Transfer a content image into the style of a style image.
-
A pre-trained VGG19 model is employed as feature extactor of content and style of an image.
-
A weighted loss function of content loss, style loss, and total variation loss is used as the optimization objective.
-
Visualize the feature maps to choose content and style layers.
-
Adam optimizer is used for training.
Try the code in google colab if you are interested in. However, you need to prepare some images youselves. I did it by copying the images stored in my google drive.
Only the first 10 feature maps of each conv layer are displayed.