GeoDE: a Geographically Diverse Evaluation Dataset for Object Recognition
This addresses geographic bias in AI datasets, which affects fairness and performance in object recognition globally, though it is incremental as it builds on existing dataset collection efforts.
The authors tackled the problem of geographic bias in object recognition datasets by introducing GeoDE, a geographically diverse dataset with 61,940 images from 40 classes and 6 world regions, collected without personally identifiable information. They demonstrated its use for evaluation and training to highlight and mitigate shortcomings in current models.
Current dataset collection methods typically scrape large amounts of data from the web. While this technique is extremely scalable, data collected in this way tends to reinforce stereotypical biases, can contain personally identifiable information, and typically originates from Europe and North America. In this work, we rethink the dataset collection paradigm and introduce GeoDE, a geographically diverse dataset with 61,940 images from 40 classes and 6 world regions, with no personally identifiable information, collected by soliciting images from people around the world. We analyse GeoDE to understand differences in images collected in this manner compared to web-scraping. We demonstrate its use as both an evaluation and training dataset, allowing us to highlight and begin to mitigate the shortcomings in current models, despite GeoDE's relatively small size. We release the full dataset and code at https://geodiverse-data-collection.cs.princeton.edu