Barlow Twins and VICReg are self-supervised representation learning models that use regularizers to decorrelate features. Although these models are as effective as conventional representation learning models, their training can be computationally demanding if the dimension$d$of the projected embeddings is high. As the regularizers are defined in terms of individual elements of a cross-correlation or covariance matrix, computing the loss for$n$samples takes$O(nd^{2})$time. In this paper, we propose a relaxed decorre-lating regularizer that can be computed in$O(nd\log d)$time by Fast Fourier Transform. We also propose an inexpensive technique to mitigate undesirable local minima that develop with the relaxation. The proposed regularizer exhibits accuracy comparable to that of existing regularizers in down-stream tasks, whereas their training requires less memory and is faster for large$d$. The source code is available.11https://github.com/yutaro-s/scalable-decorrelation-ssl.git

A computer-aided diagnosis (CAD) system that characterizes nodules in medical images can help radiologists determine its malignancy. Preparing large volumes of labeled data for CAD systems, however, requires advanced medical knowledge. This makes it extremely difficult to develop such systems, despite their growing demand. In this paper, we propose a new training method to build an image classifier for characterization of nodules utilizing pseudo-labels, i.e., image labels automatically retrieved from radiology reports. A radiology report is a type of record in which radiologists present a summary of lesion characteristics and diagnosis. Labeling radiology reports is much easier than labeling radiology images, and can be done without high expertise. Using several thousand labeled reports, we constructed a hierarchical attention network-based text classifier to assign pseudo-labels of the characteristics of pulmonary nodules with high accuracy (macro F1-score of 0.941). Experimental results show that the image classifier trained with the pseudo-labels can achieve almost the same performance as the one trained with the labels annotated by radiologists: AUC 0.848 for the model trained with the pseudo-labels on 3,000 computed tomography (CT) images and 0.847 for the model trained with the manual labels on 800 CT images.

Numerous practical datasets have been developed to recognize human actions from videos. However, many of them were constructed by collecting videos within a limited domain; thus, a model trained using one of the existing datasets often fails to classify videos in a different domain accurately. A possible solution for this drawback is to enhance the domain of each action label, i.e., to import videos associated with a given action label from the other datasets, and then, to train a model using the enhanced dataset. To realize this solution, we constructed a meta video dataset from the existing datasets for human action recognition, referred to as MetaVD. MetaVD comprises six popular human action recognition datasets, which we integrated by annotating 568,015 relation labels in total. These relation labels reflect equality, similarity, and hierarchy between action labels of the original datasets. We further present simple yet effective dataset enhancement methods using MetaVD, which are useful for training models with higher generalization performance, as established by experiments on human action classification. As a further contribution of MetaVD, we show that its analysis can provide useful insight into the datasets.

In recent years, automatic video caption generation has attracted considerable attention. This paper focuses on the generation of Japanese captions for describing human actions. While most currently available video caption datasets have been constructed for English, there is no equivalent Japanese dataset. To address this, we constructed a large-scale Japanese video caption dataset consisting of 79,822 videos and 399,233 captions. Each caption in our dataset describes a video in the form of “who does what and where.” To describe human actions, it is important to identify the details of a person, place, and action. Indeed, when we describe human actions, we usually mention the scene, person, and action. In our experiments, we evaluated two caption generation methods to obtain benchmark results. Further, we investigated whether those generation methods could specify “who does what and where.”