Big Transfer (BiT): General Visual Representation Learning

TL;DR

This paper introduces Big Transfer (BiT), a simple yet effective approach for large-scale pre-training and transfer learning in vision tasks, achieving state-of-the-art results across diverse datasets and data regimes.

Contribution

The paper presents BiT, a scalable pre-training recipe that significantly improves transfer learning performance across various vision datasets and data sizes.

Findings

BiT achieves 87.5% top-1 accuracy on ILSVRC-2012.

BiT attains 99.4% accuracy on CIFAR-10.

BiT performs well even with very limited data, such as 10 examples per class.

Abstract

Transfer of pre-trained representations improves sample efficiency and simplifies hyperparameter tuning when training deep neural networks for vision. We revisit the paradigm of pre-training on large supervised datasets and fine-tuning the model on a target task. We scale up pre-training, and propose a simple recipe that we call Big Transfer (BiT). By combining a few carefully selected components, and transferring using a simple heuristic, we achieve strong performance on over 20 datasets. BiT performs well across a surprisingly wide range of data regimes -- from 1 example per class to 1M total examples. BiT achieves 87.5% top-1 accuracy on ILSVRC-2012, 99.4% on CIFAR-10, and 76.3% on the 19 task Visual Task Adaptation Benchmark (VTAB). On small datasets, BiT attains 76.8% on ILSVRC-2012 with 10 examples per class, and 97.0% on CIFAR-10 with 10 examples per class. We conduct detailed analysis of the main components that lead to high transfer performance.