Reproduce “Intra-Architecture Source Attribution for Stable Diffusion: A New Task with Benchmark Dataset and Baseline” 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
Ashutosh Dayal, Mallika Garg, Debashis Sen
The authors introduce IntrAA-SD, a system for attributing images to one of ten fine-tuned SDXL models, and a 140,000-image benchmark. It combines style features from CLIP with structural features from LoRA-adapted DINOv2, routing the latter through experts. Its average accuracy is 99.03% on in-domain prompts and 88.55% on held-out prompts, within this benchmark and not unseen model families.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.