Relation Graph Reasoning for Image-Text Matching
Lists
Only you see this
Abstract
Image-text matching is one of the research hotspots in cross-modal field. Most of the current methods focus on digging the semantic matching degree between image and text to depict the external relations. However, the extraction of implicit semantic features is still not accurate enough to effectively model the internal relationships between entities. To address this problem, we propose a novel Relation Graph Reasoning Network(RGRN). RGRN can not only learn the concepts of entities and attributes through content reasoning, but also its topological structure reasoning can better obtain the implicit action and orientation relationships between visual entities. Besides, we also integrate the overall semantics to obtain the final visual representation, helping to cross the semantic gap between image and text. We conduct experiments on Flickr30K and MS-COCOCO datasets to verify the effectiveness of our model.
BibTeX
@inproceedings{li2022relation,
author = {Zhao Li and Xiaohong Nian and Chao Pan and Dayin Yang and Hongyun Xiong and Haibo Wang},
title = {{Relation Graph Reasoning for Image-Text Matching}},
booktitle = {2022},
year = {2022},
doi = {10.1109/icivc55077.2022.9887071},
}Loading related content...