Interactive View
Chart
Table
Interpretation
What The Learned PDE Shows
The learned diffusion model consistently beats the classical baselines by a wide margin. It is the strongest PDE-style denoiser in the project and stays tied to an interpretable iterative process.
That makes it a good benchmark for structured denoising research even when a more flexible black-box model wins on raw metrics.
What The U-Net Baseline Shows
The plain U-Net is the strongest overall denoiser in the final run. The page keeps that result visible so the comparison stays honest.
The project claim is therefore narrower and stronger: learned anisotropic diffusion narrows the gap to modern neural denoisers while preserving the structure of a classical PDE.
Figures
Final metric summary.
Full method comparison grid.
Representative denoising examples.