SNAFusion-MM: Distilling sparse sampled measurements by 2D axial diffusion priors with multi-step matching for 3D inverse problem

Open Webpage

Xiaoyue Li, Tielong Cai, Jun Dan, Sizhao Ma, Kai Shang, Mark D. Butala, Gaoang Wang

Published: 01 Jan 2025, Last Modified: 15 Jul 2025Inf. Fusion 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Unifies 2D diffusions & 3D INR via novel score distillation.•Solves 3D inverse problems in various cases, no retraining.•Tackles 3D radial CS-MRI unaddressed by SOTAs.•Outperforms SOTAs with higher PSNR on ultra-sparse views.•Reduces 3D memory costs while enhancing global coherence.