Scalable Autoregressive Image Generation with Mamba

TL;DR

AiM introduces a scalable autoregressive image generation model based on the Mamba architecture, achieving superior quality and faster inference by leveraging long-sequence modeling without complex 2D adaptations.

Contribution

The paper presents AiM, a novel autoregressive image generator using Mamba, which simplifies 2D image modeling and outperforms existing AR models in quality and speed.

Findings

Achieves state-of-the-art FID of 2.21 on ImageNet1K 256x256.

Offers models ranging from 148M to 1.3B parameters.

Demonstrates 2 to 10 times faster inference than diffusion models.

Abstract

We introduce AiM, an autoregressive (AR) image generative model based on Mamba architecture. AiM employs Mamba, a novel state-space model characterized by its exceptional performance for long-sequence modeling with linear time complexity, to supplant the commonly utilized Transformers in AR image generation models, aiming to achieve both superior generation quality and enhanced inference speed. Unlike existing methods that adapt Mamba to handle two-dimensional signals via multi-directional scan, AiM directly utilizes the next-token prediction paradigm for autoregressive image generation. This approach circumvents the need for extensive modifications to enable Mamba to learn 2D spatial representations. By implementing straightforward yet strategically targeted modifications for visual generative tasks, we preserve Mamba's core structure, fully exploiting its efficient long-sequence modeling capabilities and scalability. We provide AiM models in various scales, with parameter counts ranging from 148M to 1.3B. On the ImageNet1K 256*256 benchmark, our best AiM model achieves a FID of 2.21, surpassing all existing AR models of comparable parameter counts and demonstrating significant competitiveness against diffusion models, with 2 to 10 times faster inference speed. Code is available at https://github.com/hp-l33/AiM