Quantifying images’ “conceptual similarity”
Amazon just formalized what humans could only judge by gut: a mathematical way to measure if two images are 'conceptually similar.'

Why it matters
Amazon Science developed a quantifiable method using vision-language models to standardize image comparison—moving subjective human evaluation into objective, measurable metrics. This matters for enterprises building scalable visual search, content moderation, and asset management systems.
The key facts
9 to knowNew method quantifies 'conceptual similarity' between images
Leverages vision-language models for formalization
Replaces previously subjective human judgment with objective measurement
Published by Amazon Science
September 2024
New method leverages vision-language models
Formalizes previously subjective human judgment task
Applies to image comparison and conceptual similarity measurement
Potential automation of manual visual assessment workflows
Go to the source
Amazon Scienceamazon.science
Publisher excerpt: New method leverages vision-language models to formalize a comparison that had previously required human judgment.

