Constant Acceleration Flow

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Park, Dogyun, Lee, Sojin, Kim, Sihyeon, Lee, Taehoon, Hong, Youngjoon, Kim, Hyunwoo J.
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866913569582350336
author Park, Dogyun
Lee, Sojin
Kim, Sihyeon
Lee, Taehoon
Hong, Youngjoon
Kim, Hyunwoo J.
author_facet Park, Dogyun
Lee, Sojin
Kim, Sihyeon
Lee, Taehoon
Hong, Youngjoon
Kim, Hyunwoo J.
contents Rectified flow and reflow procedures have significantly advanced fast generation by progressively straightening ordinary differential equation (ODE) flows. They operate under the assumption that image and noise pairs, known as couplings, can be approximated by straight trajectories with constant velocity. However, we observe that modeling with constant velocity and using reflow procedures have limitations in accurately learning straight trajectories between pairs, resulting in suboptimal performance in few-step generation. To address these limitations, we introduce Constant Acceleration Flow (CAF), a novel framework based on a simple constant acceleration equation. CAF introduces acceleration as an additional learnable variable, allowing for more expressive and accurate estimation of the ODE flow. Moreover, we propose two techniques to further improve estimation accuracy: initial velocity conditioning for the acceleration model and a reflow process for the initial velocity. Our comprehensive studies on toy datasets, CIFAR-10, and ImageNet 64x64 demonstrate that CAF outperforms state-of-the-art baselines for one-step generation. We also show that CAF dramatically improves few-step coupling preservation and inversion over Rectified flow. Code is available at \href{https://github.com/mlvlab/CAF}{https://github.com/mlvlab/CAF}.
format Preprint
id arxiv_https___arxiv_org_abs_2411_00322
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Constant Acceleration Flow
Park, Dogyun
Lee, Sojin
Kim, Sihyeon
Lee, Taehoon
Hong, Youngjoon
Kim, Hyunwoo J.
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Rectified flow and reflow procedures have significantly advanced fast generation by progressively straightening ordinary differential equation (ODE) flows. They operate under the assumption that image and noise pairs, known as couplings, can be approximated by straight trajectories with constant velocity. However, we observe that modeling with constant velocity and using reflow procedures have limitations in accurately learning straight trajectories between pairs, resulting in suboptimal performance in few-step generation. To address these limitations, we introduce Constant Acceleration Flow (CAF), a novel framework based on a simple constant acceleration equation. CAF introduces acceleration as an additional learnable variable, allowing for more expressive and accurate estimation of the ODE flow. Moreover, we propose two techniques to further improve estimation accuracy: initial velocity conditioning for the acceleration model and a reflow process for the initial velocity. Our comprehensive studies on toy datasets, CIFAR-10, and ImageNet 64x64 demonstrate that CAF outperforms state-of-the-art baselines for one-step generation. We also show that CAF dramatically improves few-step coupling preservation and inversion over Rectified flow. Code is available at \href{https://github.com/mlvlab/CAF}{https://github.com/mlvlab/CAF}.
title Constant Acceleration Flow
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2411.00322