[Paper Review] Deep Learning Techniques for Future Intelligent Cross-Media Retrieval

Sadaqat Ur Rehman, Muhammad Waqas+6 more|arXiv (Cornell University)|Jul 21, 2020

Advanced Image and Video Retrieval Techniques182 references190 citations

TL;DR

This paper provides a comprehensive survey of deep learning methods for cross-media retrieval, presenting a taxonomy based on representation, alignment, and translation, and reviewing datasets and challenges.

ABSTRACT

With the advancement in technology and the expansion of broadcasting, cross-media retrieval has gained much attention. It plays a significant role in big data applications and consists in searching and finding data from different types of media. In this paper, we provide a novel taxonomy according to the challenges faced by multi-modal deep learning approaches in solving cross-media retrieval, namely: representation, alignment, and translation. These challenges are evaluated on deep learning (DL) based methods, which are categorized into four main groups: 1) unsupervised methods, 2) supervised methods, 3) pairwise based methods, and 4) rank based methods. Then, we present some well-known cross-media datasets used for retrieval, considering the importance of these datasets in the context in of deep learning based cross-media retrieval approaches. Moreover, we also present an extensive review of the state-of-the-art problems and its corresponding solutions for encouraging deep learning in cross-media retrieval. The fundamental objective of this work is to exploit Deep Neural Networks (DNNs) for bridging the "media gap", and provide researchers and developers with a better understanding of the underlying problems and the potential solutions of deep learning assisted cross-media retrieval. To the best of our knowledge, this is the first comprehensive survey to address cross-media retrieval under deep learning methods.

Motivation & Objective

• Propose a taxonomy of challenges in cross-media retrieval focused on representation, alignment, and translation.
• Audit deep learning methods for cross-media retrieval across unsupervised, supervised, pairwise, and rank-based paradigms.
• Review well-known cross-media datasets and their suitability for DL-based retrieval methods.
• Identify current problems, gaps, and future research opportunities in cross-media DL-based retrieval.

Proposed method

• Define a taxonomy of cross-media retrieval challenges: representation, alignment, and translation.
• Categorize DL-based cross-media retrieval methods into four groups: unsupervised, supervised, pairwise, and rank-based.
• Survey cross-media datasets and summarize their characteristics and relevance for DL methods.
• Discuss state-of-the-art problems and proposed DL-based solutions for bridging the media gap.
• Argue for end-to-end DL frameworks and multi-modal representations as enabling factors for cross-media retrieval.

Experimental results

Research questions

• RQ1 What are the key challenges (representation, alignment, translation) in cross-media retrieval when using deep learning?
• RQ2 How do DL-based methods (unsupervised, supervised, pairwise, rank-based) address these challenges?
• RQ3 What datasets best support evaluation and advancement of DL-based cross-media retrieval methods?
• RQ4 What are the primary gaps and future directions for DL-enabled cross-media retrieval?

Key findings

• Introduces a novel taxonomy addressing representation, alignment, and translation in DL-based cross-media retrieval.
• Provides an up-to-date survey of DL approaches across unsupervised, supervised, pairwise, and rank-based methods.
• Offers a detailed review of widely used cross-media datasets and their pros/cons for DL evaluation.
• Highlights state-of-the-art problems and opportunities, guiding future research in cross-media DL retrieval.
• Emphasizes end-to-end DL models and multi-modal representations as central to bridging the media gap.