ImageNet

The ImageNet project is a large visual database designed for use in visual object recognition software research. More than 14 million^[1][2] images have been hand-annotated by the project to indicate what objects are pictured and in at least one million of the images, bounding boxes are also provided.^[3] ImageNet contains more than 20,000 categories,^[2] with a typical category, such as "balloon" or "strawberry", consisting of several hundred images.^[4] The database of annotations of third-party image URLs is freely available directly from ImageNet, though the actual images are not owned by ImageNet.^[5] Since 2010, the ImageNet project runs an annual software contest, the ImageNet Large Scale Visual Recognition Challenge (ILSVRC), where software programs compete to correctly classify and detect objects and scenes. The challenge uses a "trimmed" list of one thousand non-overlapping classes.^[6]

History

AI researcher Fei-Fei Li began working on the idea for ImageNet in 2006. At a time when most AI research focused on models and algorithms, Li wanted to expand and improve the data available to train AI algorithms.^[7] In 2007, Li met with Princeton professor Christiane Fellbaum, one of the creators of WordNet , to discuss the project. As a result of this meeting, Li went on to build ImageNet starting from the roughly 22,000 nouns of WordNet and using many of its features.^[8] She was also inspired by a 1987 estimate^[9] that the average person recognizes roughly 30,000 different kinds of objects.^[10]

As an assistant professor at Princeton, Li assembled a team of researchers to work on the ImageNet project. They used Amazon Mechanical Turk to help with the classification of images. Labeling started in July 2008 and ended in April 2010. It took 2.5 years to complete the labeling.^[8] They had enough budget to have each of the 14 million images labelled three times.^[10]

The original plan called for 10,000 images per category, for 40,000 categories at 400 million images, each verified 3 times. They found that humans can classify at most 2 images/sec. At this rate, it was estimated to take 19 human-years of labor (without rest).^[11]

They presented their 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".^{[12][8][13][14]} The poster was reused at Vision Sciences Society 2009.^[15]

In 2009, Alex Berg suggested adding object localization as a task. Li approached PASCAL Visual Object Classes contest in 2009 for a collaboration. It resulted in the subsequent ImageNet Large Scale Visual Recognition Challenge starting in 2010, which has 1000 classes and object localization, as compared to PASCAL VOC which had just 20 classes and 19,737 images (in 2010).^[6][8]

Significance for deep learning

On 30 September 2012, a convolutional neural network (CNN) called AlexNet^[16] achieved a top-5 error of 15.3% in the ImageNet 2012 Challenge, more than 10.8 percentage points lower than that of the runner up. Using convolutional neural networks was feasible due to the use of graphics processing units (GPUs) during training,^[16] 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."^[4][17][18]

In 2015, AlexNet was outperformed by Microsoft's very deep CNN with over 100 layers, which won the ImageNet 2015 contest.^[19]

Dataset

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.^[6]

In 2012, ImageNet was the world's largest academic user of Mechanical Turk. The average worker identified 50 images per minute.^[2]

The original plan of the full ImageNet would have roughly 50M clean, diverse and full resolution images spread over approximately 50K synsets.^[13] This was not achieved.

The summary statistics given on April 30, 2010:^[20]

• Total number of non-empty synsets: 21841
• Total number of images: 14,197,122
• Number of images with bounding box annotations: 1,034,908
• Number of synsets with SIFT features: 1000
• Number of images with SIFT features: 1.2 million

Categories

The categories of ImageNet were filtered from the WordNet concepts. Each concept, since it can contain multiple synonyms (for example, "kitty" and "young cat"), so each concept is called a "synonym set" or "synset". There were more than 100,000 synsets in WordNet 3.0, majority of them are nouns (80,000+). The ImageNet dataset filtered these to 21,841 synsets that are countable nouns that can be visually illustrated.

