Stochastic positional embeddings improve masked image modeling
This work addresses a specific bottleneck in self-supervised learning for computer vision, offering incremental improvements to MIM methods.
The paper tackles the challenge of learning good representations in Masked Image Modeling (MIM) by incorporating location uncertainty through stochastic positional embeddings (StoP), resulting in improved downstream performance, such as +1.7% on ImageNet linear probing with ViT-B and +2.5% for ViT-H with 1% of the data.
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.