US-20260127852-A1 - BALANCED GENERATIVE IMAGE MODEL TRAINING

Abstract

A method for training an image generation model includes receiving a first set of training data comprising multiple training images and corresponding image captions, using the first set of training data to perform a first training process to train the image generation model resulting in a trained image generation model, and defining sensitive categories and protected attributes associated with the training images. The method further includes using the sensitive categories and the protected attributes to generate a second set of training data that is balanced across at least one of the protected attributes with respect to at least one of the sensitive categories, and using the second set of training data to perform a second training process to de-bias the trained image generation model resulting in a de-biased image generation model.

Inventors

• Raúl Gómez Bruballa
• Alessandra Sala

Assignees

• SHUTTERSTOCK, INC.

Dates

Publication Date

20260507

Application Date

20241104

Claims (20)

1. 1 . A method for training an image generation model, comprising: receiving a first set of training data comprising a plurality of training images and corresponding image captions; using the first set of training data, performing a first training process to train the image generation model resulting in a trained image generation model; defining a plurality of sensitive categories associated with the plurality of training images; defining a plurality of protected attributes associated with the plurality of training images; using the plurality of sensitive categories and the plurality of protected attributes, generating a second set of training data that is balanced across at least one of the plurality of protected attributes with respect to at least one of the plurality of sensitive categories; and using the second set of training data, performing a second training process to de-bias the trained image generation model resulting in a de-biased image generation model.
2. 2 . The method of claim 1 , wherein the plurality of training images is a first plurality of training images, and generating the second set of training data comprises: receiving a third set of training data comprising a plurality of refinement images; annotating the plurality of refinement images with a plurality of tags to associate each image in the plurality of refinement images with at least one sensitive category from the plurality of sensitive categories and at least one protected attribute from the plurality of protected attributes; and performing a selection operation on the plurality of refinement images resulting in a plurality of balanced images, wherein the plurality of balanced images are balanced across at least one of the plurality of protected attributes with respect to at least one of the plurality of sensitive categories, and wherein the second set of training data comprises the plurality of balanced images.
3. 3 . The method of claim 2 , wherein annotating the plurality of refinement images comprises performing a classification operation on the plurality of refinement images.
4. 4 . The method of claim 2 , wherein the first set of training data comprises the third set of training data.
5. 5 . The method of claim 1 , wherein the first set of training data comprises the second set of training data.
6. 6 . The method of claim 1 , wherein the image generation model is one of a Generative Adversarial Network (GAN), a Variational Autoencoder (VAE), an autoregressive model, a diffusion-based model, or a transformer-based architecture.
7. 7 . The method of claim 1 , wherein the plurality of sensitive categories comprise one or more of personalities, conditions, actions, income levels, occupations, and socioeconomic status.
8. 8 . The method of claim 1 , wherein the plurality of protected attributes comprise one or more of gender, skin color, race, ethnicity, age, and religion.
9. 9 . A non-transitory computer-readable medium storing a program for training an image generation model, which when executed by a computer, configures the computer to: receive a first set of training data comprising a plurality of training images and corresponding image captions; using the first set of training data, perform a first training process to train the image generation model resulting in a trained image generation model; define a plurality of sensitive categories associated with the plurality of training images; define a plurality of protected attributes associated with the plurality of training images; using the plurality of sensitive categories and the plurality of protected attributes, generate a second set of training data that is balanced across at least one of the plurality of protected attributes with respect to at least one of the plurality of sensitive categories; and using the second set of training data, perform a second training process to de-bias the trained image generation model resulting in a de-biased image generation model.
10. 10 . The non-transitory computer-readable medium of claim 9 , wherein the plurality of training images is a first plurality of training images, and wherein the program, when executed by the computer, further configures the computer to generate the second set of training data by further configuring the computer to: receive a third set of training data comprising a plurality of refinement images; annotate the plurality of refinement images with a plurality of tags to associate each image in the plurality of refinement images with at least one sensitive category from the plurality of sensitive categories and at least one protected attribute from the plurality of protected attributes; and perform a selection operation on the plurality of refinement images resulting in a plurality of balanced images, wherein the plurality of balanced images are balanced across at least one of the plurality of protected attributes with respect to at least one of the plurality of sensitive categories, and wherein the second set of training data comprises the plurality of balanced images.
11. 11 . The non-transitory computer-readable medium of claim 10 , wherein the first set of training data comprises the third set of training data.
12. 12 . The non-transitory computer-readable medium of claim 10 , wherein the program, when executed by the computer, further configures the computer to annotate the plurality of refinement images by further configuring the computer to perform a classification operation on the plurality of refinement images.
13. 13 . The non-transitory computer-readable medium of claim 9 , wherein the first set of training data comprises the second set of training data.
14. 14 . The non-transitory computer-readable medium of claim 9 , wherein the image generation model is one of a Generative Adversarial Network (GAN), a Variational Autoencoder (VAE), an autoregressive model, a diffusion-based model, or a transformer-based architecture.
15. 15 . The non-transitory computer-readable medium of claim 9 , wherein the plurality of sensitive categories comprise one or more of personalities, conditions, actions, income levels, occupations, and socioeconomic status, and wherein the plurality of protected attributes comprise one or more of gender, skin color, race, ethnicity, age, and religion.
16. 16 . A system for training an image generation model, comprising: a processor; and a non-transitory computer-readable medium storing a set of instructions, which when executed by the processor, configure the system to: receive a first set of training data comprising a plurality of training images and corresponding image captions; using the first set of training data, perform a first training process to train the image generation model resulting in a trained image generation model; define a plurality of sensitive categories associated with the plurality of training images; define a plurality of protected attributes associated with the plurality of training images; using the plurality of sensitive categories and the plurality of protected attributes, generate a second set of training data that is balanced across at least one of the plurality of protected attributes with respect to at least one of the plurality of sensitive categories; and using the second set of training data, perform a second training process to de-bias the trained image generation model resulting in a de-biased image generation model.
17. 17 . The system of claim 16 , wherein the plurality of training images is a first plurality of training images, and wherein the instructions, when executed by the processor, further configures the system to generate the second set of training data by further configuring the system to: receive a third set of training data comprising a plurality of refinement images; annotate the plurality of refinement images with a plurality of tags to associate each image in the plurality of refinement images with at least one sensitive category from the plurality of sensitive categories and at least one protected attribute from the plurality of protected attributes; and perform a selection operation on the plurality of refinement images resulting in a plurality of balanced images, wherein the plurality of balanced images are balanced across at least one of the plurality of protected attributes with respect to at least one of the plurality of sensitive categories, and wherein the second set of training data comprises the plurality of balanced images.
18. 18 . The system of claim 17 , wherein the first set of training data comprises the third set of training data.
19. 19 . The system of claim 17 , wherein the instructions, when executed by the processor, further configure the system to annotate the plurality of refinement images by further configuring the system to perform a classification operation on the plurality of refinement images.
20. 20 . The system of claim 16 , wherein the first set of training data comprises the second set of training data.

