# US 12,602,914 B2 · Providing user guidance to use and train a generative adversarial network

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,602,914 B2
- Document ID: US12602914B2
- Status: Granted
- Kind: B2
- Application number: 18/358,375
- Filed: 2023-07-25
- Published: 2025-01-30
- Granted: 2026-04-14
- Earliest claimed: 2023-07-25
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Sarbajit K. Rakshit; Sudhanshu Sekher Sar; Sudheesh S. Kairali; Satyam Jakkula
- Web page: https://patentgrep.com/patent/US12602914B2

## CPC

- G06V 10/7788: {the supervisor being a human, e.g. interactive learning with a human teacher}
- G06V 10/82: using neural networks
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

Provided are a computer program product, system, and method for providing user guidance to use and train a generative adversarial network. A discriminator, implementing machine learning, is trained to output a high probability from input comprising an input image comprising a real image representing a desirable design item superimposed on a body in a pose represented in the input image, a context of the input image and a pose of the body represented in the input image. A generator, implementing machine learning, generates an output image based on an input context, an input pose, and random noise. The discriminator outputs a probability the output image, from the generator, represents a real image having a desirable design item from input comprising the output image, the input context, and the input pose. The generator is trained to output the output image for the input pose and the input context with the probability.

## Claims

### Claim 1

A computer program product for providing user guidance to use and train a generative adversarial network, the computer program product comprising a computer readable storage medium having computer readable program code embodied therein that is executable to perform operations, the operations comprising:
training a discriminator, implementing machine learning, to output a high probability from input comprising an input image comprising a real image representing a desirable design item superimposed on a body in a pose represented in the input image, metadata of a context of the input image, and metadata of the pose of the body represented in the input image, wherein the metadata of the pose indicates positions of body keypoints that represent the pose on which the desirable design item will be superimposed, wherein the desirable design item comprises a design item desirable to a target audience;
executing a generator, implementing machine learning, to generate an output image based on an input context of a design item to generate, an input pose of a body on which the design item will be superimposed in the output image, and random noise;
outputting, by the discriminator, a probability the output image, from the generator, represents a real image having a desirable design item from input comprising the output image, the input context, and the input pose; and
training the generator to output the output image for the input pose and the input context with the probability outputted by the discriminator.

### Claim 2 (depends on 1)

The computer program product of claim 1, wherein the operations further comprise:
generating, during training, by a computer process, input sets, wherein an input set includes a training pose, a training context, and random nose; and
inputting the input sets into the generator during training to generate training output images for the input sets, wherein each output image is inputted into the discriminator with the training pose and the training context used to generate the training output images to determine probabilities the output images comprise real images representing desirable design items.

### Claim 3 (depends on 1)

The computer program product of claim 1, wherein the training the discriminator comprises:
providing a training data set of training images that have been labeled as desirable to the target audience, wherein the training images represent bodies clad with design items;
inputting the training images into a pose estimation model to determine poses of bodies represented in the training images; and
inputting the training images into a context model to determine contexts for the training images, wherein the discriminator is trained to output the high probability for input comprising the training images, the determined poses, and the determined contexts.

### Claim 4 (depends on 1)

The computer program product of claim 1, wherein the operations further comprise:
receiving a user selected pose and a user selected context in human understandable form;
converting the user selected pose and the user selected context to a converted pose and a converted context, respectively; and
inputting the user selected pose and the user selected context to the generator to generate the output image.

### Claim 5 (depends on 1)

The computer program product of claim 1, wherein the operations further comprise:
inputting the output image, the pose, and the context used to generate the output image into a materials model, implementing machine learning, to output materials to use to manufacture the design item represented in the output image.

### Claim 6 (depends on 1)

The computer program product of claim 1, wherein the operations further comprise:
inputting a user selected pose and a user selected context into the generator to generate a product research image representing a design item superimposed onto a body in the user selected pose;
posting the product research image onto a social network; and
receiving feedback scores from the social network indicating desirability ratings of the posted product research image, wherein the discriminator and the generator are trained based on the product research image, the user selected context, the user selected pose, and an aggregated feedback score from the received feedback scores for the posted product research image.

### Claim 7 (depends on 6)

The computer program product of claim 6, wherein the operations further comprise:
generating, with the generator, a plurality of product research images from multiple instances of user selected poses and user selected contexts; and
arrange the product research images in a portfolio to publish on the social network to receive feedback scores for the plurality of product research images, wherein the generator and the discriminator are trained from the product research images, the user selected contexts, the user selected poses, and aggregated feedback scores for the product research images.

### Claim 8 (depends on 6)

The computer program product of claim 6, wherein the training the discriminator further comprises:
determining whether the aggregated feedback score indicates that the design item represented in the product research image is desirable or not desirable to reviewers on the social network, wherein the training the discriminator comprises training the discriminator to output the high probability from input comprising the product research image, the user selected context, and the user selected pose in response to the aggregated feedback score indicating that the design item in the product research image is desirable; and
training the discriminator to output a low probability from input comprising the product research image, the user selected context, and the user selected pose in response to the aggregated feedback score indicating that the design item in the product research image is not desirable.

### Claim 9 (depends on 6)

The computer program product of claim 6, wherein the training the generator further comprises:
training the generator to output the posted product research image with the high probability for inputs comprising the user selected context and the user selected pose in response to the aggregated feedback score from the social network indicating a high desirability of the design item in the posted product research image; and
training the generator to output the posted product research image with a low probability for inputs comprising the user selected context and the user selected pose in response to the aggregated feedback score from the social network indicating a low desirability of the design item in the posted product research image.

