Learning Visual Relation Priors for Image-Text Matching and Image Captioning with Neural Scene Graph Generators

TL;DR

This paper introduces the use of neural scene graph generators to learn visual relation features, significantly enhancing image-text matching and captioning by grounding language in visual relations, as demonstrated on Flickr30K and MSCOCO datasets.

Contribution

The work presents a novel approach of integrating relation features from scene graph generators into existing models, improving their ability to capture visual relations in language-and-vision tasks.

Findings

Relation features improve model performance on benchmarks

Scene graph generators with relevant relations are crucial for effectiveness

Significant improvements on Flickr30K and MSCOCO datasets

Abstract

Grounding language to visual relations is critical to various language-and-vision applications. In this work, we tackle two fundamental language-and-vision tasks: image-text matching and image captioning, and demonstrate that neural scene graph generators can learn effective visual relation features to facilitate grounding language to visual relations and subsequently improve the two end applications. By combining relation features with the state-of-the-art models, our experiments show significant improvement on the standard Flickr30K and MSCOCO benchmarks. Our experimental results and analysis show that relation features improve downstream models' capability of capturing visual relations in end vision-and-language applications. We also demonstrate the importance of learning scene graph generators with visually relevant relations to the effectiveness of relation features.