Search

Contrastive learning predicts whether two images belong to the same category by training a model to make their feature representations as close (positive samples) or as far away as possible (negative samples). Selecting appropriate samples is critical to effectively train a model, however, existing methods suffer from false or uninformative sample problems. This paper rethinks how to mine samples in contrastive learning and the proposed method is more comprehensive. It takes into account both positive and negative samples, and mines potential samples from two aspects. First, for positive samples, this paper incorporates both the augmented sample views and the mined sample views. A weighted combination of these positive samples is achieved by using both hard and soft weighting strategies simultaneously. Second, considering the existence of false and uninformative negative samples, this paper analyzes the negative samples from the perspective of gradient and mines negative samples that are neither too difficult nor too easy as potential negative samples, i.e., those negative samples that are close to positive samples. Compared with previous state-of-the-art self-supervised methods, experiments show the obvious advantages of the proposed method, and the corresponding top-1 accuracies of linear classification are improved by 0.77%, 2.39%, and 1.01% on CIFAR10, CIFAR100, and TinyImageNet, respectively. Source code and pretrained models are available at https://github.com/dhkdhk/PSM. [ABSTRACT FROM AUTHOR]

Self-supervised methods based on contrastive learning have achieved great success in unsupervised visual representation learning. However, most methods under this framework suffer from the problem of false negative samples. Inspired by the mean shift for self-supervised learning, we propose a new simple framework, namely Multiple Sample Views and Queues (MSVQ). We jointly construct three soft labels on-the-fly by utilizing two complementary and symmetric approaches: multiple augmented positive views and two momentum encoders that generate various semantic features for negative samples. Two teacher networks perform similarity relationship calculations with negative samples and then transfer this knowledge to the student network. Let the student network mimic the similarity relationships between the samples, thus giving the student network a more flexible ability to identify false negative samples in the dataset. The classification results on four benchmark image datasets demonstrate the high effectiveness and efficiency of our approach compared to some classical methods. Source code and pretrained models are available here. [ABSTRACT FROM AUTHOR]