### Claim 10 (depends on 1)

The computer program product of claim 1, wherein the metadata of the pose comprises a pose mask of the body keypoints, wherein the metadata of the context indicates demographics of a person represented in the input image and the output image in the pose, a style of design items superimposed over the body of the person in the pose, and an environment in which the person is situated.

### Claim 11

A system for providing user guidance to use and train a generative adversarial network, comprising:
a discriminator;
a generator;
a computer readable storage medium having computer readable program code embodied therein that is executable to perform operations, the operations comprising:
training the discriminator, implementing machine learning, to output a high probability from input comprising an input image comprising a real image representing a desirable design item superimposed on a body in a pose represented in the input image, metadata of a context of the input image, and metadata of the pose of the body represented in the input image, wherein the metadata of the pose indicates positions of body keypoints that represent the pose on which the desirable design item will be superimposed, wherein the desirable design item comprises a design item desirable to a target audience;
executing the generator, implementing machine learning, to generate an output image based on an input context of a design item to generate, an input pose of a body on which the design item will be superimposed in the output image, and random noise;
outputting, by the discriminator, a probability the output image, from the generator, represents a real image having a desirable design item from input comprising the output image, the input context, and the input pose; and
training the generator to output the output image for the input pose and the input context with the probability outputted by the discriminator.

### Claim 12 (depends on 11)

The system of claim 11, wherein the operations further comprise:
generating, during training, by a computer process, input sets, wherein an input set includes a training pose, a training context, and random nose; and
inputting the input sets into the generator during training to generate training output images for the input sets, wherein each output image is inputted into the discriminator with the training pose and the training context used to generate the training output images to determine probabilities the output images comprise real images representing desirable design items.

### Claim 13 (depends on 11)

The system of claim 11, wherein the training the discriminator further comprises:
providing a training data set of training images that have been labeled as desirable to the target audience, wherein the training images represent bodies clad with design items;
inputting the training images into a pose estimation model to determine poses of bodies represented in the training images; and
inputting the training images into a context model to determine contexts for the training images, wherein the discriminator is trained to output the high probability for input comprising the training images, the determined poses, and the determined contexts.

### Claim 14 (depends on 11)

The system of claim 11, wherein the operations further comprise:
inputting a user selected pose and a user selected context into the generator to generate a product research image representing a design item superimposed onto a body in the user selected pose;
posting the product research image onto a social network; and
receiving feedback scores from the social network indicating desirability ratings of the posted product research image, wherein the discriminator and the generator are trained based on the product research image, the user selected context, the user selected pose, and an aggregated feedback score from the received feedback score for the posted product research image.

### Claim 15 (depends on 14)

The system of claim 14, wherein the training the generator further comprises:
training the generator to output the posted product research image with the high probability for inputs comprising the user selected context and the user selected pose in response to the aggregated feedback score from the social network indicating a high desirability of the design item in the posted product research image; and
training the generator to output the posted product research image with a low probability for inputs comprising the user selected context and the user selected pose in response to the aggregated feedback score from the social network indicating a low desirability of the design item in the posted product research image.

### Claim 16

A method for providing user guidance to use and train a generative adversarial network, comprising:
training a discriminator, implementing machine learning, to output a high probability from input comprising an input image comprising a real image representing a desirable design item superimposed on a body in a pose represented in the input image, metadata of a context of the input image, and metadata of the pose of the body represented in the input image, wherein the metadata of the pose indicates positions of body keypoints that represent the pose on which the desirable design item will be superimposed, wherein the desirable design item comprises a design item desirable to a target audience;
executing a generator, implementing machine learning, to generate an output image based on an input context of a design item to generate, an input pose of a body on which the design item will be superimposed in the output image, and random noise;
outputting, by the discriminator, a probability the output image, from the generator, represents a real image having a desirable design item from input comprising the output image, the input context, and the input pose; and
training the generator to output the output image for the input pose and the input context with the probability outputted by the discriminator.

### Claim 17 (depends on 16)

The method of claim 16, further comprising:
generating, during training, by a computer process, input sets, wherein an input set includes a training pose, a training context, and random nose; and
inputting the input sets into the generator during training to generate training output images for the input sets, wherein each output image is inputted into the discriminator with the training pose and the training context used to generate the training output images to determine probabilities the output images comprise real images representing desirable design items.

### Claim 18 (depends on 16)

The method of claim 16, wherein the training the discriminator further comprises:
providing a training data set of training images that have been labeled as desirable to the target audience, wherein the training images represent bodies clad with design items;
inputting the training images into a pose estimation model to determine poses of bodies represented in the training images; and
inputting the training images into a context model to determine contexts for the training images, wherein the discriminator is trained to output the high probability for input comprising the training images, the determined poses, and the determined contexts.

### Claim 19 (depends on 16)

The method of claim 16, further comprising:
inputting a user selected pose and a user selected context into the generator to generate a product research image representing a design item superimposed onto a body in the user selected pose;
posting the product research image onto a social network; and
receiving feedback scores from the social network indicating desirability ratings of the posted product research image, wherein the discriminator and the generator are trained based on the product research image, the user selected context, the user selected pose, and an aggregated feedback score from the received feedback score for the posted product research image.

### Claim 20 (depends on 19)

The method of claim 19, wherein the training the generator further comprises:
training the generator to output the posted product research image with the high probability for inputs comprising the user selected context and the user selected pose in response to the aggregated feedback score from the social network indicating a high desirability of the design item in the posted product research image; and
training the generator to output the posted product research image with a low probability for inputs comprising the user selected context and the user selected pose in response to the aggregated feedback score from the social network indicating a low desirability of the design item in the posted product research image.