Diffusion-based Latent Image Generation with Label Conditioning: A Score-Based Approach

Abstract
We present a novel score-based strategy for image generation using label conditioning, diffusion models, and latent space modeling. By integrating label information into the diffusion models and iteratively revising the latent space with score estimation, our method enables the generation of high-quality images that correspond to specific labels. Through extensive experimentation, we demonstrate that our method is effective at producing realistic images with precise label control and fine-grained details. This score-based method not only provides a flexible framework for label-guided image synthesis, but it also opens up new opportunities in computer vision, artistic design, and creative applications. The combination of label conditioning, diffusion models, and latent space modeling allows us to stretch the limits of image generation and achieve impressive visual outcomes.
Description
Keywords
Citation
Department Name
Electrical and Computer Engineering
Publisher
North South University
Printed Thesis
DOI
ISSN
ISBN