In recent years, the rapid development of deep learning approaches has paved the way to explore the underlying factors that explain the data. In particular, several methods have been proposed to learn to identify and disentangle these underlying explanatory factors in order to improve the learning process and model generalization. However, extracting this representation with little or no supervision remains a key challenge in machine learning. In this paper, we provide a theoretical outlook on recent advances in the field of unsupervised representation learning with a focus on auto-encoding-based approaches and on the most well-known supervised disentanglement metrics. We cover the current state-of-the-art methods for learning disentangled representation in an unsupervised manner while pointing out the connection between each method and its added value on disentanglement. Further, we discuss how to quantify disentanglement and present an in-depth analysis of associated metrics. We conclude by carrying out a comparative evaluation of these metrics according to three criteria, (i) modularity, (ii) compactness and (iii) informativeness. Finally, we show that only the Mutual Information Gap score (MIG) meets all three criteria.

Over the past decade, several approaches have been proposed to learn disentangled representations for video prediction. However, reported experiments are mostly based on standard benchmark datasets such as Moving MNIST and Bouncing Balls. In this work, we address the problem of learning disentangled representation for video prediction in an industrial environment. To this end, we use decompositional disentangled variational autoencoder, a deep generative model that aims to decompose and recognize overlapped boxes on a pallet. Specifically, this approach disentangles each frame into a dynamic component (box appearance) and a temporally variant component (box location). We evaluate this approach on a new dataset, which contains 40000 video sequences. The experimental results demonstrate the ability to learn both the decomposition of the bounding boxes and their reconstruction without explicit supervision.