Get Started
Research questionHow should diffusion samplers allocate limited neural evaluations without sacrificing generated image quality?Diffusion generation requires many sequential denoising evaluations, while different parts of a sampling trajectory may contain different amounts of useful change. Uniform schedules therefore make it difficult to decide where steps can be removed without degrading samples.
Diffusion Models
Image Generation
Inference Optimization
Latest papersRecent research connected to this question, newest first.GeoSPRINT: Geometric Redundancy-Aware Step Pruning for Inference in Diffusion TrajectoriesThe evidence covers CIFAR-10 at 32×32, LSUN Church at 256×256, and Stable Diffusion v1.5 at 512×512 latent resolution, using trajectory-based schedules and matched NFE comparisons. Reported outcomes focus on FID and trajectory-geometry diagnostics, including a first-order DDIM sampling setting; the evidence does not establish behavior beyond these evaluated models and configurations.research paper · Sep 2, 2026
Related questions
How can few-step diffusion defer high-resolution denoising without causing transition artifacts?How can masked diffusion language models retain generation quality with only a few denoising steps?How can continuous diffusion language models reduce denoising steps without sacrificing text-generation quality?How can high-resolution diffusion Transformers prune tokens without sacrificing image fidelity or predictable compute?
Home
Topics
Search
Library