OPERA: Omni-Supervised Representation Learning with Hierarchical Supervisions

TL;DR

OPERA introduces a hierarchical supervision framework that combines self-supervised and fully supervised learning to improve transferability and performance across various vision tasks.

Contribution

It presents a unified framework for integrating self and full supervision via hierarchical proxy representations, enhancing model training with combined supervision signals.

Findings

OPERA outperforms existing methods in image classification.

It improves segmentation and object detection results.

Hierarchical supervision enhances transferability of learned representations.

Abstract

The pretrain-finetune paradigm in modern computer vision facilitates the success of self-supervised learning, which tends to achieve better transferability than supervised learning. However, with the availability of massive labeled data, a natural question emerges: how to train a better model with both self and full supervision signals? In this paper, we propose Omni-suPErvised Representation leArning with hierarchical supervisions (OPERA) as a solution. We provide a unified perspective of supervisions from labeled and unlabeled data and propose a unified framework of fully supervised and self-supervised learning. We extract a set of hierarchical proxy representations for each image and impose self and full supervisions on the corresponding proxy representations. Extensive experiments on both convolutional neural networks and vision transformers demonstrate the superiority of OPERA in image classification, segmentation, and object detection. Code is available at: https://github.com/wangck20/OPERA.