Each synset in WordNet 3.0 has a "WordNet ID" (wnid), which is a concatenation of part of speech and an "offset" (a unique identifying number). Every wnid starts with "n" because ImageNet only includes nouns. For example, the wnid of synset "dog, domestic dog, Canis familiaris" is "n02084071".^[21]

The categories in ImageNet fall into 9 levels, from level 1 (such as "mammal") to level 9 (such as "German shepherd").^[11]

Image format

The images were scraped from online image search (Google, Picsearch, MSN, Yahoo, Flickr, etc) using synonyms in multiple languages. For example: German shepherd, German police dog, German shepherd dog, Alsatian, ovejero alemán, pastore tedesco, 德国牧羊犬.^[22]

ImageNet consists of images in RGB format with varying resolutions. For example, in ImageNet 2012, "fish" category, the resolution ranges 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 these whitens the input data.^[23]

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.^[24]

The bounding boxes of objects were available for about 3000 popular synsets^[25] with on average 150 images in each synset.^[26]

Furthermore, some images have attributes. They released 25 attributes for ~400 popular synsets:^[27][28]

• Color: black, blue, brown, gray, green, orange, pink, red, violet, white, yellow
• Pattern: spotted, striped
• Shape: long, round, rectangular, square
• Texture: furry, smooth, rough, shiny, metallic, vegetation, wooden, wet

ImageNet-21K

The full original dataset is referred to as ImageNet-21K. ImageNet-21k contains 14,197,122 images divided into 21,841 classes. Some papers round this up and name it ImageNet-22k.^[29]

The full ImageNet-21k was released in Fall of 2011, as fall11_whole.tar. There is no official train-validation-test split for ImageNet-21k. Some classes contain only 1-10 samples, while others contain thousands.^[29]

ImageNet-1K

There are various subsets of the ImageNet dataset used in various context, sometimes referred to as "versions".^[16]

One of the most highly used subset of ImageNet is the "ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2012–2017 image classification and localization dataset". This is also referred to in the research literature as ImageNet-1K or ILSVRC2017, reflecting the original ILSVRC challenge that involved 1,000 classes. ImageNet-1K contains 1,281,167 training images, 50,000 validation images and 100,000 test images.^[30]

Each category in ImageNet-1K is a leaf category, meaning that there are no child nodes below it, unlike ImageNet-21K. For example, in ImageNet-21K, there are some images categorized as simply "mammal", whereas in ImageNet-1K, there are only images categorized as things like "German shepherd", since there are no child-words below "German shepherd".^[22]

Later developments

In 2021 winter, ImageNet-21k was updated. 2,702 categories in the "person" subtree were filtered to prevent "problematic behaviors" in a trained model. In 2021, ImageNet-1k was updated by annotating faces appearing in the 997 non-person categories. They found training models on the dataset with these faces blurred caused minimal loss in performance.^[31]

ImageNetV2 was a new dataset containing three test sets with 10,000 each, constructed by the same methodology as the original ImageNet.^[32]

ImageNet-21K-P was a filtered and cleaned subset of ImageNet-21K, with 12,358,688 images from 11,221 categories.^[29]

Table of datasets

Name	Published	Classes	Training	Validation	Test	Size
PASCAL VOC	2005	20
ImageNet-1K	2009	1,000	1,281,167	50,000	100,000	130 GB
ImageNet-21K	2011	21,841	14,197,122			1.31 TB
ImageNetV2	2019				30,000
ImageNet-21K-P	2021	11,221	11,797,632		561,052

History of the ImageNet challenge

Error rate history on ImageNet (showing best result per team and up to 10 entries per year)

The ILSVRC aims to "follow in the footsteps" of the smaller-scale PASCAL VOC challenge, established in 2005, which contained only about 20,000 images and twenty object classes.^[6] To "democratize" ImageNet, Fei-Fei Li proposed to the PASCAL VOC team a collaboration, beginning in 2010, where research teams would evaluate their algorithms on the given data set, and compete to achieve higher accuracy on several visual recognition tasks.^[8]

The resulting annual competition is now known as the ImageNet Large Scale Visual Recognition Challenge (ILSVRC). The ILSVRC uses a "trimmed" list of only 1000 image categories or "classes", including 90 of the 120 dog breeds classified by the full ImageNet schema.^[6]

The 2010s saw dramatic progress in image processing.

