ImageNet

Assembled from 16 blocks · 1 evidence · 15 related

1. Story
2. The labelled image dataset, begun in 2007 under Fei-Fei Li and released from 2009, that supplied the missing ingredient for deep learning in vision and set the pattern for everything that followed. Its size was the argument: millions of images, hand-labelled across thousands of categories, organised on the WordNet noun hierarchy, built largely through crowdsourced annotation. The annual ImageNet Large Scale Visual Recognition Challenge that ran on a subset of it turned image classification into a public, repeatable, externally scored contest, which is what made the 2012 result legible as a result rather than as a claim. ImageNet is also the origin of a habit the field has not escaped: progress came to be defined as movement on a fixed, public, downloadable test set — a design that rewards optimisation against the set itself and that later benchmarks inherited wholesale.
3. Knowledge
4. ImageNet
5. Fei-Fei Li
6. Connections
7. Deep learning
8. Fei-Fei Li
9. ImageNet is assembled and released
10. A deep convolutional network wins the ImageNet challenge
11. ImageNet
12. Benchmark saturation
13. ImageNet is assembled and released
14. The AI Index records both fast benchmark movement and doubts about benchmarks
15. Evidence
16. An eight-layer convolutional network — five convolutional and three fully connected layers — trained on graphics processors with dropout regularisation achieved a top-5 error rate of 15.3% at ILSVRC 2012 against 26.2% for the second-placed entry, learning hierarchical features rather than using hand-designed ones. V55 verification basis: the paper was not fetched. The 15.3% and 26.2% figures and the architecture description were carried in multiple retrieved summaries and in a hosted copy of the paper itself, but the attribution in those summaries is to secondary explainers as often as to the paper, and a curator should confirm both error rates against the ILSVRC 2012 results table before publication.