search

somepago / DCR

Official Pytorch repo of CVPR'23 and NeurIPS'23 papers on understanding replication in diffusion models.
Apache License 2.0
104 stars 5 forks source link

Mitigation strategies：Training-time mitigation strategy and inference time mitigation strategy do not use the same Stable diffusion？ #7

Closed liuxiao-guan closed 7 months ago

liuxiao-guan commented 7 months ago

Hey,

thank you for providing the code to reproduce your experiments. I want to confirm if my understanding of the different strategies using Stable diffusion versions in the 'Mitigation strategies' section of the article is correct. In the training-time mitigation strategy, you fine-tuned on Stable diffusion 2.1 and then analyzed using the generated model. However, in the inference time mitigation strategy, you directly incorporated the mitigation strategy on Stable diffusion 1.4?

somepago commented 7 months ago

In the paper, we used SD 2.1 for the inference time mitigation as well. (I guess I pushed the code version when I was testing the same for SD 1.4.)