README.md

metadata

language:
  - en
license: mit
size_categories:
  - 1M<n<10M
task_categories:
  - visual-question-answering
  - image-text-to-text
pretty_name: ABC-Pretraining-Data
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: caption
      dtype: string
    - name: url
      dtype: string
    - name: id
      dtype: int64
    - name: image
      dtype: string
    - name: negatives
      sequence: int64
  splits:
    - name: train
      num_bytes: 2289772991
      num_examples: 2252041
  download_size: 1855548818
  dataset_size: 2289772991
tags:
  - visual

ABC Pretraining Data

This the the pretraining data for ABC. This dataset is derived from Google's Conceptual Captions dataset. The each item in the dataset contain a URL where the corresponding image can be downloaded and mined negatives for each item. Full dataaset is ~300 GB of images. For a detailed description of how we mined the negatives please check out our ppaer ;).
Update I have added the images to this repository, for an example of how to use and download this dataset see our repository.

Paper and Website

For more information, please refer to Website.

Citation

If you find any of our work helpful please connsider citing:

@misc{schneider2025abcachievingbettercontrol,
      title={ABC: Achieving Better Control of Multimodal Embeddings using VLMs}, 
      author={Benjamin Schneider and Florian Kerschbaum and Wenhu Chen},
      year={2025},
      eprint={2503.00329},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.00329}, 
}