28
9

SemPPL: Predicting pseudo-labels for better contrastive representations

Abstract

Learning from large amounts of unsupervised data and a small amount of supervision is an important open problem in computer vision. We propose a new semi-supervised learning method, Semantic Positives via Pseudo-Labels (SemPPL), that combines labelled and unlabelled data to learn informative representations. Our method extends self-supervised contrastive learning -- where representations are shaped by distinguishing whether two samples represent the same underlying datum (positives) or not (negatives) -- with a novel approach to selecting positives. To enrich the set of positives, we leverage the few existing ground-truth labels to predict the missing ones through a kk-nearest neighbours classifier by using the learned embeddings of the labelled data. We thus extend the set of positives with datapoints having the same pseudo-label and call these semantic positives. We jointly learn the representation and predict bootstrapped pseudo-labels. This creates a reinforcing cycle. Strong initial representations enable better pseudo-label predictions which then improve the selection of semantic positives and lead to even better representations. SemPPL outperforms competing semi-supervised methods setting new state-of-the-art performance of 68.5%68.5\% and 76%76\% top-11 accuracy when using a ResNet-5050 and training on 1%1\% and 10%10\% of labels on ImageNet, respectively. Furthermore, when using selective kernels, SemPPL significantly outperforms previous state-of-the-art achieving 72.3%72.3\% and 78.3%78.3\% top-11 accuracy on ImageNet with 1%1\% and 10%10\% labels, respectively, which improves absolute +7.8%+7.8\% and +6.2%+6.2\% over previous work. SemPPL also exhibits state-of-the-art performance over larger ResNet models as well as strong robustness, out-of-distribution and transfer performance. We release the checkpoints and the evaluation code at https://github.com/deepmind/semppl .

View on arXiv
Comments on this paper