A convolutional neural style transfer film
|in cooperation with Alexandra Verzier
In this project, we used a convolutional neural network designed for artisc style transfer. This application was first conceived by Gatys et al and further developed by Gene Kogan.
The video footage was collected in the city of Leiden while trying to capture details of the urban environment. We were curious on how the style transfer would act on more abstract input material. During filming as well as post-production, we were purposely trying to locate images that we thought that were interesting or beautiful because of their abstract shape instead of semantic content. Then, the original video was edited and segmented into individual frames. These were then processed through the neural style transfer network using three different style models of which, the first was trained by us and the other two were pretrained models.
Music: "Minotaur Shock - This Plane is Going to Fall"