisms, and downstream tasks well verify the efficacy as well asthe superior performance of our proposed method with respect to severalstate-of-the-art baselines.

Masking Improves Contrastive Self-Supervised Learning for ConvNets, and Saliency Tells You Where