The first competition in 2010 had 11 participating teams. The winning team was a linear support vector machine (SVM). The features are a dense grid of HoG and LBP, sparsified by local coordinate coding and pooling.^[33] It achieved 52.9% in classification accuracy and 71.8% in top-5 accuracy. It was trained for 4 days on three 8-core machines (dual quad-core 2GHz Intel Xeon CPU).^[34]

The second competition in 2011 had fewer teams, with another SVM winning at top-5 error rate 25%.^[10] The winning team was XRCE by Florent Perronnin, Jorge Sanchez. The system was another linear SVM, running on quantized^[35] Fisher vectors.^[36][37] It achieved 74.2% in top-5 accuracy.

In 2012, a deep convolutional neural net called AlexNet achieved 84.7% in top-5 accuracy, a great leap forward.^[38] In the next couple of years, top-5 accuracy grew to above 90%. While the 2012 breakthrough "combined pieces that were all there before", the dramatic quantitative improvement marked the start of an industry-wide artificial intelligence boom.^[4]

By 2014, more than fifty institutions participated in the ILSVRC.^[6] In 2017, 29 of 38 competing teams had greater than 95% accuracy.^[39] In 2017 ImageNet stated it would roll out a new, much more difficult challenge in 2018 that involves classifying 3D objects using natural language. Because creating 3D data is more costly than annotating a pre-existing 2D image, the dataset is expected to be smaller. The applications of progress in this area would range from robotic navigation to augmented reality.^[1]

By 2015, researchers at Microsoft reported that their CNNs exceeded human ability at the narrow ILSVRC tasks.^[19][40] However, as one of the challenge's organizers, Olga Russakovsky, pointed out in 2015, the contest is over only 1000 categories; humans can recognize a larger number of categories, and also (unlike the programs) can judge the context of an image.^[41]

Bias in ImageNet

It is estimated that over 6% of labels in the ImageNet-1k validation set are wrong.^[42] It is also found that around 10% of ImageNet-1k contains ambiguous or erroneous labels, and that, when presented with a model's prediction and the original ImageNet label, human annotators prefer the prediction of a state of the art model in 2020 trained on the original ImageNet, suggesting that ImageNet-1k has been saturated.^[43]

A study of the history of the multiple layers (taxonomy, object classes and labeling) of ImageNet and WordNet in 2019 described how bias^{[clarification needed]} is deeply embedded in most classification approaches for all sorts of images.^{[44][45][46][47]} ImageNet is working to address various sources of bias.^[48]

One downside of WordNet use is the categories may be more "elevated" than would be optimal for ImageNet: "Most people are more interested in Lady Gaga or the iPod Mini than in this rare kind of diplodocus."^{[clarification needed]}

