Multi-task Self-Supervised Visual Learning

TL;DR

This paper explores combining multiple self-supervised tasks to learn unified visual representations, demonstrating improved performance across various vision benchmarks with deeper networks and task combination strategies.

Contribution

It provides a comprehensive comparison of self-supervised tasks, introduces methods for task combination and input harmonization, and achieves near-supervised performance without manual labels.

Findings

Deeper networks outperform shallower ones.

Combining multiple self-supervised tasks improves performance.

Naive multi-task training nearly matches supervised pre-training.

Abstract

We investigate methods for combining multiple self-supervised tasks--i.e., supervised tasks where data can be collected without manual labeling--in order to train a single visual representation. First, we provide an apples-to-apples comparison of four different self-supervised tasks using the very deep ResNet-101 architecture. We then combine tasks to jointly train a network. We also explore lasso regularization to encourage the network to factorize the information in its representation, and methods for "harmonizing" network inputs in order to learn a more unified representation. We evaluate all methods on ImageNet classification, PASCAL VOC detection, and NYU depth prediction. Our results show that deeper networks work better, and that combining tasks--even via a naive multi-head architecture--always improves performance. Our best joint network nearly matches the PASCAL performance of a model pre-trained on ImageNet classification, and matches the ImageNet network on NYU depth prediction.