New AI Method Adapts Image Denoising to Unknown Noise Levels in Real Time

Researchers developed a quantitative flow matching framework that estimates noise levels from images and adjusts denoising trajectories accordingly, improving restoration quality while reducing computation. The approach generalizes across natural, medical, and microscopy images with varying corruption levels, addressing a key limitation of current diffusion models that struggle with mismatched training and inference conditions.