Skip to content

Data set for the IEEE TGRS paper "Mutual Attention Inception Network for Remote Sensing Visual Question Answering"

Notifications You must be signed in to change notification settings

spectralpublic/RSIVQA

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

25 Commits
 
 
 
 
 
 
 
 

Repository files navigation

RSIVQA (Remote Sensing Image Visual Question Answering)

RSIVQA is a remote sensing visual question answering dataset proposed in the paper "Mutual Attention Inception Network for Remote Sensing Visual Question Answering". If this data set is used in your work, please cite our paper.

[1] X. Zheng, B. Wang, X. Du, and X. Lu, “Mutual Attention Inception Network for Remote Sensing Visual Question Answering,” IEEE Transactions on Geoscience Remote Sensing, 2021.

@article{ 9444570,
author = {Zheng, Xiangtao and Wang, Binqiang and Du, Xingqian and Lu, Xiaoqiang},
doi = {10.1109/TGRS.2021.3079918},
issn = {0196-2892},
journal = {IEEE Transactions on Geoscience and Remote Sensing},
title = {{Mutual Attention Inception Network for Remote Sensing Visual Question Answering}},
year = {2021}
}

Overview

RSIVQA dataset is derived from existing remote sensing image (RSI) datasets with a specially designed generation method.

Currently, images of RSIVQA come from three RSI classification datasets (UCM, Sydney and AID) and two RSI object detection datasets (HRRSD and DOTA). Questions and answers are generated based on images to form image-question-answer triplets. Questions, answers and their correspondence can be found in txt files in this repository. Images of the datasets can be downloaded from the link of each dataset.

Deatails

There are 37264 images and 111134 image-question-answer triplets in the dataset. Detaled information is summarized in the table below.

Item Amount
The number of images 37264
The number of unique questions 91
The number of unique answers 574
The total number of VQA triplets 111134
The number of yes or no VQA triplets 64610
The number of number VQA triplets 32331
The number of others VQA triplets 14193

A small part of RSIVQA is annotated by human. Others are automatically generated using existing scene classification datasets and object detection datasets. For more detailed information of the generation method, please refer to the paper.
Note that 559 triplets are added in current version, which makes up to 111693 vqa triplets in total.

Files

The question and answers are saved in txt files. Each line includes an IQA triplet and three parts are seprated by colon and question mark.
For example, in sydney_vqa.txt, the first line is

1.tif:what is the theme of this picture?residential

which means the question for the image "1.tif" is "what is the theme of this picture?". The answer for the question is "residential". "1.tif" is file name of the image.

Image-question-answer triplets for each origin scene classification or object detection dataset are saved in corresponding folders.


About

Data set for the IEEE TGRS paper "Mutual Attention Inception Network for Remote Sensing Visual Question Answering"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published