Three Guidelines You Should Know for Universally Slimmable Self-Supervised Learning

TL;DR

US3L introduces a novel approach for self-supervised learning that maintains accuracy-efficiency balance across diverse devices by ensuring temporal consistency and employing dynamic strategies, validated on multiple vision tasks.

Contribution

The paper presents three guidelines for loss design to ensure temporal consistency in universally slimmable SSL, along with dynamic sampling and group regularization strategies, enabling effective training across models.

Findings

Outperforms state-of-the-art methods on recognition, detection, and segmentation benchmarks.

Achieves better accuracy-efficiency trade-offs with a single training process.

Validates effectiveness on both CNNs and vision transformers.

Abstract

We propose universally slimmable self-supervised learning (dubbed as US3L) to achieve better accuracy-efficiency trade-offs for deploying self-supervised models across different devices. We observe that direct adaptation of self-supervised learning (SSL) to universally slimmable networks misbehaves as the training process frequently collapses. We then discover that temporal consistent guidance is the key to the success of SSL for universally slimmable networks, and we propose three guidelines for the loss design to ensure this temporal consistency from a unified gradient perspective. Moreover, we propose dynamic sampling and group regularization strategies to simultaneously improve training efficiency and accuracy. Our US3L method has been empirically validated on both convolutional neural networks and vision transformers. With only once training and one copy of weights, our method outperforms various state-of-the-art methods (individually trained or not) on benchmarks including recognition, object detection and instance segmentation. Our code is available at https://github.com/megvii-research/US3L-CVPR2023.