Separators in Enhancing Autoregressive Pretraining for Vision Mamba

TL;DR

This paper introduces STAR, a separator-based method that significantly extends the sequence length for Vision Mamba pretraining, leading to improved accuracy on ImageNet-1k by better capturing long-range dependencies.

Contribution

The paper proposes a novel separator technique called STAR that enables longer sequence pretraining for Vision Mamba, enhancing its ability to model extended visual data.

Findings

STAR allows quadrupling of input sequence length.

STAR-B achieved 83.5% accuracy on ImageNet-1k.

The method improves long-range dependency modeling in vision models.

Abstract

The state space model Mamba has recently emerged as a promising paradigm in computer vision, attracting significant attention due to its efficient processing of long sequence tasks. Mamba's inherent causal mechanism renders it particularly suitable for autoregressive pretraining. However, current autoregressive pretraining methods are constrained to short sequence tasks, failing to fully exploit Mamba's prowess in handling extended sequences. To address this limitation, we introduce an innovative autoregressive pretraining method for Vision Mamba that substantially extends the input sequence length. We introduce new \textbf{S}epara\textbf{T}ors for \textbf{A}uto\textbf{R}egressive pretraining to demarcate and differentiate between different images, known as \textbf{STAR}. Specifically, we insert identical separators before each image to demarcate its inception. This strategy enables us to quadruple the input sequence length of Vision Mamba while preserving the original dimensions of the dataset images. Employing this long sequence pretraining technique, our STAR-B model achieved an impressive accuracy of 83.5\% on ImageNet-1k, which is highly competitive in Vision Mamba. These results underscore the potential of our method in enhancing the performance of vision models through improved leveraging of long-range dependencies.