Evaluating the Robustness of Text-to-image Diffusion Models against Real-world Attacks

Hongcheng Gao, Hao Zhang, Yinpeng Dong, Zhijie Deng

Primary Area: societal considerations including fairness, safety, privacy

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Diffusion Models, Text to Image Generation, Adversarial Attack, Robustness Evaluation

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract:

Text-to-image (T2I) diffusion models (DMs) have shown promise in generating high-quality images from textual descriptions. The real-world applications of these models require particular attention to their safety and fidelity, but this has not been sufficiently explored.
One fundamental question is whether the existing T2I DMs are robust against variations over input texts. To answer it, this work provides the first robustness evaluation of T2I DMs against real-world perturbations. Unlike malicious attacks that involve apocryphal alterations to the input texts, we consider a perturbation space spanned by realistic errors (e.g., typo, glyph, phonetic) that humans can make and adopt adversarial attacks to generate worst-case perturbations for robustness evaluation. Given the inherent randomness of the generation process, we develop novel distribution-based objectives to mislead T2I DMs. We optimize the objectives by black-box attacks without any knowledge of the model. Extensive experiments demonstrate the effectiveness of our method for attacking popular T2I DMs and simultaneously reveal their non-trivial robustness issues. Moreover, we provide an in-depth analysis of our method to show that it is not designed to attack the text encoder in T2I DMs solely.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 3821