Masked Siamese Networks for Label-Efficient Learning
This addresses the need for scalable, high-performance self-supervised learning in image classification, particularly for low-shot scenarios, representing a strong incremental advance in joint-embedding architectures.
The paper tackles the problem of label-efficient learning in computer vision by proposing Masked Siamese Networks (MSN), a self-supervised framework that matches representations of masked and unmasked image patches, achieving 75.7% top-1 accuracy on ImageNet-1K with only 1% labels.
We propose Masked Siamese Networks (MSN), a self-supervised learning framework for learning image representations. Our approach matches the representation of an image view containing randomly masked patches to the representation of the original unmasked image. This self-supervised pre-training strategy is particularly scalable when applied to Vision Transformers since only the unmasked patches are processed by the network. As a result, MSNs improve the scalability of joint-embedding architectures, while producing representations of a high semantic level that perform competitively on low-shot image classification. For instance, on ImageNet-1K, with only 5,000 annotated images, our base MSN model achieves 72.4% top-1 accuracy, and with 1% of ImageNet-1K labels, we achieve 75.7% top-1 accuracy, setting a new state-of-the-art for self-supervised learning on this benchmark. Our code is publicly available.