---
dataset_info:
  features:
  - name: image
    dtype: image
  - name: caption
    dtype: string
  splits:
  - name: train
    num_bytes: 4993980142.0
    num_examples: 30000
  download_size: 4898811398
  dataset_size: 4993980142.0
---
# Dataset Card for "coco-30-val-2014"

This is 30k randomly sampled image-captioned pairs from the [COCO](https://cocodataset.org/) 2014 `val` split. This is useful for image generation benchmarks (FID, CLIPScore, etc.).

Refer to the gist to know how the dataset was created: https://gist.github.com/sayakpaul/0c4435a1df6eb6193f824f9198cabaa5.