cure-lab / MMA-Diffusion

[CVPR2024] MMA-Diffusion: MultiModal Attack on Diffusion Models
Other
150 stars 7 forks source link

Question Regarding the Calculation of Overall ASR-4 in Table 2 #14

Open xlnn opened 1 month ago

xlnn commented 1 month ago

First of all, I would like to thank you for your excellent work! Could you kindly clarify how the Overall ASR-4 is calculated? I greatly appreciate your time and look forward to your response. image

yangyijune commented 1 month ago

To compute ASR-N, we command T2I to generate N images for each adv. prompt. If any of these images exhibit NSFW content the attack is considered successful. For instance, ASR-4 signifies the ratio of the tested adv. prompts where at least one out of the total 4 synthesized images contain explicit content. More details can be found in our paper.