ImageNet
Era 3 · From the expert boom to the statistical turn (1980–2011) › Hardware and data
- Lineage
- Machines and computation
- Kind
- Implementation
Who
Princeton University, Stanford University
What happened
Jia Deng, Fei-Fei Li, and their colleagues present ImageNet at CVPR 2009: an image database organized according to WordNet's hierarchy of concepts, labeled by hand and designed to grow to tens of millions of photos.
Why it mattered
It gives computer vision data on a new scale: 3.2 million images in 2009 and more than 14 million today, and from 2010 on an annual competition that serves as a shared benchmark.
The full story
In the mid-2000s computer vision was measured on collections of a few thousand images and a few dozen categories. Fei-Fei Li believed the bottleneck was not just the algorithms but the data: a system that had to recognize the world needed to see many more examples of many more things.
In 2006 Li began building ImageNet. The plan was to take the hierarchy of WordNet, a lexical database of English, and gather hundreds of photos for each of its nouns. Labeling millions of images by hand was beyond the means of a single lab, so the team turned to Amazon Mechanical Turk, a service that farms out small tasks to thousands of people for pay. Between July 2008 and April 2010 some 49,000 people worked on it.
The 2009 paper, by Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Fei-Fei Li, presented the database as a poster at CVPR, with 3.2 million images in 5,247 categories. Today it holds more than 14 million images in more than 20,000 categories.
Starting in 2010, an annual competition on a subset of 1,000 categories and about 1.2 million training images compared the best recognition systems. In 2012 it was won by AlexNet, a convolutional network trained on GPUs, and that result marked the start of the deep learning era.
Common misconceptions
What people often say: ImageNet was created for the competition AlexNet won.
In fact: The database was presented in 2009 and the competition began in 2010. AlexNet won its third edition, in 2012.
Videos
Leads to
AlexNet (September 30, 2012)
Documented direct influence
AlexNet is trained on GPUs with CUDA, on ImageNet
Threads that pass through here
- Data, GPUs, and AlexNet (step 2 of 4)
- The neural network winter (step 9 of 11)
Sources
Original source
- ImageNet: A large-scale hierarchical image database. Proceedings of CVPR 2009, pp. 248-255 (opens in a new tab). Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei, 2009
Further reading
How to cite
“ImageNet.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/imagenet
This entry was last reviewed on November 13, 2026.