Learning Visually-Grounded Semantics from Contrastive Adversarial Samples
Abstract
We study the problem of grounding distributional representations of texts on the visual domain, namely visual-semantic embeddings (VSE for short). Begin with an insightful adversarial attack on VSE embeddings, we show the limitation of current frameworks and image-text datasets (e.g., MS-COCO) both quantitatively and qualitatively. The large gap between the number of possible constitutions of real-world semantics and the size of parallel data, to a large extent, restricts the model to establish the link between textual semantics and visual concepts. We alleviate this problem by augmenting the MS-COCO image captioning datasets with textual contrastive adversarial samples. These samples are synthesized using linguistic rules and the WordNet knowledge base. The construction procedure is both syntax- and semantics-aware. The samples enforce the model to ground learned embeddings to concrete concepts within the image. This simple but powerful technique brings a noticeable improvement over the baselines on a diverse set of downstream tasks, in addition to defending known-type adversarial attacks. We release the codes at https://github.com/ExplorerFreda/VSE-C.
Cite
@article{arxiv.1806.10348,
title = {Learning Visually-Grounded Semantics from Contrastive Adversarial Samples},
author = {Haoyue Shi and Jiayuan Mao and Tete Xiao and Yuning Jiang and Jian Sun},
journal= {arXiv preprint arXiv:1806.10348},
year = {2018}
}
Comments
To Appear at COLING 2018