Conceptual Captions¶

Attribute	Value
pretty_name	Conceptual Captions
annotations_creators
language_creators
languages
licenses	custom
multilinguality
size_categories	10M<n<100M
source_datasets
task_categories
task_ids
paperswithcode_id

Dataset Description¶

Paper: Conceptual Captions Paper
Licenses: Attribution

Dataset Summary¶

Pairs of images and captions.

Download and prepare data¶

The dataset can be loaded directly via the squirrel Catalog API. Make sure that squirrel-dataset-core is installed via pip, which will register this dataset. Use the following code to load the data:

from squirrel.catalog import Catalog

plugin_catalog = Catalog.from_plugins()
it = plugin_catalog["conceptual-captions-12m"].get_driver().get_iter()

Dataset Structure¶

Data Instances¶

A sample from the training set is provided below:

{
    'url': 'https://i.pinimg.com...',
    'error': False,
    'image': array(...)
    'caption': 'Peterbilt 359 custom built show me how to find this Large Cars kits...'
}

Dataset Schema¶

error: True if there was an http error accessing the data.

Data Splits¶

name
CC12M	12M