October 31, 2023, Tuesday, 03:00PM (GMT+03.00, Moscow)

Abstract:

In the presentation, we will explore the application of the Fréchet Inception Distance (FID) metric for assessing the quality of historical image reconstruction. FID is an effective quantitative method for comparing visual feature distributions in the absence of ground truth data or true values, allowing for the evaluation of content generation quality. The presentation will provide an introduction to the FID concept, describe its mathematical framework and operating principles, as well as demonstrate how FID can be utilized for assessing image quality. We plan to discuss practical aspects of using FID in real-life restoration projects, including parameter selection and the interpretation of results. The aim of the presentation is to introduce FID as a powerful tool for assessing the quality of historical image restoration and to discuss its potential for objective quantitative evaluation of the restoration process.