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

arXiv:2308.15692v2 Announce Type: replace
Abstract: 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. However, we find that this technology is actually a double-edged sword: We identify a new type of attack, called the Natural Denoising Diffusion (NDD) attack based on the finding that state-of-the-art deep neural network (DNN) models still …

abstract applications arxiv capability cs.cr cs.cv denoising diffusion diffusion models gans generate generative generative models human illustrations image images natural performance photo prior study text text-to-image type