ImageNet Large Scale Visual Recognition Challenge
Level 5 - mechanism / opinion, no new human data
Level 5 by design analogy (technical benchmark description and overview; non-clinical)
OpenAlex W2546241758 · doi:10.48550/arxiv.1409.0575
What was done
The authors describe the creation, ground-truth annotation process, and organization of the ImageNet Large Scale Visual Recognition Challenge (ILSVRC), an annual benchmark running from 2010 to 2014 evaluating object category classification and detection across hundreds of categories and millions of images with participation from over fifty institutions. They analyze advances in categorical recognition, assess the state of large-scale computer vision, and compare computer vision accuracy with human accuracy.
What was found
The abstract reports no numerical results, accuracy percentages, or quantitative human-versus-machine comparisons; it outlines categorical breakthroughs and lessons learned over five years of the challenge qualitatively.
Why it matters
ILSVRC served as the defining benchmark for evaluating deep learning and computer vision architectures during a transformative period in artificial intelligence.
Limits
The abstract provides no quantitative error rates, metrics, or detailed experimental parameters. Specific methodologies for measuring human accuracy and evaluating dataset biases are not detailed in the abstract text.
Cited by
- supports Human benchmark error rate on the ImageNet 1,000-class object recognition task was measured at roughly 4%.