Look, Imagine and Match: Improving Textual-Visual Cross-Modal Retrieval with Generative Models

TL;DR

This paper introduces a generative model-based framework for textual-visual cross-modal retrieval that enhances feature embedding by learning both global and local grounded features, achieving state-of-the-art results on MSCOCO.

Contribution

It proposes a novel generative approach to improve cross-modal retrieval by incorporating local grounded features into the embedding process.

Findings

Achieves state-of-the-art retrieval performance on MSCOCO dataset.

Effectively models complex image-sentence content.

Enhances feature representations with generative processes.

Abstract

Textual-visual cross-modal retrieval has been a hot research topic in both computer vision and natural language processing communities. Learning appropriate representations for multi-modal data is crucial for the cross-modal retrieval performance. Unlike existing image-text retrieval approaches that embed image-text pairs as single feature vectors in a common representational space, we propose to incorporate generative processes into the cross-modal feature embedding, through which we are able to learn not only the global abstract features but also the local grounded features. Extensive experiments show that our framework can well match images and sentences with complex content, and achieve the state-of-the-art cross-modal retrieval results on MSCOCO dataset.