metadata

task_categories:
  - image-to-text
  - text-to-image
size_categories:
  - n<1K

NASA Earth Instagram

This dataset is a moderately curated subset of the posts shown on NASA Earth's Instagram, with an emphasis on finding image-text pairs where the text associated is as close as possible to being a direct caption of the image in question.

This dataset has a variety of use cases, but the one which it is originally intended for is to provide a fine-tuning dataset for image captioning models, to be better equipped for describing the exact pheonomena in satellite imagery.

The owner of all images and text in this data is NASA.