Fast Autoregressive Video Generation with Diagonal Decoding

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ye, Yang, Guo, Junliang, Wu, Haoyu, He, Tianyu, Pearce, Tim, Rashid, Tabish, Hofmann, Katja, Bian, Jiang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916655455535104
author Ye, Yang
Guo, Junliang
Wu, Haoyu
He, Tianyu
Pearce, Tim
Rashid, Tabish
Hofmann, Katja
Bian, Jiang
author_facet Ye, Yang
Guo, Junliang
Wu, Haoyu
He, Tianyu
Pearce, Tim
Rashid, Tabish
Hofmann, Katja
Bian, Jiang
contents Autoregressive Transformer models have demonstrated impressive performance in video generation, but their sequential token-by-token decoding process poses a major bottleneck, particularly for long videos represented by tens of thousands of tokens. In this paper, we propose Diagonal Decoding (DiagD), a training-free inference acceleration algorithm for autoregressively pre-trained models that exploits spatial and temporal correlations in videos. Our method generates tokens along diagonal paths in the spatial-temporal token grid, enabling parallel decoding within each frame as well as partially overlapping across consecutive frames. The proposed algorithm is versatile and adaptive to various generative models and tasks, while providing flexible control over the trade-off between inference speed and visual quality. Furthermore, we propose a cost-effective finetuning strategy that aligns the attention patterns of the model with our decoding order, further mitigating the training-inference gap on small-scale models. Experiments on multiple autoregressive video generation models and datasets demonstrate that DiagD achieves up to $10\times$ speedup compared to naive sequential decoding, while maintaining comparable visual fidelity.
format Preprint
id arxiv_https___arxiv_org_abs_2503_14070
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Fast Autoregressive Video Generation with Diagonal Decoding
Ye, Yang
Guo, Junliang
Wu, Haoyu
He, Tianyu
Pearce, Tim
Rashid, Tabish
Hofmann, Katja
Bian, Jiang
Computer Vision and Pattern Recognition
Artificial Intelligence
Autoregressive Transformer models have demonstrated impressive performance in video generation, but their sequential token-by-token decoding process poses a major bottleneck, particularly for long videos represented by tens of thousands of tokens. In this paper, we propose Diagonal Decoding (DiagD), a training-free inference acceleration algorithm for autoregressively pre-trained models that exploits spatial and temporal correlations in videos. Our method generates tokens along diagonal paths in the spatial-temporal token grid, enabling parallel decoding within each frame as well as partially overlapping across consecutive frames. The proposed algorithm is versatile and adaptive to various generative models and tasks, while providing flexible control over the trade-off between inference speed and visual quality. Furthermore, we propose a cost-effective finetuning strategy that aligns the attention patterns of the model with our decoding order, further mitigating the training-inference gap on small-scale models. Experiments on multiple autoregressive video generation models and datasets demonstrate that DiagD achieves up to $10\times$ speedup compared to naive sequential decoding, while maintaining comparable visual fidelity.
title Fast Autoregressive Video Generation with Diagonal Decoding
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2503.14070