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

Deep learning techniques have shown promising results in image compression,with competitive bitrate and image reconstruction quality from compressedlatent. However, while image compression has progressed towards a higher peaksignal-to-noise ratio (PSNR) and fewer bits per pixel (bpp), their robustnessto adversarial images has never received deliberation. In this work, we, forthe first time, investigate the robustness of image compression systems whereimperceptible perturbation of input images can precipitate a significantincrease in the bitrate of their compressed latent. To characterize therobustness of state-of-the-art learned image compression, we mount white-boxand black-box attacks. Our white-box attack employs fast gradient sign methodon the entropy estimation of the bitstream as its bitrate approximation. Wepropose DCT-Net simulating JPEG compression with architectural simplicity andlightweight training as the substitute in the black-box attack and enable fastadversarial transferability. Our results on six image compression models, eachwith six different bitrate qualities (thirty-six models in total), show thatthey are surprisingly fragile, where the white-box attack achieves up to56.326x and black-box 1.947x bpp change. To improve robustness, we propose anovel compression architecture factorAtn which incorporates attention modulesand a basic factorized entropy model, resulting in a promising trade-offbetween the rate-distortion performance and robustness to adversarial attacksthat surpasses existing learned image compressors.

MALICE: Manipulation Attacks on Learned Image ComprEssion