HOW IT WORKS

The reliability of our image recognition algorithm stems from the image generation domain.
Diffusion-based generative models can be repurposed for image recognition, enabling :

• Higher image recognition reliability
• Out-Of-Distribution image recognition
• No majority label bias
• Image segmentation without human-annotated masks (similarly to https://landing.malt.com/projects/segmentation-bottleneck)
• Abolishing the supervised vs unsupervised anomaly detection dilemma