On Inference Stability for Diffusion Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Nguyen, Viet, Vu, Giang, Thanh, Tung Nguyen, Than, Khoat, Tran, Toan
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910311820296192
author Nguyen, Viet
Vu, Giang
Thanh, Tung Nguyen
Than, Khoat
Tran, Toan
author_facet Nguyen, Viet
Vu, Giang
Thanh, Tung Nguyen
Than, Khoat
Tran, Toan
contents Denoising Probabilistic Models (DPMs) represent an emerging domain of generative models that excel in generating diverse and high-quality images. However, most current training methods for DPMs often neglect the correlation between timesteps, limiting the model's performance in generating images effectively. Notably, we theoretically point out that this issue can be caused by the cumulative estimation gap between the predicted and the actual trajectory. To minimize that gap, we propose a novel \textit{sequence-aware} loss that aims to reduce the estimation gap to enhance the sampling quality. Furthermore, we theoretically show that our proposed loss function is a tighter upper bound of the estimation loss in comparison with the conventional loss in DPMs. Experimental results on several benchmark datasets including CIFAR10, CelebA, and CelebA-HQ consistently show a remarkable improvement of our proposed method regarding the image generalization quality measured by FID and Inception Score compared to several DPM baselines. Our code and pre-trained checkpoints are available at \url{https://github.com/VinAIResearch/SA-DPM}.
format Preprint
id arxiv_https___arxiv_org_abs_2312_12431
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle On Inference Stability for Diffusion Models
Nguyen, Viet
Vu, Giang
Thanh, Tung Nguyen
Than, Khoat
Tran, Toan
Computer Vision and Pattern Recognition
Denoising Probabilistic Models (DPMs) represent an emerging domain of generative models that excel in generating diverse and high-quality images. However, most current training methods for DPMs often neglect the correlation between timesteps, limiting the model's performance in generating images effectively. Notably, we theoretically point out that this issue can be caused by the cumulative estimation gap between the predicted and the actual trajectory. To minimize that gap, we propose a novel \textit{sequence-aware} loss that aims to reduce the estimation gap to enhance the sampling quality. Furthermore, we theoretically show that our proposed loss function is a tighter upper bound of the estimation loss in comparison with the conventional loss in DPMs. Experimental results on several benchmark datasets including CIFAR10, CelebA, and CelebA-HQ consistently show a remarkable improvement of our proposed method regarding the image generalization quality measured by FID and Inception Score compared to several DPM baselines. Our code and pre-trained checkpoints are available at \url{https://github.com/VinAIResearch/SA-DPM}.
title On Inference Stability for Diffusion Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2312.12431