Data augmentation techniques for the Video Question Answering task

TL;DR

This paper explores data augmentation methods to improve the performance of Egocentric Video Question Answering models, addressing overfitting issues due to limited dataset size and achieving a 5.5% accuracy boost.

Contribution

The authors introduce several data augmentation techniques specifically designed for Egocentric VideoQA to enhance model generalization on small datasets.

Findings

Augmentation techniques improved accuracy by 5.5%.

Proposed methods reduce overfitting on small datasets.

Enhanced model robustness in Egocentric VideoQA tasks.

Abstract

Video Question Answering (VideoQA) is a task that requires a model to analyze and understand both the visual content given by the input video and the textual part given by the question, and the interaction between them in order to produce a meaningful answer. In our work we focus on the Egocentric VideoQA task, which exploits first-person videos, because of the importance of such task which can have impact on many different fields, such as those pertaining the social assistance and the industrial training. Recently, an Egocentric VideoQA dataset, called EgoVQA, has been released. Given its small size, models tend to overfit quickly. To alleviate this problem, we propose several augmentation techniques which give us a +5.5% improvement on the final accuracy over the considered baseline.