Saved in:
Bibliographic Details
Main Authors: Gambashidze, Alexander, Kulikov, Anton, Sosnin, Yuriy, Makarov, Ilya
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2406.17636
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909939073548288
author Gambashidze, Alexander
Kulikov, Anton
Sosnin, Yuriy
Makarov, Ilya
author_facet Gambashidze, Alexander
Kulikov, Anton
Sosnin, Yuriy
Makarov, Ilya
contents Recent advancements in human preference optimization, initially developed for Language Models (LMs), have shown promise for text-to-image Diffusion Models, enhancing prompt alignment, visual appeal, and user preference. Unlike LMs, Diffusion Models typically optimize in pixel or VAE space, which does not align well with human perception, leading to slower and less efficient training during the preference alignment stage. We propose using a perceptual objective in the U-Net embedding space of the diffusion model to address these issues. Our approach involves fine-tuning Stable Diffusion 1.5 and XL using Direct Preference Optimization (DPO), Contrastive Preference Optimization (CPO), and supervised fine-tuning (SFT) within this embedding space. This method significantly outperforms standard latent-space implementations across various metrics, including quality and computational cost. For SDXL, our approach provides 60.8\% general preference, 62.2\% visual appeal, and 52.1\% prompt following against original open-sourced SDXL-DPO on the PartiPrompts dataset, while significantly reducing compute. Our approach not only improves the efficiency and quality of human preference alignment for diffusion models but is also easily integrable with other optimization techniques. The training code and LoRA weights will be available here: https://huggingface.co/alexgambashidze/SDXL\_NCP-DPO\_v0.1
format Preprint
id arxiv_https___arxiv_org_abs_2406_17636
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Aligning Diffusion Models with Noise-Conditioned Perception
Gambashidze, Alexander
Kulikov, Anton
Sosnin, Yuriy
Makarov, Ilya
Computer Vision and Pattern Recognition
Artificial Intelligence
Recent advancements in human preference optimization, initially developed for Language Models (LMs), have shown promise for text-to-image Diffusion Models, enhancing prompt alignment, visual appeal, and user preference. Unlike LMs, Diffusion Models typically optimize in pixel or VAE space, which does not align well with human perception, leading to slower and less efficient training during the preference alignment stage. We propose using a perceptual objective in the U-Net embedding space of the diffusion model to address these issues. Our approach involves fine-tuning Stable Diffusion 1.5 and XL using Direct Preference Optimization (DPO), Contrastive Preference Optimization (CPO), and supervised fine-tuning (SFT) within this embedding space. This method significantly outperforms standard latent-space implementations across various metrics, including quality and computational cost. For SDXL, our approach provides 60.8\% general preference, 62.2\% visual appeal, and 52.1\% prompt following against original open-sourced SDXL-DPO on the PartiPrompts dataset, while significantly reducing compute. Our approach not only improves the efficiency and quality of human preference alignment for diffusion models but is also easily integrable with other optimization techniques. The training code and LoRA weights will be available here: https://huggingface.co/alexgambashidze/SDXL\_NCP-DPO\_v0.1
title Aligning Diffusion Models with Noise-Conditioned Perception
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2406.17636