Improving Fine-tuning of Self-supervised Models with Contrastive Initialization

TL;DR

This paper introduces COIN, a contrastive initialization method that enhances semantic information in self-supervised models, leading to improved fine-tuning performance across multiple downstream tasks without extra training cost.

Contribution

The paper proposes a novel contrastive initialization stage that enriches semantic information in self-supervised models before fine-tuning, outperforming existing methods.

Findings

COIN significantly improves downstream task performance.

It outperforms existing methods without additional training cost.

Sets new state-of-the-art results on multiple benchmarks.

Abstract

Self-supervised learning (SSL) has achieved remarkable performance in pretraining the models that can be further used in downstream tasks via fine-tuning. However, these self-supervised models may not capture meaningful semantic information since the images belonging to the same class are always regarded as negative pairs in the contrastive loss. Consequently, the images of the same class are often located far away from each other in learned feature space, which would inevitably hamper the fine-tuning process. To address this issue, we seek to provide a better initialization for the self-supervised models by enhancing the semantic information. To this end, we propose a Contrastive Initialization (COIN) method that breaks the standard fine-tuning pipeline by introducing an extra initialization stage before fine-tuning. Extensive experiments show that, with the enriched semantics, our COIN significantly outperforms existing methods without introducing extra training cost and sets new state-of-the-arts on multiple downstream tasks.