English

Stochastic positional embeddings improve masked image modeling

Computer Vision and Pattern Recognition 2024-02-28 v2 Artificial Intelligence Machine Learning

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%+1.7\% on ImageNet linear probing using ViT-B, and +2.5%+2.5\% for ViT-H using 1%1\% of the data.

Keywords

Cite

@article{arxiv.2308.00566,
  title  = {Stochastic positional embeddings improve masked image modeling},
  author = {Amir Bar and Florian Bordes and Assaf Shocher and Mahmoud Assran and Pascal Vincent and Nicolas Ballas and Trevor Darrell and Amir Globerson and Yann LeCun},
  journal= {arXiv preprint arXiv:2308.00566},
  year   = {2024}
}

Comments

Code and models available in https://github.com/amirbar/StoP

R2 v1 2026-06-28T11:45:35.425Z