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

Following their success in visual recognition tasks, VisionTransformers(ViTs) are being increasingly employed for image restoration. As afew recent works claim that ViTs for image classification also have betterrobustness properties, we investigate whether the improved adversarialrobustness of ViTs extends to image restoration. We consider the recentlyproposed Restormer model, as well as NAFNet and the "Baseline network" whichare both simplified versions of a Restormer. We use Projected Gradient Descent(PGD) and CosPGD, a recently proposed adversarial attack tailored to pixel-wiseprediction tasks for our robustness evaluation. Our experiments are performedon real-world images from the GoPro dataset for image deblurring. Our analysisindicates that contrary to as advocated by ViTs in image classification works,these models are highly susceptible to adversarial attacks. We attempt toimprove their robustness through adversarial training. While this yields asignificant increase in robustness for Restormer, results on other networks areless promising. Interestingly, the design choices in NAFNet and Baselines,which were based on iid performance, and not on robust generalization, seem tobe at odds with the model robustness. Thus, we investigate this further andfind a fix.

On the unreasonable vulnerability of transformers for image restoration -- and an easy fix