Contrast to Divide: Self-Supervised Pre-Training for Learning with Noisy Labels

TL;DR

This paper introduces C2D, a self-supervised pre-training framework that enhances learning with noisy labels by improving feature quality and reducing warm-up time, leading to significant performance gains especially under high noise conditions.

Contribution

The paper proposes a novel self-supervised pre-training method called Contrast to Divide (C2D) that addresses the warm-up obstacle in noisy label learning, improving robustness and accuracy.

Findings

C2D boosts performance in high noise regimes, e.g., over 27% improvement on CIFAR-100 with 90% noise.

C2D outperforms previous methods on WebVision and ImageNet, with 3% higher top-1 accuracy.

Self-supervised pre-training reduces warm-up duration and noise susceptibility in label-noise learning.

Abstract

The success of learning with noisy labels (LNL) methods relies heavily on the success of a warm-up stage where standard supervised training is performed using the full (noisy) training set. In this paper, we identify a "warm-up obstacle": the inability of standard warm-up stages to train high quality feature extractors and avert memorization of noisy labels. We propose "Contrast to Divide" (C2D), a simple framework that solves this problem by pre-training the feature extractor in a self-supervised fashion. Using self-supervised pre-training boosts the performance of existing LNL approaches by drastically reducing the warm-up stage's susceptibility to noise level, shortening its duration, and improving extracted feature quality. C2D works out of the box with existing methods and demonstrates markedly improved performance, especially in the high noise regime, where we get a boost of more than 27% for CIFAR-100 with 90% noise over the previous state of the art. In real-life noise settings, C2D trained on mini-WebVision outperforms previous works both in WebVision and ImageNet validation sets by 3% top-1 accuracy. We perform an in-depth analysis of the framework, including investigating the performance of different pre-training approaches and estimating the effective upper bound of the LNL performance with semi-supervised learning. Code for reproducing our experiments is available at https://github.com/ContrastToDivide/C2D