sayakpaul/coco-30-val-2014

Dataset Card for "coco-30-val-2014" This is 30k randomly sampled image-captioned pairs from the COCO 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.