Efficient Self-supervised Vision Transformers for Representation Learning

ICLR 2022 |

Publication

This paper investigates two techniques for developing efficient self-supervised vision transformers (EsViT) for visual representation learning. First, we show through a comprehensive empirical study that multi-stage architectures with sparse self-attentions can significantly reduce modeling complexity but with a cost of losing the ability to capture fine-grained correspondences between image regions. Second, we propose a new pre-training task of region matching which allows the model to capture fine-grained region dependencies and as a result significantly improves the quality of the learned vision representations. Our results show that combining the two techniques, EsViT achieves 81.3% top-1 on the ImageNet linear probe evaluation, outperforming prior arts with around an order magnitude of higher throughput. When transferring to downstream linear classification tasks, EsViT outperforms its supervised counterpart on 17 out of 18 datasets. The code and models will be publicly available.

论文与出版物下载

Efficient Self-Supervised Vision Transformers (EsViT)

9 7 月, 2021

This is a research project in exploring self-supervised learning (SSL) for computer vision. It aims to learn general-purpose image features from raw pixels without relying on manual supervisions, and the learned networks serve as the backbone of various downstream tasks. Aiming to improve the efficiency of Transformer-based SSL, this project presents Efficient self-supervised Vision Transformers (EsViT), by using a multi-stage architecture and a region-based pre-training task for unsupervised representation learning.