Evaluating Unsupervised Denoising Requires Unsupervised Metrics

Abstract:  Denoising is a crucial challenge in imaging applications. In recent years, unsupervised methods based on deep neural networks have demonstrated impressive performance on natural-image benchmarks. However, no quantitative metrics are available to evaluate these methods in an unsupervised fashion. This is highly problematic for the many real-world applications where ground-truth clean images are not readily available. In this talk we introduce two novel metrics: the unsupervised mean squared error (MSE) and the unsupervised peak signal-to-noise ratio (PSNR), which are computed using only noisy data. We provide a theoretical analysis of these metrics, which shows that they are asymptotically consistent estimators of the supervised MSE and PSNR. Controlled numerical experiments with synthetic noise confirm that they provide accurate approximations in practice. In addition, we report experiments on real-world data from two imaging modalities: videos in raw format and transmission electron microscopy.

Bio:  Carlos Fernandez-Granda is an Associate Professor at the Courant Institute of Mathematical Sciences and the Center for Data Science in New York University. His research focuses on the design and analysis of data-science methodology, with a particular focus on machine learning, and its application to medicine, climate science and scientific imaging.

***This event will take place only in the remote/virtual format.

Join Zoom Meeting https: https://umich.zoom.us/j/91414297851

Meeting ID: 914 1429 7851

Passcode: XXXXXX (Will be sent via e-mail to attendees)

Zoom Passcode information is also available upon request to Michele Feldkamp ([email protected]) or Sher Nickrand ([email protected])