hayden-donnelly's picture
Update README.md
19b68b8 verified
|
raw
history blame
1.56 kB
metadata
task_categories:
  - image-classification
  - text-to-image
  - unconditional-image-generation
pretty_name: Danbooru SFW 512 Filtered and Cropped Dataset
size_categories:
  - 1M<n<10M

Danbooru SFW 512 Filtered and Cropped

A version of Danbooru SFW which has been automatically filtered and cropped to 128x128 so that the resulting images focus almost entirely on characters. First filtering was applied to remove non-character focused images, then cropping was applied to remove horizontal/vertical bars and further focus on the central character(s) of each image. Both steps were performed automatically by two different vision models trained on manually labelled subsets of the original dataset. This technique provided satisfactory results, however there is much room for improvement.

Preprocessing Datasets

db-sfw-512-character-filter-dataset was used to train the character filter model and db-sfw-512-crop-dataset was used to train the crop model.

Original Dataset Citation

@misc{danbooru2021,
    author={Anonymous and Danbooru community and Gwern Branwen},
    title={Danbooru2021: A Large-Scale Crowdsourced and Tagged Anime Illustration Dataset},
    howpublished={\url{https://gwern.net/danbooru2021}},
    url={https://gwern.net/danbooru2021},
    type={dataset},
    year={2022},
    month={January},
    timestamp={2022-01-21},
    note={Accessed: 2023-12-06}
}