Controlled style-content image generation based on disentangling content and style

One or more computer storage media storing computer-useable instructions that, when used by one or more computing devices, cause the one or more computing devices to perform operations comprising:

• receiving a class label and a domain label;
• matching an input latent vector to a content vector generated based on the class label using coarse layers of a neural network, wherein the coarse layers of the neural network are used to learn coarse layer parameters that align low-resolution images to both a source domain and a target domain in relation to a corresponding class label;
• and
• generating an image using the neural network, wherein the image is based on the input latent vector matched to the content vector using the coarse layers of the neural network and the input latent vector matched to the domain label using the fine layers of the neural network and the image is low-resolution representation of the source domain and the target domain for a class label.

A computing system comprising:

• means for receiving class label and a domain label;
• means for matching a latent vector to a content vector generated based on the class label in relation to coarse network parameters of a neural network, wherein the coarse network parameters control low-resolutions in an image and the class label adjusts coarse layer parameters related to coarse layers;
• means for matching the latent vector to a style vector generated based on the domain label in relation to fine network parameters of a neural network, wherein the fine network parameters control high-resolutions in the image and the domain label adjusts fine layer parameters related to fine layers;
• and
• means for generating an image, based on the latent vector matched to the content vector and the domain label matched to the style vector.