Salience-Based Adaptive Masking: Revisiting Token Dynamics for Enhanced Pre-training

TL;DR

This paper presents Saliency-Based Adaptive Masking (SBAM), a novel approach that improves Masked Image Modeling by dynamically adjusting masking ratios based on token salience, leading to more stable and effective pre-training.

Contribution

Introduces SBAM and AMR strategies that adapt masking ratios per sample, enhancing robustness and performance in Masked Image Modeling.

Findings

Outperforms state-of-the-art in ImageNet-1K pre-training

Provides robustness against masking ratio variations

Enables adaptive, sample-specific masking ratios

Abstract

In this paper, we introduce Saliency-Based Adaptive Masking (SBAM), a novel and cost-effective approach that significantly enhances the pre-training performance of Masked Image Modeling (MIM) approaches by prioritizing token salience. Our method provides robustness against variations in masking ratios, effectively mitigating the performance instability issues common in existing methods. This relaxes the sensitivity of MIM-based pre-training to masking ratios, which in turn allows us to propose an adaptive strategy for `tailored' masking ratios for each data sample, which no existing method can provide. Toward this goal, we propose an Adaptive Masking Ratio (AMR) strategy that dynamically adjusts the proportion of masking for the unique content of each image based on token salience. We show that our method significantly improves over the state-of-the-art in mask-based pre-training on the ImageNet-1K dataset.