Friday, November 17, 2023 | 10:00 AM - 11:00 AM
Preston M. Green Hall, Rodin Auditorium, L0120
135 N Skinker Blvd, St. Louis, MO 63112, USA
Image Restoration through Inversion by Direct Iteration (InDI)
Abstract: Deep learning has revolutionized image restoration, but how to train models to restore images that are realistic and faithful to the input remains a challenge. This talk introduces Inversion by Direct Iteration (InDI), a new formulation for supervised image restoration that avoids the "regression to the mean" effect and produces more realistic and detailed images than existing methods. InDI works by gradually improving image quality in small steps, similar to generative denoising diffusion models. It can be applied to virtually any image degradation, given paired training data. We show that InDI achieves state-of-the-art results on a variety of image restoration tasks, including motion and out-of-focus deblurring, super-resolution, compression artifact removal, and denoising. We conclude by discussing the current status and open challenges in generative image restoration and showcasing the newly introduced Unblur feature in the Google Pixel 7/Pro.