metadata

license: cc0-1.0
task_categories:
  - text-to-image
pretty_name: DiffusionDB Captions
size_categories:
  - 1M<n<10M
dataset_info:
  features:
    - name: description
      dtype: string
  splits:
    - name: train
      num_bytes: 50735678
      num_examples: 967163
    - name: validation
      num_bytes: 101940
      num_examples: 1948
    - name: test
      num_bytes: 256037
      num_examples: 4870
  download_size: 35930405
  dataset_size: 51093655
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*

DiffusionDB Captions

Captions corpora processed from DiffusionDB dataset, the goal is to create a dataset similar to flickr30k but much larger

Filtered from diffusionDB + diffusionDB_large, total 14M data, into ~1M pre-processing includes:

remove style prompts
remove too long and specific prompts
remove prompts with many specific names

Citation

If you find this curated dataset helpful, feel free to give us a cite.

@misc{rodriguez2025renderingawarereinforcementlearningvector,
      title={Rendering-Aware Reinforcement Learning for Vector Graphics Generation}, 
      author={Juan A. Rodriguez and Haotian Zhang and Abhay Puri and Aarash Feizi and Rishav Pramanik and Pascal Wichmann and Arnab Mondal and Mohammad Reza Samsami and Rabiul Awal and Perouz Taslakian and Spandana Gella and Sai Rajeswar and David Vazquez and Christopher Pal and Marco Pedersoli},
      year={2025},
      eprint={2505.20793},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2505.20793}, 
}