Back to Feed
Benchmarks & Evals / Computer Vision

Building the ImageNet Large Scale Database

Original: ImageNet: A large-scale hierarchical image database

Listen to the summary

Uses a voice available on your device

Audio options
On this page 4 sections

Key Takeaways

  • Built a large scale ontology of images called ImageNet upon the backbone of the WordNet structure.
  • Populated the majority of the 80,000 synsets of WordNet with an average of 500 to 1000 clean and full resolution images.
  • Utilized Amazon Mechanical Turk for the data collection scheme.
  • ImageNet currently contains 12 subtrees with 5247 synsets and 3.2 million images in total.

Summary & Methodology Analysis

How massive Internet image data can be harnessed and organized remains a critical problem for fostering more sophisticated and robust models and algorithms. To address this, the authors built a large scale ontology of images called ImageNet upon the backbone of the WordNet structure. This database populates the majority of the 80,000 synsets of WordNet with an average of 500 to 1000 clean and full resolution images, using Amazon Mechanical Turk for the data collection scheme. Currently, ImageNet contains 12 subtrees with 5247 synsets and 3.2 million images in total. The paper does not specify any further architectural or methodological details beyond these data structures and collection methods. Similarly, the paper does not specify any specific limitations regarding the ImageNet dataset.

Interactive System Flowchart

Click diagram to expand and zoom

Cross-Examination & FAQs

A deeper dive clarifying mechanics, constraints, and baseline evaluations.

Q1. What is the main problem addressed in the paper?

How massive Internet image data can be harnessed and organized remains a critical problem for fostering more sophisticated and robust models and algorithms.

Q2. What is the core contribution of the paper?

The paper introduces ImageNet, a large scale hierarchical image database built upon the backbone of the WordNet structure.

Q3. What tools or platforms were used for data collection?

Amazon Mechanical Turk was utilized for the data collection scheme.

Q4. What structure serves as the backbone for ImageNet?

The WordNet structure serves as the backbone for ImageNet.

Q5. How many synsets of WordNet were populated with images?

The majority of the 80,000 synsets of WordNet were populated.

Q6. How many images are typically assigned to each populated synset?

An average of 500 to 1000 clean and full resolution images populate the synsets.

Q7. What are the current contents and scale of ImageNet according to the key results?

ImageNet currently contains 12 subtrees with 5247 synsets and 3.2 million images in total.

Q8. What limitations of the dataset or method are reported in the paper?

The paper does not specify any limitations.

Q9. Are there any specific hardware or training cost figures provided in the paper?

The paper does not specify any hardware requirements, training costs, or other performance metrics beyond the dataset statistics.

Flag an issue

What is wrong with this summary?

What is wrong?