Reproduce “MOST-GAN: 3D Morphable StyleGAN for Disentangled Face Image Manipulation” 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
Safa C. Medin, Bernhard Egger, Anoop Cherian, Ye Wang, Joshua B. Tenenbaum, Xiaoming Liu, Tim K. Marks
MOST-GAN edits portrait faces by controlling 3D shape, albedo, lighting, and pose separately, while handling hair with a 2D pathway. Its face representation combines a 3D morphable model with StyleGAN2 synthesis and a differentiable renderer. On 100 test images at +90° yaw, it reports 6.7° mean absolute pose error versus PIE’s 81.0°, using estimated—not ground-truth—output pose.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.