VictorG-028's picture
Update train proportion
24621c0 verified
metadata
license: unknown
language:
  - en
task_categories:
  - image-to-text
tags:
  - code
pretty_name: mjsynth
size_categories:
  - 10K<n<100K

This dataset contains:

  • 40351 images (71.39%) in train dataset
  • 6378 images (14.29%) in validation dataset
  • 6391 images (14.32%) in test dataset
  • Total: 53120 images
  • sourced from the extensive Synthetic Word Dataset, a large-scale word-image dataset.

The original and complete dataset (9 million images, 10.68GB) can be found and downloaded at this academic torrent.