Multimodal grid features and cell pointers for Scene Text Visual Question Answering

TL;DR

This paper introduces a novel multimodal attention-based model for scene text visual question answering, effectively integrating visual and textual cues to locate and interpret scene text in images.

Contribution

It proposes a new model utilizing attention over multi-modal features conditioned on questions, with a novel analysis of the ST-VQA dataset based on human performance.

Findings

Achieves competitive results on standard datasets

Provides insights into human performance on scene text VQA

Introduces a new interpretability approach for attention weights

Abstract

This paper presents a new model for the task of scene text visual question answering, in which questions about a given image can only be answered by reading and understanding scene text that is present in it. The proposed model is based on an attention mechanism that attends to multi-modal features conditioned to the question, allowing it to reason jointly about the textual and visual modalities in the scene. The output weights of this attention module over the grid of multi-modal spatial features are interpreted as the probability that a certain spatial location of the image contains the answer text the to the given question. Our experiments demonstrate competitive performance in two standard datasets. Furthermore, this paper provides a novel analysis of the ST-VQA dataset based on a human performance study.