Parametric composite image harmonization

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

• generating, by a first machine learning model, color curve parameters by providing a down-sampled version of a composite image as input to the first machine learning model;
• generating, by a second machine learning model, a shadow map by providing the down-sampled version of the composite image with the first color curve parameters as input to the second machine learning model;
• up-sampling the color curve parameters to provide up-sampled color curve parameters;
• up-sampling the shadow map to provide an up-sampled shadow map;
• and
• generating a harmonized composite image by applying the up-sampled color curve parameters and the up-sampled shadow map to a foreground of the composite image.

A computer system comprising:

• a processor;
• and
• a computer storage medium storing computer-useable instructions that, when used by the processor, causes the computer system to perform operations comprising:
• receiving a training dataset comprising a plurality of training images;
• and
• training a model to predict color curve parameters and shadow maps for composite images using a reconstruction loss determined using a first training image and an adversarial loss determined using a second training image.