Who created ImageNet?
Fei-Fei Li began developing ImageNet in 2006 and, starting in 2007, worked with Princeton professor Christiane Fellbaum, one of WordNet's creators, to build the project from WordNet's roughly 22,000 nouns.
Questions about ImageNet
Short answers, pulled from the story.
What is ImageNet used for?
ImageNet is a large visual database used to train and test software that recognizes objects in images. It is organized into more than 20,000 categories and is used to run the annual ImageNet Large Scale Visual Recognition Challenge.
When did the ImageNet Large Scale Visual Recognition Challenge start?
The ILSVRC launched in 2010 as a collaboration between Fei-Fei Li and organizers of the PASCAL Visual Object Classes contest, using a trimmed list of 1,000 classes.
How many images are in ImageNet's original dataset?
ImageNet's original collection, known as ImageNet-21K, contains 14,197,122 images across 21,841 classes, gathered through crowdsourced labeling between July 2008 and April 2010.
Why was AlexNet significant for ImageNet?
On the 30th of September 2012, the convolutional neural network AlexNet reached a top-5 error of 15.3% in the ImageNet challenge, more than 10.8 percentage points better than the runner-up. The result is credited with helping start the deep learning revolution.
What bias issues has ImageNet faced?
Studies found more than 6% of labels in the ImageNet-1k validation set are wrong and about 10% of the set contains ambiguous or erroneous labels. ImageNet later removed thousands of "person" subtree categories and blurred faces across its non-person categories after review found many were "potentially offensive" or not truly visual.