In this paper, we investigate two deep learning approaches to recovering initial temperature profiles from thermographicimages in non-destructive material testing.