README.md

metadata

dataset_info:
  - config_name: bird
    features:
      - name: image
        dtype: image
      - name: entity_description
        dtype: string
      - name: prompt
        dtype: string
    splits:
      - name: train
        num_bytes: 1153768
        num_examples: 50
    download_size: 1077545
    dataset_size: 1153768
  - config_name: car
    features:
      - name: image
        dtype: image
      - name: entity_description
        dtype: string
      - name: prompt
        dtype: string
    splits:
      - name: train
        num_bytes: 5285261
        num_examples: 50
    download_size: 4433750
    dataset_size: 5285261
  - config_name: dish
    features:
      - name: image
        dtype: image
      - name: entity_description
        dtype: string
      - name: prompt
        dtype: string
    splits:
      - name: train
        num_bytes: 24535127
        num_examples: 50
    download_size: 14104477
    dataset_size: 24535127
  - config_name: dog
    features:
      - name: image
        dtype: image
      - name: entity_description
        dtype: string
      - name: prompt
        dtype: string
    splits:
      - name: train
        num_bytes: 1144143
        num_examples: 50
    download_size: 918873
    dataset_size: 1144143
  - config_name: landmark
    features:
      - name: image
        dtype: image
      - name: entity_description
        dtype: string
      - name: prompt
        dtype: string
    splits:
      - name: train
        num_bytes: 305009678
        num_examples: 50
    download_size: 258293099
    dataset_size: 305009678
  - config_name: starwar
    features:
      - name: image
        dtype: image
      - name: entity_description
        dtype: string
      - name: prompt
        dtype: string
    splits:
      - name: train
        num_bytes: 14444301
        num_examples: 50
    download_size: 9059561
    dataset_size: 14444301
configs:
  - config_name: bird
    data_files:
      - split: train
        path: bird/train-*
  - config_name: car
    data_files:
      - split: train
        path: car/train-*
  - config_name: dish
    data_files:
      - split: train
        path: dish/train-*
  - config_name: dog
    data_files:
      - split: train
        path: dog/train-*
  - config_name: landmark
    data_files:
      - split: train
        path: landmark/train-*
  - config_name: starwar
    data_files:
      - split: train
        path: starwar/train-*
license: mit
task_categories:
  - text-to-image
language:
  - en
size_categories:
  - n<1K

This dataset is a replication evaluation set from Re-Imagen (https://arxiv.org/pdf/2209.14491.pdf). The dataset is aimed for evaluating text-to-image generation on rare visual entities.