See also

• Computer vision
• List of datasets for machine learning research
• WordNet

References

1. "New computer vision challenge wants to teach robots to see in 3D". New Scientist. 7 April 2017. Retrieved 3 February 2018.
2. Markoff, John (19 November 2012). "For Web Images, Creating New Technology to Seek and Find". The New York Times. Retrieved 3 February 2018.
3. "ImageNet". 7 September 2020. Archived from the original on 7 September 2020. Retrieved 11 October 2022.
4. "From not working to neural networking". The Economist. 25 June 2016. Retrieved 3 February 2018.
5. "ImageNet Overview". ImageNet. Retrieved 15 October 2022.
6. Olga Russakovsky*, Jia Deng*, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg and Li Fei-Fei. (* = equal contribution) ImageNet Large Scale Visual Recognition Challenge. IJCV, 2015.
7. Hempel, Jesse (13 November 2018). "Fei-Fei Li's Quest to Make AI Better for Humanity". Wired. Retrieved 5 May 2019. When Li, who had moved back to Princeton to take a job as an assistant professor in 2007, talked up her idea for ImageNet, she had a hard time getting faculty members to help out. Finally, a professor who specialized in computer architecture agreed to join her as a collaborator.
8. Gershgorn, Dave (26 July 2017). "The data that transformed AI research—and possibly the world". Quartz. Atlantic Media Co. Retrieved 26 July 2017. Having read about WordNet's approach, Li met with professor Christiane Fellbaum, a researcher influential in the continued work on WordNet, during a 2006 visit to Princeton.
9. Biederman, Irving (1987). "Recognition-by-components: A theory of human image understanding". Psychological Review. 94 (2): 115–117. doi:10.1037/0033-295x.94.2.115. ISSN 0033-295X. PMID 3575582.
10. Lee, Timothy B. (11 November 2024). "How a stubborn computer scientist accidentally launched the deep learning boom". Ars Technica. Retrieved 12 November 2024.
11. Li, F-F. ImageNet. "Crowdsourcing, benchmarking & other cool things." CMU VASC Semin 16 (2010): 18-25.
12. "CVPR 2009: IEEE Computer Society Conference on Computer Vision and Pattern Recognition". tab.computer.org. Retrieved 13 November 2024.
13. Deng, Jia; Dong, Wei; Socher, Richard; Li, Li-Jia; Li, Kai; Fei-Fei, Li (2009), "ImageNet: A Large-Scale Hierarchical Image Database" (PDF), 2009 conference on Computer Vision and Pattern Recognition, archived from the original (PDF) on 15 January 2021, retrieved 26 July 2017
14. Li, Fei-Fei (23 March 2015), How we're teaching computers to understand pictures, retrieved 16 December 2018
15. Deng, Jia, et al. "Construction and analysis of a large scale image ontology." Vision Sciences Society 186.2 (2009).
16. Krizhevsky, Alex; Sutskever, Ilya; Hinton, Geoffrey E. (June 2017). "ImageNet classification with deep convolutional neural networks" (PDF). Communications of the ACM. 60 (6): 84–90. doi:10.1145/3065386. ISSN 0001-0782. S2CID 195908774. Retrieved 24 May 2017.
17. "Machines 'beat humans' for a growing number of tasks". Financial Times. 30 November 2017. Retrieved 3 February 2018.
18. Gershgorn, Dave (18 June 2018). "The inside story of how AI got good enough to dominate Silicon Valley". Quartz. Retrieved 10 December 2018.
19. He, Kaiming; Zhang, Xiangyu; Ren, Shaoqing; Sun, Jian (2016). "Deep Residual Learning for Image Recognition". 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 770–778. arXiv:1512.03385. doi:10.1109/CVPR.2016.90. ISBN 978-1-4673-8851-1. S2CID 206594692.
20. "ImageNet Summary and Statistics (updated on April 30, 2010)". 15 January 2013. Archived from the original on 15 January 2013. Retrieved 13 November 2024.
21. "ImageNet API documentation". 22 January 2013. Archived from the original on 22 January 2013. Retrieved 13 November 2024.
22. Berg, Alex, Jia Deng, and L. Fei-Fei. "Large scale visual recognition challenge 2010." November 2010.
23. "std and mean for image normalization different from ImageNet · Issue #20 · openai/CLIP". GitHub. Retrieved 19 September 2024.
