Stochastic positional embeddings improve masked image modeling

TL;DR

This paper introduces stochastic positional embeddings (StoP) in masked image modeling to incorporate location uncertainty, resulting in more robust feature learning and improved downstream task performance.

Contribution

The paper proposes stochastic positional embeddings (StoP) that incorporate location uncertainty into MIM, enhancing robustness and accuracy in learned representations.

Findings

StoP improves ImageNet linear probing accuracy by 1.7% with ViT-B.

StoP achieves 2.5% improvement with ViT-H using only 1% of data.

StoP reduces overfitting to location features in masked image modeling.

Abstract

Masked Image Modeling (MIM) is a promising self-supervised learning approach that enables learning from unlabeled images. Despite its recent success, learning good representations through MIM remains challenging because it requires predicting the right semantic content in accurate locations. For example, given an incomplete picture of a dog, we can guess that there is a tail, but we cannot determine its exact location. In this work, we propose to incorporate location uncertainty into MIM by using stochastic positional embeddings (StoP). Specifically, we condition the model on stochastic masked token positions drawn from a Gaussian distribution. StoP reduces overfitting to location features and guides the model toward learning features that are more robust to location uncertainties. Quantitatively, StoP improves downstream MIM performance on a variety of downstream tasks, including $+1.7\%$ on ImageNet linear probing using ViT-B, and $+2.5\%$ for ViT-H using $1\%$ of the data.