The representation of multimedia content experienced tremendous advances in the last decades. Mainly supported by deep learning models, impressive results have been obtained. However, despite such advances in representation, the definition of similarity has been neglected. Effectively computing the similarity between representations remains a challenge. Traditional distance functions, such as the Euclidean distance, are not able to properly consider the relevant similarity information encoded in the dataset manifold. In fact, manifolds are essential to perception in many scenarios, such that exploiting the underlying structure of dataset manifolds plays a central role in multimedia content understanding and retrieval. In this paper, we present a framework for unsupervised distance learning which provides easy and uniform access to methods capable of considering the dataset manifold for redefining similarity. Such methods perform context-sensitive similarity learning based on more global measures, capable of improving the effectiveness of retrieval and machine learning tasks. The framework can use distance, similarity, or ranking information both as input and output and compute traditional retrieval effectiveness measures. Implemented as a wrapper in Python, the framework allows integration with a large number of Python libraries while keeping a back-end in C++ for efficiency. The paper also discusses diverse applications of the methods available in the pyUDLF framework, including image re-ranking, video retrieval, person re-ID, and pre-processing of distance measurements for clustering and classification.

In spite of the huge advances in supervised learning, the common requirement for extensive labeled datasets represents a severe bottleneck. In this scenario, other learning paradigms capable of addressing the challenge associated with the scarcity of labeled data represent a relevant alternative solution. This paper presents a novel clustering method called Self-Supervised Graph Convolutional Clustering (SGCC)1, which aims to exploit the strengths of different learning paradigms, combining unsupervised, semi-supervised, and self-supervised perspectives. An unsupervised manifold learning algorithm based on hypergraphs and ranking information is used to provide more effective and global similarity information. The hypergraph structures allow identifying representative items for each cluster, which are used to derive a set of small but high-confident clusters. Such clusters are taken as soft-labels for training a Graph Convolutional Network (GCN) in a semi-supervised classification task. Once trained in a self-supervised setting, the GCN is used to predict the cluster of remaining items. The proposed SGCC method was evaluated both in image and citation networks datasets and compared with classic and recent clustering methods, obtaining high-effective results in all scenarios.