Reproduce “Unrestricted Adversarial Examples via Semantic Manipulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Anand Bhattad, Min Jin Chong, Kaizhao Liang, Bo Li, D. A. Forsyth
The authors introduce cAdv and tAdv, attacks that redirect image classifiers by changing color with a pretrained colorizer or transferring texture from another image. They test classification on ImageNet, captioning on MSCOCO, selected defenses, and human judgments. In one ImageNet transfer test, cAdv reaches 28.56% on DenseNet121 and 31.00% on VGG19; these results apply only to the tested settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.