FrontierThe story, in brief
Learning to segment images without manually segmented training data
NOBODY TALKING: Everyone focuses on GPT breakthroughs. Nobody talks about Amazon's computer vision advancing without expensive labeled data.
The KeyNews take
Why it matters
Amazon's breakthrough in image segmentation without manual labeling could dramatically reduce AI training costs for enterprises deploying computer vision at scale.
The key facts
3 to knowState-of-the-art segmentation results achieved
Uses only coarse bounding-box labels instead of manual segmentation
Published September 2020
Go to the source
Amazon Scienceamazon.science
Publisher excerpt: Machine learning method relies on coarse “bounding-box” image labels but still delivers state-of-the-art segmentation results.