Dataset Card for Dataception

Dataset of datasets.

Dataset Details

Dataset Description

Dataception is a dataset of image-text training datasets.

Datasets were collected from CivitAI. Thanks to all the users who curated the original datasets and shared them ❤️

Datasets have been processed and converted into WebDataset format.

To ensure consistent format .png images are re-encoded to jpg with 90% quality and .jpeg extension renamed to .jpg.

Contents of each WebDataset are named with zero-filled sequence number e.g. 00000.txt.

Each dataset is accompanied by the original model metadata in .json format.

Notes

Some datasets were excluded from this release:

• If multiple subdirectories containing image-text pairs were detected; this structure is used in some training UIs to denote multiple concepts. These datasets will be reviewed to determine appropriate processing into WebDataset format.
• If no image-text pairs were detected. These datasets will be reviewed to determine whether captions are in the filenames or just not available.
• If the file is not actually a .zip; some cases of .rar files have been detected which is unsupported by zipfile library that the current processing script uses.
• If the .zip uses an unknown compression type that is unsupported by zipfile library. This excludes deflate64 (used by Windows), which is unsupported by default, thanks to patching from zipfile_deflate64 library.
• Both cases of unsupported files will be processed at a later date.

Curated by: hlky

License: Open Data Commons Attribution License (ODC-By) v1.0

Dataset Structure

Dataception

• id: Model Version ID.
• modelId: Model ID.
• name: Model name.
• type: Model type.
• baseModel: Base model used for the original training.
• nsfwLevel: Content rating for the original model and dataset.
• dataset_wds: Path to the WebDataset.
• dataset_json: Path to the .json metadata.

Individual datasets

WebDataset with .jpg and .txt.

Uses

Potential uses include:

• Training practice; easily compare your result to the original model.
• Retraining; with modern tooling and newer base models.

Citation Information

@misc{Dataception,
  author = {hlky},
  title = {Dataception},
  year = {2024},
  publisher = {hlky},
  journal = {Hugging Face repository},
  howpublished = {\url{[https://huggingface.co/datasets/bigdata-pw/Dataception](https://huggingface.co/datasets/bigdata-pw/Dataception)}}
}

Attribution Information

Contains information from [Dataception](https://huggingface.co/datasets/bigdata-pw/Dataception) which is made available
under the [ODC Attribution License](https://opendatacommons.org/licenses/by/1-0/).