About the dataset

This dataset contains a around 10K satellite images split into 3 folders -

• 8734 train images
• 1093 test images
• 1094 validation images

Each of the images are accompanied by 5 different captions - annotated by 5 different annotators .

*For example : *

Sample image 1

Has 5 captions describing it

• "'Many planes are near a large building in an airport.'",
• "'the terminal building with a large number of aircraft is surrounded by several tracks.'",
• "'the terminal building wih a great many of airplanes aisde is surrounded by several runways .'",
• "'Many planes are close to a large building at an airport.'",
• "'many planes are near a large building in an airport .'"

Sample image 2

-"'There is a large playground in the gym.'",

• "'Many cars were parked outside the stadium.'",
• "'many cars were parked outside the stadium .'",
• '"There's a big playground in the gym."',
• "'there is a big playground in the gym .'

Here's a starter kernel to view the images - https://www.kaggle.com/code/tomtillo/viewing-the-images-and-captions

See more details in these two papers https://arxiv.org/pdf/1712.07835.pdf https://www.sciencedirect.com/science/article/pii/S1877050920300752

Potential uses of this dataset

1. Automatic Caption generation

Train the images and the corresponding captions using any of the CLIP models

2. Semantic search using embedding