1 paper · 1 filter
Raman Dutt, Pedro Sanchez, Ondrej Bohdal +2
In this work, we present compelling evidence that controlling model capacity during fine-tuning can effectively mitigate memorization in diffusion models. Specifically, we demonstr…