Ke Li Announces ROMS-IMLE One-Step Model
Assistant professor presents simple one-step model as alternative to diffusion methods.
TLDR
Ke Li, Assistant Professor of Computing Science at Simon Fraser University, posted on X that diffusion and flow matching methods are slow, sample inefficient, and complicated. The researcher and collaborator @researchirag introduced ROMS-IMLE, a small and simple one-step model. The post states the model attains an FID of 2.56 on ImageNet. The announcement forms the first of a six-post thread that includes a link and a 20-second video attachment of a static presentation slide.
Combined views
34.7K
1 Source, first seen 26d ago