Efficient Sampling of Diffusion Models through Adaptive Timesteps

This project proposes an adaptive timestep method for diffusion models that adjusts sampling dynamically based on intermediate states. Compared to existing global approaches, it improves efficiency while maintaining or improving output quality.