Reproduce “MULTI: Disentangling Camera Lens, Sensor, View, and Domain for Novel Image Generation” 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
Sonali Godavarthy, Matthias Neuwirth-Trapp, Tim-Felix Faasch, Maarten Bieshaar, Michael Moeller, Danda Pani Paudel
MULTI learns separate controls for lens, sensor, viewpoint, and domain in text-to-image generation. It uses two-stage textual inversion across partially overlapping datasets, keeping the pretrained diffusion model frozen. In a user study of novel combinations, participants preferred MULTI at 53.53% over DreamBooth at 32.37%. However, unseen combinations remain inconsistent and can contain artifacts; this does not establish independent factor control.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.