When Does Contrastive Visual Representation Learning Work?

TL;DR

This paper investigates the effectiveness of contrastive self-supervised learning across diverse datasets, revealing key factors influencing success and limitations compared to supervised methods, especially on fine-grained tasks.

Contribution

It provides new insights into data quantity, domain, quality, and task granularity affecting contrastive learning, guiding best practices beyond ImageNet.

Findings

Additional data beyond 500k images offers limited benefits.

Using images from different domains does not improve generalization.

Corrupted images impact supervised and self-supervised learning differently.

Abstract

Recent self-supervised representation learning techniques have largely closed the gap between supervised and unsupervised learning on ImageNet classification. While the particulars of pretraining on ImageNet are now relatively well understood, the field still lacks widely accepted best practices for replicating this success on other datasets. As a first step in this direction, we study contrastive self-supervised learning on four diverse large-scale datasets. By looking through the lenses of data quantity, data domain, data quality, and task granularity, we provide new insights into the necessary conditions for successful self-supervised learning. Our key findings include observations such as: (i) the benefit of additional pretraining data beyond 500k images is modest, (ii) adding pretraining images from another domain does not lead to more general representations, (iii) corrupted pretraining images have a disparate impact on supervised and self-supervised pretraining, and (iv) contrastive learning lags far behind supervised learning on fine-grained visual classification tasks.