An Efficient Modern Baseline for FloodNet VQA

TL;DR

This paper introduces a simple, efficient VQA system for flood disaster management that outperforms existing methods on FloodNet with less training and inference time.

Contribution

It revisits basic combination methods with modern features to create a lightweight, high-performing VQA baseline for flood-related visual question answering.

Findings

Outperforms previous methods on FloodNet dataset

Requires less training and inference time

Analyzes various backbone architectures

Abstract

Designing efficient and reliable VQA systems remains a challenging problem, more so in the case of disaster management and response systems. In this work, we revisit fundamental combination methods like concatenation, addition and element-wise multiplication with modern image and text feature abstraction models. We design a simple and efficient system which outperforms pre-existing methods on the FloodNet dataset and achieves state-of-the-art performance. This simplified system requires significantly less training and inference time than modern VQA architectures. We also study the performance of various backbones and report their consolidated results. Code is available at https://github.com/sahilkhose/floodnet_vqa.