Finding complementary digital images using a conditional generative adversarial network

A method for retrieving one or more output images, implemented by one or more computing devices, comprising:

• identifying a first-part image in response to at least one input signal provided by a user, using an input device;
• receiving first-part image information that describes the first-part image;
• using a generator component implemented by said one or more computing devices to transform, using a neural network, the first-part image information into one or more instances of second-part generated image information, each instance of second-part generated image information complementing the first-part image information and being fake image information created by the generator component rather than being retrieved from a preexisting empirical distribution of images;
• retrieving one or more second-part output images from a data store, based on said one more instances of second-part generated image information;
• generating a user interface presentation that presents the first-part image and said one or more second-part output images, the user interface presentation including one or more graphical prompts that allow a user to select any of said one or more second-part output images;
• and
• displaying the user interface presentation on a display device, for presentation to the user,
• the generator component having been trained by a computer-implemented training system using a conditional generative adversarial network (cGAN) by:
• identifying a set of composite images, each composite image being composed of at least a particular first-part image and a particular second-part image;
• segmenting the composite images to generate a plurality of first-part extracted images and a plurality of second-part extracted images;
• and
• using the cGAN to train the generator component based on the plurality of first-part extracted images and the plurality of second-part extracted images.

A computer-readable storage medium for storing computer-readable instructions, the computer-readable instructions, when executed by one or more hardware processors, performing a method that comprises:

• identifying, using a matching component, a set of composite images, each composite image being composed of at least a first-part image and a second-part image;
• segmenting the composite images to generate a plurality of first-part extracted images and a plurality of second-part extracted images;
• and
• using a conditional generative adversarial network (cGAN) to train a generator component based on the plurality of first-part extracted images and the plurality of second-part extracted images,
• wherein said using the cGAN comprises, for a plurality of instances of first-part image information:
• using the generator component to map an instance of first-part image information, together with an instance of random information taken from a distribution, into an instance of second-part generated image information;
• using a discriminator component to provide a score which reflects a level of confidence that the instance of second-part generated image information is representative of a real preexisting second-part image;
• and
• based at least on the score, adjusting weight values of the generator component and the discriminator component,
• the matching component, generator component, and discriminator component being implemented by the computer-readable instructions.