Curiosity some details on CLUE-MARK

[chengyi-chris]

Hi, thank you for sharing this interesting project! I’m also exploring latent watermarking methods based on the LWE problem in diffusion models.

It would be great to see evaluation results on standard benchmark experiments, such as:

Geometry attacks

Removal attacks

Forgery attacks (e.g., regeneration attacks)

I’m also curious whether CLUE-MARK supports multi-bit scenarios. Could you share the FPR, TPR, and detection threshold values used in your work?

Recently, I’ve been looking into undetectable (in-processing) watermarking methods for diffusion models, such as PRCW (ICLR 2025), and I’d like to further investigate their robustness against adversarial attacks.

I’d be happy to discuss potential directions or ideas if you’re open to it. Thanks again for your contribution!

Best,
ChengYi