Reproduce “LIME: Localized Image Editing via Attention Regularization in Diffusion Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Enis Simsar, Alessio Tonioni, Yongqin Xian, Thomas Hofmann, Federico Tombari
LIME localizes image edits from an instruction and input image, then steers diffusion model attention to the inferred region without user-provided masks or retraining. It fuses multi-resolution features with instruction-conditioned cross-attention to select that region. On MagicBrush single-turn IP2P, L1 falls from 0.112 to 0.058; a controlled ablation favors attention regularization over noise-space editing, though IP2P’s entanglement remains.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.