1 paper across 1 session
We propose blockwise flow modeling framework that aim for optimal Pareto-frontier between generation performance (FID) and inference complexity (FLOPs).