Stacked Attention Networks for Image Question Answering

TLDR

Image question answering often requires multiple steps of reasoning. The paper proposes stacked attention networks that learn to answer natural language questions from images by repeatedly querying the image across multiple layers to infer the answer progressively. SANs use the question’s semantic representation as a query to iteratively search image regions across multiple layers, progressively refining the answer. Experiments on four image QA datasets show that SANs significantly outperform prior state‑of‑the‑art methods, and attention visualizations reveal progressive localization of relevant visual clues.

Abstract

This paper presents stacked attention networks (SANs) that learn to answer natural language questions from images. SANs use semantic representation of a question as query to search for the regions in an image that are related to the answer. We argue that image question answering (QA) often requires multiple steps of reasoning. Thus, we develop a multiple-layer SAN in which we query an image multiple times to infer the answer progressively. Experiments conducted on four image QA data sets demonstrate that the proposed SANs significantly outperform previous state-of-the-art approaches. The visualization of the attention layers illustrates the progress that the SAN locates the relevant visual clues that lead to the answer of the question layer-by-layer.