# ImageNet LSVRC 2012

ImageNet LSVRC 2012 was the annual ImageNet Large Scale Visual Recognition Challenge where AlexNet, a convolutional neural network, achieved a top-5 error of 15.3%, outperforming all other entries by over 10.8 percentage points and triggering the deep learning boom.

The ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2012 was the third edition of an annual software contest in which programs compete to correctly classify and detect objects and scenes from the ImageNet database. The 2012 challenge is widely regarded as a turning point in [artificial intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) history, as the winning entry, a [neural network](https://www.wikiprompt.org/wiki/neural-network) called AlexNet, achieved a top-5 error of 15.3% on the test set, more than 10.8 percentage points lower than the runner-up. This dramatic improvement demonstrated the power of [deep learning](https://www.wikiprompt.org/wiki/deep-learning) and [machine learning](https://www.wikiprompt.org/wiki/machine-learning) techniques, sparking a wave of research and investment that continues to shape the field.

The ImageNet project itself was initiated by AI researcher Fei-Fei Li in 2006, with the goal of expanding the data available for training AI algorithms. Li met with Princeton professor Christiane Fellbaum, one of the creators of WordNet, in 2007, and subsequently built ImageNet using the roughly 22,000 nouns of WordNet as a starting point. The database was constructed with the help of crowdsourced annotation via Amazon Mechanical Turk, with labeling starting in July 2008 and ending in April 2010. The effort involved 49,000 workers from 167 countries filtering and labeling over 160 million candidate images, resulting in more than 14 million hand-annotated images across over 20,000 categories.

## History and Development

The idea for ImageNet emerged from Li's observation that most AI research at the time focused on models and algorithms rather than data. She was inspired by a 1987 estimate that the average person recognizes roughly 30,000 different kinds of objects. In 2007, Li met with Christiane Fellbaum to discuss the project, and the meeting led to the decision to build ImageNet starting from WordNet's noun synsets. Li assembled a team of researchers at Princeton, where she was an assistant professor, and they used Amazon Mechanical Turk to classify images.

The original plan called for 10,000 images per category, for 40,000 categories at 400 million images, each verified three times. However, humans can classify at most 2 images per second, and at that rate it was estimated to take 19 human-years of labor without rest. The team presented the database for the first time as a poster at the 2009 Conference on Computer Vision and Pattern Recognition (CVPR) in Florida, titled "ImageNet: A Preview of a Large-scale Hierarchical Dataset."

In 2009, Alex Berg suggested adding object localization as a task. Li approached the PASCAL Visual Object Classes contest for a collaboration, which resulted in the ImageNet Large Scale Visual Recognition Challenge starting in 2010. The challenge uses a trimmed list of 1,000 non-overlapping classes, compared to PASCAL VOC's 20 classes and 19,737 images in 2010.

## Significance for Deep Learning

On 30 September 2012, AlexNet, a convolutional neural network (CNN), achieved a top-5 error of 15.3% in the ImageNet 2012 Challenge. This was more than 10.8 percentage points lower than the runner-up. The use of convolutional neural networks was made feasible by the use of graphics processing units (GPUs) during training, which was an essential ingredient of the deep learning revolution. According to The Economist, "Suddenly people started to pay attention, not just within the AI community but across the technology industry as a whole."

The victory of AlexNet is often cited as the trigger for the deep learning boom. It demonstrated that [neural networks](https://www.wikiprompt.org/wiki/neural-network) could achieve state-of-the-art results on complex visual tasks, leading to rapid adoption of deep learning techniques across various domains, including [generative AI](https://www.wikiprompt.org/wiki/generative-ai) and [large language models](https://www.wikiprompt.org/wiki/large-language-model). In 2015, AlexNet was outperformed by Microsoft's very deep CNN with over 100 layers, which won the ImageNet 2015 contest with a 3.57% error rate on the test set.

Andrej Karpathy estimated in 2014 that with concentrated effort, he could reach a 5.1% error rate, and about 10 people from his lab reached around 12-13% with less effort. It was estimated that with maximal effort, a human could reach 2.4% error.

## Dataset Composition

ImageNet crowdsources its annotation process. Image-level annotations indicate the presence or absence of an object class in an image, such as "there are tigers in this image" or "there are no tigers in this image." Object-level annotations provide a bounding box around the visible part of the indicated object. ImageNet uses a variant of the broad WordNet schema to categorize objects, augmented with 120 categories of dog breeds to showcase fine-grained classification.

In 2012, ImageNet was the world's largest academic user of Mechanical Turk, with the average worker identifying 50 images per minute. The summary statistics given on April 30, 2010, showed a total of 21,841 non-empty synsets, 14,197,122 images, 1,034,908 images with bounding box annotations, and 1.2 million images with SIFT features.

The categories of ImageNet were filtered from WordNet concepts. Each concept is called a "synset" (synonym set), and there were more than 100,000 synsets in WordNet 3.0, with the majority being nouns (80,000+). ImageNet filtered these to 21,841 synsets that are countable nouns that can be visually illustrated. Each synset has a WordNet ID (wnid), which starts with "n" because ImageNet only includes nouns. The categories fall into 9 levels, from level 1 (such as "mammal") to level 9 (such as "German shepherd").

## Image Format and Preprocessing

The images were scraped from online image search engines (Google, Picsearch, MSN, Yahoo, Flickr) using synonyms in multiple languages. ImageNet consists of images in RGB format with varying resolutions. For example, in ImageNet 2012, the "fish" category has resolutions ranging from 4288 x 2848 to 75 x 56. In machine learning, these are typically preprocessed into a standard constant resolution and whitened before further processing by neural networks.

For example, in PyTorch, ImageNet images are by default normalized by dividing the pixel values so that they fall between 0 and 1, then subtracting by [0.485, 0.456, 0.406], then dividing by [0.229, 0.224, 0.225]. These are the mean and standard deviations for ImageNet, so this whitens the input data.

## Labels and Annotations

Each image is labelled with exactly one wnid. Dense SIFT features (raw SIFT descriptors, quantized codewords, and coordinates of each descriptor/codeword) for ImageNet-1K were available for download, designed for bag of visual words. The bounding boxes of objects were available for about 3,000 popular synsets with on average 150 images in each synset. Some images also have attribute annotations, though the full details of these are not specified in the available sources.

---
Source: https://www.wikiprompt.org/wiki/imagenet-lsvrc-2012
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-13T03:50:55.355818+00:00