Generation of synthetic images of infants for training and comparison of analysis algorithms
Igor Mikhailov, Igor Gajniyarov, Ksenia I. Kunnikova, I. Tuktareva, Alexander I. Kotyusov, E. Suleymanova, Alexander B. Palchik · 2021
This article is devoted to generating synthetic datasets with computer models of infants with normal and abnormal general movement activity. Our solution is a software for creating data sets similar to real ones for use in video analysis algorithms. One of the important issues in video content analysis is comparison of different algorithms. Image and video dataset are availabel for many tasks like object classification, pose recognition, etc. This allows comparing different algorithms on the same data by various metrics like accuracy or time of work. However, there is no free, public, and large enough dataset for babies at the given instant. The reason for this is both in principle that videos of babies are less common, and because videos of babies are harder to look at from a legal point of view.