RandAR: Decoder-only Autoregressive Visual Generation in Random Orders

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Pang, Ziqi, Zhang, Tianyuan, Luan, Fujun, Man, Yunze, Tan, Hao, Zhang, Kai, Freeman, William T., Wang, Yu-Xiong
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915375252242432
author Pang, Ziqi
Zhang, Tianyuan
Luan, Fujun
Man, Yunze
Tan, Hao
Zhang, Kai
Freeman, William T.
Wang, Yu-Xiong
author_facet Pang, Ziqi
Zhang, Tianyuan
Luan, Fujun
Man, Yunze
Tan, Hao
Zhang, Kai
Freeman, William T.
Wang, Yu-Xiong
contents We introduce RandAR, a decoder-only visual autoregressive (AR) model capable of generating images in arbitrary token orders. Unlike previous decoder-only AR models that rely on a predefined generation order, RandAR removes this inductive bias, unlocking new capabilities in decoder-only generation. Our essential design enables random order by inserting a "position instruction token" before each image token to be predicted, representing the spatial location of the next image token. Trained on randomly permuted token sequences -- a more challenging task than fixed-order generation, RandAR achieves comparable performance to its conventional raster-order counterpart. More importantly, decoder-only transformers trained from random orders acquire new capabilities. For the efficiency bottleneck of AR models, RandAR adopts parallel decoding with KV-Cache at inference time, enjoying 2.5x acceleration without sacrificing generation quality. Additionally, RandAR supports inpainting, outpainting and resolution extrapolation in a zero-shot manner. We hope RandAR inspires new directions for decoder-only visual generation models and broadens their applications across diverse scenarios. Our project page is at https://rand-ar.github.io/.
format Preprint
id arxiv_https___arxiv_org_abs_2412_01827
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle RandAR: Decoder-only Autoregressive Visual Generation in Random Orders
Pang, Ziqi
Zhang, Tianyuan
Luan, Fujun
Man, Yunze
Tan, Hao
Zhang, Kai
Freeman, William T.
Wang, Yu-Xiong
Computer Vision and Pattern Recognition
Artificial Intelligence
We introduce RandAR, a decoder-only visual autoregressive (AR) model capable of generating images in arbitrary token orders. Unlike previous decoder-only AR models that rely on a predefined generation order, RandAR removes this inductive bias, unlocking new capabilities in decoder-only generation. Our essential design enables random order by inserting a "position instruction token" before each image token to be predicted, representing the spatial location of the next image token. Trained on randomly permuted token sequences -- a more challenging task than fixed-order generation, RandAR achieves comparable performance to its conventional raster-order counterpart. More importantly, decoder-only transformers trained from random orders acquire new capabilities. For the efficiency bottleneck of AR models, RandAR adopts parallel decoding with KV-Cache at inference time, enjoying 2.5x acceleration without sacrificing generation quality. Additionally, RandAR supports inpainting, outpainting and resolution extrapolation in a zero-shot manner. We hope RandAR inspires new directions for decoder-only visual generation models and broadens their applications across diverse scenarios. Our project page is at https://rand-ar.github.io/.
title RandAR: Decoder-only Autoregressive Visual Generation in Random Orders
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2412.01827