> ImageNet_

Before ImageNet, computer vision was starving. Algorithms existed, architectures were proposed, but without enough data to train on, progress was painfully slow. Most image datasets contained tens of thousands of images at best, organized into a handful of categories. Fei-Fei Li, a professor at Stanford who had emigrated from China as a teenager, working at a dry cleaner to help support her family while studying physics at Princeton, saw the fundamental problem clearly: the bottleneck was not algorithms but data. In 2006, she began what many colleagues considered a quixotic project to build the largest image dataset the world had ever seen.

The scale of the ambition was staggering. Li and her team set out to create a dataset covering more than 20,000 categories from WordNet, the lexical database of English, with hundreds or thousands of images per category. The solution to labeling millions of images came from an unlikely source: Amazon Mechanical Turk. By breaking the task into tiny labeling jobs distributed to workers around the world, the team was able to annotate roughly 14 million images at a fraction of what traditional labeling would have cost. At its peak, nearly 50,000 workers from 167 countries contributed to the effort, making ImageNet one of the largest crowd-sourced scientific projects in history.

In 2010, Li and her colleagues launched the ImageNet Large Scale Visual Recognition Challenge (ILSVRC), an annual competition that invited researchers to build the best image classifier on a subset of 1,000 categories. The first few years saw modest improvements using traditional computer vision techniques like SIFT features and support vector machines. Error rates hovered around 25-28%. Then in 2012, everything changed when a deep learning entry shattered the competition, but that is a story for the next chapter.

ImageNet's impact transcended any single competition. It validated the now-famous maxim that "data is the new oil," demonstrating that the quality and scale of training data could matter more than algorithmic novelty. The dataset became the standard benchmark against which all image recognition systems were measured, catalyzing an explosion of deep learning research. Fei-Fei Li's journey from an immigrant teenager working odd jobs to the scientist who arguably did more than anyone to ignite the deep learning revolution remains one of the most inspiring stories in AI, a testament to the idea that the most transformative breakthroughs sometimes come not from a clever algorithm but from the patient, unglamorous work of building the right foundation.