The story
ImageNet is a database of labeled photographs that Fei-Fei Li and her colleagues at Princeton presented at a computer vision conference in 2009. It grew to about 14 million images in more than 20,000 categories, such as 'golden retriever' or 'fire engine'. The categories came from WordNet, an English-language dictionary organized by meaning, and the pictures were collected from the web and labeled by tens of thousands of crowd workers on Amazon Mechanical Turk, who checked whether each image really showed the thing named.
At the time most researchers believed better algorithms mattered more than more data, and Li bet the opposite. From 2010 a yearly contest, the ImageNet Large Scale Visual Recognition Challenge, used a subset of 1.2 million images in 1,000 categories and gave labs a shared scoreboard. In 2012 a neural network called AlexNet, built by Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton, cut the top-five error rate to about 15 percent, against 26 percent for the runner-up, and it ran on graphics processors. That result convinced the field that deep learning worked. Error fell below human-level estimates by 2015. The contest ended in 2017, and researchers later found that the labels contained biases and errors, and in 2019 the team removed many offensive person categories after criticism.
Why it mattered
- AlexNet's 2012 win began the deep learning boom in computer vision and then in other fields.
- The idea that large datasets drive progress shaped the later development of language and image models.
- Pretraining models on ImageNet became a standard step for many vision applications.
- Debates about dataset bias, consent and labeling errors led to audits and the removal of some categories.
Sources
- ImageNet Large Scale Visual Recognition Challenge International Journal of Computer Vision
- Fei-Fei Li Britannica
Drafted by AI (Claude) from the sources above and checked against them by AI; no historian has reviewed this page. Spot a mistake? Suggest a correction. How pages are made
Community notes
Loading notes…