Connecting Joint-Embedding Predictive Architecture with Contrastive Self-supervised Learning

TL;DR

This paper introduces C-JEPA, a novel framework combining JEPA with VICReg to address collapse issues and improve visual representation learning, showing enhanced stability and performance on ImageNet-1K.

Contribution

The paper proposes C-JEPA, integrating contrastive regularization with JEPA to prevent collapse and improve learning stability, advancing unsupervised visual representation methods.

Findings

C-JEPA outperforms previous JEPA variants in stability and quality.

Pre-training on ImageNet-1K shows faster convergence and better accuracy.

C-JEPA effectively prevents collapse and learns better feature representations.

Abstract

In recent advancements in unsupervised visual representation learning, the Joint-Embedding Predictive Architecture (JEPA) has emerged as a significant method for extracting visual features from unlabeled imagery through an innovative masking strategy. Despite its success, two primary limitations have been identified: the inefficacy of Exponential Moving Average (EMA) from I-JEPA in preventing entire collapse and the inadequacy of I-JEPA prediction in accurately learning the mean of patch representations. Addressing these challenges, this study introduces a novel framework, namely C-JEPA (Contrastive-JEPA), which integrates the Image-based Joint-Embedding Predictive Architecture with the Variance-Invariance-Covariance Regularization (VICReg) strategy. This integration is designed to effectively learn the variance/covariance for preventing entire collapse and ensuring invariance in the mean of augmented views, thereby overcoming the identified limitations. Through empirical and theoretical evaluations, our work demonstrates that C-JEPA significantly enhances the stability and quality of visual representation learning. When pre-trained on the ImageNet-1K dataset, C-JEPA exhibits rapid and improved convergence in both linear probing and fine-tuning performance metrics.