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/

Zendy Plus

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

Zendy Tools

Zendy Open

When a small number of poisoned samples are injected into the trainingdataset of a deep neural network, the network can be induced to exhibitmalicious behavior during inferences, which poses potential threats toreal-world applications. While they have been intensively studied inclassification, backdoor attacks on semantic segmentation have been largelyoverlooked. Unlike classification, semantic segmentation aims to classify everypixel within a given image. In this work, we explore backdoor attacks onsegmentation models to misclassify all pixels of a victim class by injecting aspecific trigger on non-victim pixels during inferences, which is dubbedInfluencer Backdoor Attack (IBA). IBA is expected to maintain theclassification accuracy of non-victim pixels and mislead classifications of allvictim pixels in every single inference and could be easily applied toreal-world scenes. Based on the context aggregation ability of segmentationmodels, we proposed a simple, yet effective, Nearest-Neighbor trigger injectionstrategy. We also introduce an innovative Pixel Random Labeling strategy whichmaintains optimal performance even when the trigger is placed far from thevictim pixels. Our extensive experiments reveal that current segmentationmodels do suffer from backdoor attacks, demonstrate IBA real-worldapplicability, and show that our proposed techniques can further increaseattack performance.

Influencer Backdoor Attack on Semantic Segmentation