sbu_captions

ds = tfds.load('huggingface:sbu_captions')

The SBU Captioned Photo Dataset is a collection of over 1 million images with associated text descriptions extracted from Flicker.

{
    "image_url": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "user_id": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "caption": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    }
}