Reproduce “MAST: Mask-Guided Attention Control for Training-Free Regional-Multi Style Transfer” 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
Dongkyung Kang, Jaeyeon Hwang, Junseo Park, Minji Kang, Yeryeong Lee, Beomseok Ko, Hanyoung Roh, Jeongmin Shin and 1 more
MAST performs regional-multi style transfer by assigning reference styles to masked regions of a content image within a diffusion model’s attention. Mask-guided logit adjustments allocate attention, adaptive sharpening preserves selectivity, and discrepancy-weighted feature injection restores details. Across two-to-five-style tests in Stable Diffusion v1.4 with 50-step DDIM, mean ArtFID, combining style-distribution and content-perceptual similarity, was 14.928 versus 16.075 for Any-to-Any, within this protocol.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.