ImageNet gives computer vision a dataset at internet scale
Fei-Fei Li's team crowdsources millions of labeled images organized by WordNet's structure
Quick facts
- Lead researcher
- Fei-Fei Li
- Institutions
- Princeton University
- Initial scale
- 3.2 million images, 5,247 synsets (2009)
What happened
Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei presented 'ImageNet: A Large-Scale Hierarchical Image Database' at the IEEE Conference on Computer Vision and Pattern Recognition in Miami. The dataset organized images according to WordNet, a lexical database that groups roughly 80,000 English nouns into hierarchical concepts, or synsets, such as mammal, then dog, then golden retriever. Using Amazon Mechanical Turk to crowdsource human verification of image labels at scale, the initial 2009 release already contained 3.2 million images across 5,247 synsets, with the stated goal of eventually reaching tens of millions of images covering the majority of WordNet's noun concepts. The team's founding insight was that computer vision progress had been limited less by algorithms than by the small size of existing training sets, some containing only a few thousand images.
Why it matters
ImageNet supplied the enormous, reliably labeled training data that deep neural networks needed to outperform older computer-vision techniques, and its annual competition became the proving ground where AlexNet demonstrated deep learning's breakthrough three years later.
How we know
The original 2009 CVPR paper, hosted on ImageNet's own site, documents the dataset's construction and scale directly, and ImageNet's official project page confirms its ongoing goal of covering WordNet's noun hierarchy.
Sources
- Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, Li Fei-Fei, CVPR 2009. ImageNet: A Large-Scale Hierarchical Image Database · Primary source (author-declared)image-net.org · Cited as a "primary" source (no stronger domain match).
- ImageNet project. About ImageNet · Primary source (author-declared)image-net.org · Cited as a "primary" source (no stronger domain match). · Link is live and its text matches the event's key terms (Jul 2026)
See something wrong? . Corrections with a source get fixed fastest.
Part of a timelineArtificial Intelligence30 events · From a wartime theory of neurons to machines that write, paint, and fold proteinsView all →