Saved in:
Bibliographic Details
Main Authors: So, Junhyuk, Lee, Jungwon, Park, Eunhyeok
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2312.03517
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917764686413824
author So, Junhyuk
Lee, Jungwon
Park, Eunhyeok
author_facet So, Junhyuk
Lee, Jungwon
Park, Eunhyeok
contents The substantial computational costs of diffusion models, especially due to the repeated denoising steps necessary for high-quality image generation, present a major obstacle to their widespread adoption. While several studies have attempted to address this issue by reducing the number of score function evaluations (NFE) using advanced ODE solvers without fine-tuning, the decreased number of denoising iterations misses the opportunity to update fine details, resulting in noticeable quality degradation. In our work, we introduce an advanced acceleration technique that leverages the temporal redundancy inherent in diffusion models. Reusing feature maps with high temporal similarity opens up a new opportunity to save computation resources without compromising output quality. To realize the practical benefits of this intuition, we conduct an extensive analysis and propose a novel method, FRDiff. FRDiff is designed to harness the advantages of both reduced NFE and feature reuse, achieving a Pareto frontier that balances fidelity and latency trade-offs in various generative tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2312_03517
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle FRDiff : Feature Reuse for Universal Training-free Acceleration of Diffusion Models
So, Junhyuk
Lee, Jungwon
Park, Eunhyeok
Computer Vision and Pattern Recognition
Artificial Intelligence
The substantial computational costs of diffusion models, especially due to the repeated denoising steps necessary for high-quality image generation, present a major obstacle to their widespread adoption. While several studies have attempted to address this issue by reducing the number of score function evaluations (NFE) using advanced ODE solvers without fine-tuning, the decreased number of denoising iterations misses the opportunity to update fine details, resulting in noticeable quality degradation. In our work, we introduce an advanced acceleration technique that leverages the temporal redundancy inherent in diffusion models. Reusing feature maps with high temporal similarity opens up a new opportunity to save computation resources without compromising output quality. To realize the practical benefits of this intuition, we conduct an extensive analysis and propose a novel method, FRDiff. FRDiff is designed to harness the advantages of both reduced NFE and feature reuse, achieving a Pareto frontier that balances fidelity and latency trade-offs in various generative tasks.
title FRDiff : Feature Reuse for Universal Training-free Acceleration of Diffusion Models
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2312.03517