What question did this study set out to answer?

This research aims to learn scene representations that effectively transfer across various modalities.

September 18, 2017

Cross-Modal Scene Networks

Key Points

This research aims to learn scene representations that effectively transfer across various modalities.
Introduced a new cross-modal scene dataset.
Applied regularization techniques to convolutional neural networks to align representations across modalities.
Conducted experiments to evaluate retrieval performance of the cross-modal representations.
The proposed scene representation improves the transfer of representations across modalities for retrieval purposes.
Visualizations indicate that shared representation units activate consistently on distinct concepts, regardless of modality.

Abstract

People can recognize scenes across many different modalities beyond natural images. In this paper, we investigate how to learn cross-modal scene representations that transfer across modalities. To study this problem, we introduce a new cross-modal scene dataset. While convolutional neural networks can categorize scenes well, they also learn an intermediate representation not aligned across modalities, which is undesirable for cross-modal transfer applications. We present methods to regularize cross-modal convolutional neural networks so that they have a shared representation that is agnostic of the modality. Our experiments suggest that our scene representation can help transfer representations across modalities for retrieval. Moreover, our visualizations suggest that units emerge in the shared representation that tend to activate on consistent concepts independently of the modality.

Bookmark

Cross-Modal Scene Networks

Key Points

Abstract

Cite This Study

Also Consider

Also Consider