Bag of Image Patch Embedding Behind the Success of Self-Supervised Learning

TL;DR

This paper reveals that self-supervised learning primarily captures image patch co-occurrence, and demonstrates that patch-based aggregation methods can match or outperform traditional SSL approaches, providing insights into the underlying principles of image representation learning.

Contribution

It introduces BagSSL, a patch aggregation approach that improves SSL performance and offers a formal connection to co-occurrence modeling, enhancing understanding of SSL mechanisms.

Findings

BagSSL achieves 62% top-1 accuracy on ImageNet with 32x32 patches.

Patch aggregation can enhance state-of-the-art SSL methods.

Local patch representations preserve locality even at invariance scales.

Abstract

Self-supervised learning (SSL) has recently achieved tremendous empirical advancements in learning image representation. However, our understanding of the principle behind learning such a representation is still limited. This work shows that joint-embedding SSL approaches primarily learn a representation of image patches, which reflects their co-occurrence. Such a connection to co-occurrence modeling can be established formally, and it supplements the prevailing invariance perspective. We empirically show that learning a representation for fixed-scale patches and aggregating local patch representations as the image representation achieves similar or even better results than the baseline methods. We denote this process as BagSSL. Even with 32x32 patch representation, BagSSL achieves 62% top-1 linear probing accuracy on ImageNet. On the other hand, with a multi-scale pretrained model, we show that the whole image embedding is approximately the average of local patch embeddings. While the SSL representation is relatively invariant at the global scale, we show that locality is preserved when we zoom into local patch-level representation. Further, we show that patch representation aggregation can improve various SOTA baseline methods by a large margin. The patch representation is considerably easier to understand, and this work makes a step to demystify self-supervised representation learning.