#content-loss

Question 1: Select all that apply

Neural style transfer uses a pretrained VGG-19 and three images: a content image, a style image, and a synthesized image that is being optimized. Which statements are correct? Select all that apply.

The content loss compares feature maps of the synthesized image and the content image at a deeper layer1
The style loss compares Gram matrices of the synthesized image and the style image at several layers2
The VGG weights stay fixed; gradient descent changes the pixels of the synthesized image3
The content loss compares feature maps of the content image and the style image4
The style loss compares feature maps position by position, as the content loss does5

Timed test