Flow Matching and Fast Sampling

This database tracks a major post-DDPM direction: preserving the quality of diffusion-like models while reducing sampling cost, simplifying the generative path, or training continuous generative dynamics more directly.

Fast Diffusion Sampling and Distillation

Year	Paper	Topic	Note
2022	Progressive Distillation for Fast Sampling of Diffusion Models	Distillation	Distills many-step diffusion sampling into fewer steps.
2022	DPM-Solver	Diffusion ODE solver	Dedicated high-order solver for fast diffusion sampling.
2023	Consistency Models (OpenAI)	One-step generation	Maps noise to data with one-step or few-step generation.

Flow Matching and Rectified Flow

Year	Paper	Topic	Note
2022	Flow Straight and Fast	Rectified flow	Learns straighter ODE transport paths for efficient generation.
2022	Flow Matching for Generative Modeling	Flow matching	Simulation-free training of continuous normalizing flows through vector-field regression.
2023	Stochastic Interpolants	Interpolant framework	General framework connecting diffusion, flows, and stochastic bridges.
2024	Scalable Interpolant Transformers	SiT	Scales interpolant-based generative modeling with Transformer backbones.
2024	Scaling Rectified Flow Transformers for High-Resolution Image Synthesis	Stable Diffusion 3 / MM-DiT	Rectified-flow Transformer architecture for high-resolution text-to-image synthesis.

Reading Path

Step	Read
1	DDIM and diffusion ODE ideas in Diffusion Models.
2	Progressive Distillation and DPM-Solver for fast sampling.
3	Consistency Models for one-step/few-step generation.
4	Rectified Flow, Flow Matching, and Stochastic Interpolants.
5	SiT and Stable Diffusion 3 for modern Transformer-based flow/rectified-flow systems.