ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2302.12469
20
22

Unsupervised Discovery of Semantic Latent Directions in Diffusion Models

24 February 2023
Yong-Hyun Park
Mingi Kwon
Junghyo Jo
Youngjung Uh
    DiffM
ArXivPDFHTML
Abstract

Despite the success of diffusion models (DMs), we still lack a thorough understanding of their latent space. While image editing with GANs builds upon latent space, DMs rely on editing the conditions such as text prompts. We present an unsupervised method to discover interpretable editing directions for the latent variables xt∈X\mathbf{x}_t \in \mathcal{X}xt​∈X of DMs. Our method adopts Riemannian geometry between X\mathcal{X}X and the intermediate feature maps H\mathcal{H}H of the U-Nets to provide a deep understanding over the geometrical structure of X\mathcal{X}X. The discovered semantic latent directions mostly yield disentangled attribute changes, and they are globally consistent across different samples. Furthermore, editing in earlier timesteps edits coarse attributes, while ones in later timesteps focus on high-frequency details. We define the curvedness of a line segment between samples to show that X\mathcal{X}X is a curved manifold. Experiments on different baselines and datasets demonstrate the effectiveness of our method even on Stable Diffusion. Our source code will be publicly available for the future researchers.

View on arXiv
Comments on this paper