search

zhaisf / BadT2I

[MM'23 Oral] "Text-to-image diffusion models can be easily backdoored through multimodal data poisoning"
https://dl.acm.org/doi/10.1145/3581783.3612108
MIT License
22 stars 0 forks source link

pretrained model #4

Closed Aemilia-xu closed 8 months ago

Aemilia-xu commented 8 months ago

Can you provide a pre-trained bacckdoor model? This will be of great help to us. Thank you!

zhaisf commented 8 months ago

Thank you for your interest in our work!

I'm sorry to say that we haven't intentionally retained the backdoored models, due to the substantial storage space needed for the SD models.

However, with the released code and the experiment setup outlined in our paper, it is not difficult to obtain backdoored models. And I'm more than happy to assist with any issues you encounter during replication.