MoPro: Webly Supervised Learning with Momentum Prototypes

TL;DR

MoPro is a contrastive learning method that effectively handles noisy web data, enabling scalable, robust image representation learning without extensive annotations, and outperforms existing models on various tasks.

Contribution

Introduces momentum prototypes (MoPro), a novel contrastive learning approach that corrects label noise and removes out-of-distribution samples in webly-supervised learning.

Findings

Achieves state-of-the-art results on WebVision dataset.

Outperforms ImageNet supervised pretraining on 1-shot VOC classification.

More robust to distribution shifts than previous methods.

Abstract

We propose a webly-supervised representation learning method that does not suffer from the annotation unscalability of supervised learning, nor the computation unscalability of self-supervised learning. Most existing works on webly-supervised representation learning adopt a vanilla supervised learning method without accounting for the prevalent noise in the training data, whereas most prior methods in learning with label noise are less effective for real-world large-scale noisy data. We propose momentum prototypes (MoPro), a simple contrastive learning method that achieves online label noise correction, out-of-distribution sample removal, and representation learning. MoPro achieves state-of-the-art performance on WebVision, a weakly-labeled noisy dataset. MoPro also shows superior performance when the pretrained model is transferred to down-stream image classification and detection tasks. It outperforms the ImageNet supervised pretrained model by +10.5 on 1-shot classification on VOC, and outperforms the best self-supervised pretrained model by +17.3 when finetuned on 1\% of ImageNet labeled samples. Furthermore, MoPro is more robust to distribution shifts. Code and pretrained models are available at https://github.com/salesforce/MoPro.