Customize Your Visual Autoregressive Recipe with Set Autoregressive Modeling

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Liu, Wenze, Zhuo, Le, Xin, Yi, Xia, Sheng, Gao, Peng, Yue, Xiangyu
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866909347802513408
author Liu, Wenze
Zhuo, Le
Xin, Yi
Xia, Sheng
Gao, Peng
Yue, Xiangyu
author_facet Liu, Wenze
Zhuo, Le
Xin, Yi
Xia, Sheng
Gao, Peng
Yue, Xiangyu
contents We introduce a new paradigm for AutoRegressive (AR) image generation, termed Set AutoRegressive Modeling (SAR). SAR generalizes the conventional AR to the next-set setting, i.e., splitting the sequence into arbitrary sets containing multiple tokens, rather than outputting each token in a fixed raster order. To accommodate SAR, we develop a straightforward architecture termed Fully Masked Transformer. We reveal that existing AR variants correspond to specific design choices of sequence order and output intervals within the SAR framework, with AR and Masked AR (MAR) as two extreme instances. Notably, SAR facilitates a seamless transition from AR to MAR, where intermediate states allow for training a causal model that benefits from both few-step inference and KV cache acceleration, thus leveraging the advantages of both AR and MAR. On the ImageNet benchmark, we carefully explore the properties of SAR by analyzing the impact of sequence order and output intervals on performance, as well as the generalization ability regarding inference order and steps. We further validate the potential of SAR by training a 900M text-to-image model capable of synthesizing photo-realistic images with any resolution. We hope our work may inspire more exploration and application of AR-based modeling across diverse modalities.
format Preprint
id arxiv_https___arxiv_org_abs_2410_10511
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Customize Your Visual Autoregressive Recipe with Set Autoregressive Modeling
Liu, Wenze
Zhuo, Le
Xin, Yi
Xia, Sheng
Gao, Peng
Yue, Xiangyu
Computer Vision and Pattern Recognition
We introduce a new paradigm for AutoRegressive (AR) image generation, termed Set AutoRegressive Modeling (SAR). SAR generalizes the conventional AR to the next-set setting, i.e., splitting the sequence into arbitrary sets containing multiple tokens, rather than outputting each token in a fixed raster order. To accommodate SAR, we develop a straightforward architecture termed Fully Masked Transformer. We reveal that existing AR variants correspond to specific design choices of sequence order and output intervals within the SAR framework, with AR and Masked AR (MAR) as two extreme instances. Notably, SAR facilitates a seamless transition from AR to MAR, where intermediate states allow for training a causal model that benefits from both few-step inference and KV cache acceleration, thus leveraging the advantages of both AR and MAR. On the ImageNet benchmark, we carefully explore the properties of SAR by analyzing the impact of sequence order and output intervals on performance, as well as the generalization ability regarding inference order and steps. We further validate the potential of SAR by training a 900M text-to-image model capable of synthesizing photo-realistic images with any resolution. We hope our work may inspire more exploration and application of AR-based modeling across diverse modalities.
title Customize Your Visual Autoregressive Recipe with Set Autoregressive Modeling
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2410.10511