Fei-Fei Li and ImageNet

In the early 2000s, Fei-Fei Li was a young researcher at Stanford studying computer vision — teaching machines to see. And she had a diagnosis for what was wrong with the field.

Everyone was working on algorithms. Better models, cleverer techniques, smarter approaches to recognizing images. But there was no shared, rigorous way to compare them. Researchers tested on small datasets. Results varied. Progress was hard to measure.

Her insight: the bottleneck wasn't the algorithms. It was the data.

She set out to build a benchmark large enough to matter. 1.2 million photographs, each carefully labeled by human workers with one of 1,000 categories — cats, airplanes, coffee mugs, volcanoes, hundreds of others. The labeling alone took years, done by thousands of workers hired online.

In 2010, she launched a competition: the ImageNet Large Scale Visual Recognition Challenge. Every year, teams around the world would compete to see whose system could classify the images most accurately. It wasn't just a competition. It was a shared measuring stick for the whole field.

The stage was set.