24. "ImageNet". 5 April 2013. Archived from the original on 5 April 2013. Retrieved 13 November 2024.
25. https://web.archive.org/web/20181030191122/http://www.image-net.org/api/text/imagenet.sbow.obtain_synset_list
26. "ImageNet". Archived from the original on 5 April 2013.
27. "ImageNet". Archived from the original on 22 December 2019.
28. Russakovsky, Olga; Fei-Fei, Li (2012). "Attribute Learning in Large-Scale Datasets". In Kutulakos, Kiriakos N. (ed.). Trends and Topics in Computer Vision. Lecture Notes in Computer Science. Vol. 6553. Berlin, Heidelberg: Springer. pp. 1–14. doi:10.1007/978-3-642-35749-7_1. ISBN 978-3-642-35749-7.
29. Ridnik, Tal; Ben-Baruch, Emanuel; Noy, Asaf; Zelnik-Manor, Lihi (5 August 2021). "ImageNet-21K Pretraining for the Masses". arXiv:2104.10972 [cs.CV].
30. "ImageNet". www.image-net.org. Retrieved 19 October 2022.
31. "An Update to the ImageNet Website and Dataset". www.image-net.org. Retrieved 13 November 2024.
32. Recht, Benjamin; Roelofs, Rebecca; Schmidt, Ludwig; Shankar, Vaishaal (24 May 2019). "Do ImageNet Classifiers Generalize to ImageNet?". Proceedings of the 36th International Conference on Machine Learning. PMLR: 5389–5400.
33. ImageNet classification: fast descriptor coding and large-scale SVM training
34. Lin, Yuanqing; Lv, Fengjun; Zhu, Shenghuo; Yang, Ming; Cour, Timothee; Yu, Kai; Cao, Liangliang; Huang, Thomas (June 2011). "Large-scale image classification: Fast feature extraction and SVM training". CVPR 2011. IEEE. pp. 1689–1696. doi:10.1109/cvpr.2011.5995477. ISBN 978-1-4577-0394-2.
35. Sanchez, Jorge; Perronnin, Florent (June 2011). "High-dimensional signature compression for large-scale image classification". CVPR 2011. IEEE. pp. 1665–1672. doi:10.1109/cvpr.2011.5995504. ISBN 978-1-4577-0394-2.
36. Perronnin, Florent; Sánchez, Jorge; Mensink, Thomas (2010). "Improving the Fisher Kernel for Large-Scale Image Classification". In Daniilidis, Kostas; Maragos, Petros; Paragios, Nikos (eds.). Computer Vision – ECCV 2010. Lecture Notes in Computer Science. Vol. 6314. Berlin, Heidelberg: Springer. pp. 143–156. doi:10.1007/978-3-642-15561-1_11. ISBN 978-3-642-15561-1.
37. "XRCE@ILSVRC2011: Compressed Fisher vectors for LSVR", Florent Perronnin and Jorge Sánchez, Xerox Research Centre Europe (XRCE)
38. https://www.image-net.org/challenges/LSVRC/2012/results
39. Gershgorn, Dave (10 September 2017). "The Quartz guide to artificial intelligence: What is it, why is it important, and should we be afraid?". Quartz. Retrieved 3 February 2018.
40. Markoff, John (10 December 2015). "A Learning Advance in Artificial Intelligence Rivals Human Abilities". The New York Times. Retrieved 22 June 2016.
41. Aron, Jacob (21 September 2015). "Forget the Turing test – there are better ways of judging AI". New Scientist. Retrieved 22 June 2016.
42. Northcutt, Curtis G.; Athalye, Anish; Mueller, Jonas (7 November 2021), Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks, arXiv:2103.14749
43. Beyer, Lucas; Hénaff, Olivier J.; Kolesnikov, Alexander; Zhai, Xiaohua; Oord, Aäron van den (12 June 2020), Are we done with ImageNet?, arXiv:2006.07159
44. "The Viral App That Labels You Isn't Quite What You Think". Wired. ISSN 1059-1028. Retrieved 22 September 2019.
45. Wong, Julia Carrie (18 September 2019). "The viral selfie app ImageNet Roulette seemed fun – until it called me a racist slur". The Guardian. ISSN 0261-3077. Retrieved 22 September 2019.
46. Crawford, Kate; Paglen, Trevor (19 September 2019). "Excavating AI: The Politics of Training Sets for Machine Learning". -. Retrieved 22 September 2019.
47. Lyons, Michael (24 December 2020). "Excavating "Excavating AI": The Elephant in the Gallery". arXiv:2009.01215. doi:10.5281/zenodo.4037538. {{cite journal}}: Cite journal requires |journal= (help)
48. "Towards Fairer Datasets: Filtering and Balancing the Distribution of the People Subtree in the ImageNet Hierarchy". image-net.org. 17 September 2019. Retrieved 22 September 2019.

External links

• Official website