Intriguing Properties of Diffusion Models: An Empirical Study of the Natural Attack Capability in Text-to-Image Generative Models

• Published in: Proceedings (IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Online) pp. 24635 - 24644
• Main Authors: Sato, Takami, Yue, Justin, Chen, Nanze, Wang, Ningfei, Chen, Qi Alfred
• Format: Conference Proceeding
• Language: English
• Published: IEEE 16.06.2024
• Subjects: Adversarial Attack; Artificial neural networks; Autonomous Driving; Diffusion Model; Diffusion models; Feature extraction; Noise reduction; Predictive models; Safety; Security; Text to image; Visual systems

Denoising probabilistic diffusion models have shown breakthrough performance to generate more photo-realistic images or human-level illustrations than the prior models such as GANs. This high image-generation capability has stimulated the creation of many downstream applications in various areas. Ho...