I counted all of the yurts in Mongolia using machine learning

a year ago

The Fall of Civilizations podcast released a 6¾-hour episode on the Mongol Empire, sparking interest in contemporary Mongolian society.
Mongolia has rapidly reduced poverty in the 21st century, with high economic growth and a healthy fertility rate, but struggles with corruption and governance issues.
Google Maps exploration revealed extensive yurt settlements (ger districts) in Ulaanbaatar, prompting a project to count yurts using machine learning.
A YOLO-based machine learning model was trained on manually labeled satellite imagery to detect yurts, with over 10,000 yurts annotated for training.
The project scaled up using Docker Swarm and rented GPUs, processing millions of tiles to count 172,689 yurts across Mongolia.
Yurts in urban areas reflect Mongolia's transition from a nomadic herder society to an urbanized industrial one, with ger districts persisting due to housing shortages.
Government efforts to develop ger districts, like the 2002 land privatization law and the Ulaanbaatar 2020 Master Plan, have seen slow progress.
The article raises further questions about urbanization, industrialization, and development challenges in Mongolia.

Hasty Briefsbeta