Description
This dataset is a new extension to Google's Open Images dataset and aims to provide researchers and developers with more geo-diverse images for model training and testing. The dataset contains images of everyday things—from household objects to plants and animals to local monuments and landmarks—contributed from countries that are often under-represented in open imagery datasets. People have shared these images using the Google Crowdsource app, in a gamified interface that awards badges and level-ups. All images are shared under the Creative Commons CC-BY-4.0 license.
This dataset was released in conjunction with the Inclusive Images competition, a collaboration between Google AI and the Kaggle community to bring geographical and demographic diversity to imagery training data. The starting dataset contains 478k images and will continue to grow with Crowdsource contributions. Accompanying this dataset is a Data Card, which describes the composition, method of collection, and annotation of the dataset.
The Google AI team would like to invite communities worldwide—especially those in Asia, Africa, and Latin America—to help build out this dataset by contributing images that showcase their cultures and environments. This can be done via the Crowdsource app, or by reaching out to open-images-extended@google.com for contributing larger imagery datasets.