2 papers across 2 sessions
This work improves CLIP’s visual detail capturing ability by inverting the unCLIP generative model, which we find suitable for achieving this goal.