OPTML-Group / UnlearnCanvas

[NeurIPS 2024 D&B Track] UnlearnCanvas: A Stylized Image Dataset to Benchmark Machine Unlearning for Diffusion Models by Yihua Zhang, Chongyu Fan, Yimeng Zhang, Yuguang Yao, Jinghan Jia, Jiancheng Liu, Gaoyuan Zhang, Gaowen Liu, Ramana Kompella, Xiaoming Liu, Sijia Liu
https://unlearn-canvas.netlify.app
51 stars 2 forks source link

SalUn Unlearning Process #11

Open NGK2110 opened 3 weeks ago

NGK2110 commented 3 weeks ago

Hi sir, it's a really great work. I had a few doubts about how you have done unlearning for the SalUn work. In the original SalUn work (another great work) they showed unlearning for classes in Imagenette and NSFW content. Could you explain the experimental setup in more detail, particularly how the unlearning was implemented for any random concept like Brad Pitt? I would really appreciate any insights on this aspect. Thanks again for your great work, and I look forward to hearing from you.

NormalUhr commented 5 days ago

Thank you very much for your recognition and appreciation of our work! In SalUn, concepts that need to be unlearned are re-labeled to other concepts. For example, if we need to unlearn images of nude individuals, we re-label them as clothed individuals. In the case of “Brad Pitt”, you can map it to either other celebrities or individuals with more generalized facial features. Along this process, you may need to prepare some images corresponding to the mapped concept as well.