Reproduce “Flickr Africa: Examining Geo-Diversity in Large-Scale, Human-Centric Visual Data” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Keziah Naggita, Julienne LaChance, Alice Xiang

An examination of 1.5 million geotagged Flickr images from Africa reveals a severe scarcity of human-centric visual data, a predominant "othering" phenomenon where non-local photographers capture most images, and widespread "All Rights Reserved" licensing, highlighting significant biases and usability challenges for computer vision research. The analysis quantifies these issues by comparing African nations to European counterparts and assessing intra-national wealth distribution to show deficiencies in current data collection methods.

Key insights

  • Relying on web scraping from platforms like Flickr for data from Africa results in insufficient and biased visual data, failing to provide truly representative human-centric content.
  • A significant "othering" phenomenon exists, where non-local (foreign) photographers disproportionately capture geotagged images from African nations, potentially leading to stereotypical or unauthentic portrayals.
  • The vast majority (80-82%) of Flickr images are marked with the "All Rights Reserved" license, posing substantial ethical and legal barriers for researchers to use this data for machine learning training.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.