PulseTrendingJournal ClubResearchersJournalsExplore
Instagram
HomeTrendingJournal ClubExplore
Synapse
⌘+K
Synapse
October 2, 2025Open Access

LowDiff: Efficient Diffusion Sampling with Low-Resolution Condition

View Full Paper
Ask AI
Bookmark
Share

Authors

JXJ. Q. XuQJQing JinMCMeida Chen

Discussion

Loading...

Member takes

Overview

LowDiff improves sampling speed in image generation tasks, suggesting significant gains in efficiency and quality.

Key Points

  • LowDiff achieves over 50% throughput improvement while maintaining image quality across various datasets.
  • On CIFAR-10, LowDiff reports an FID of 2.11 and IS of 9.87 for unconditional generation.
  • The methodology includes generating images from low resolution to higher resolutions using a unified model.
  • Results indicate LowDiff's effectiveness in both pixel space and latent space applications.

Cite This Study

Xu et al. (2025) studied this question.

synapsesocial.com/papers/68de5da283cbc991d0a20736https://doi.org/10.48550/arxiv.2509.15342
View Full Paper
Ask AI
Bookmark
Share