Description

TECHNICAL FIELD The present disclosure generally relates to image generation, and particularly to training of image generative models. BACKGROUND The data requirements to train machine learning and/or artificial intelligence (AI) image generation models are immense, involving very large training image datasets. However, all generative AI models have inherent biases and imbalances that are representative of and inherited from the image datasets they are trained on. Since the social impact of generative AI is potentially large, it is important to ensure that the generated data distribution doesn’t replicate or augment sensitive biases, by amplifying stereotypes (e.g., gender stereotypes, racial stereotypes, etc.). Studies have demonstrated that current image generation models do suffer from these biases (e.g., Leonardo Nicoletti and Dina Bass, “Bloomberg Analysis of Stable Diffusion,” https://www.bloomberg.com/graphics/2023-generative-ai-bias/, retrieved January 17, 2024). Image sets generated for every high-paying job were dominated by subjects with lighter skin tones, while subjects with darker skin tones were more commonly generated by prompts like “fast-food worker” and “social worker.” For each image depicting a perceived woman, almost three times as many images were generated of perceived men. Most occupations were dominated by men, except for low-paying jobs like housekeeper and cashier. Men with lighter skin tones represented the majority of subjects in every high-paying job, including “politician,” “lawyer," “judge” and “CEO.” The biases in image generation models are worse than reality, with women being underrepresented in high-paying occupations and overrepresented in low-paying ones, and overrepresenting people with darker skin tones while underrepresenting people with lighter skin tones in low-paying fields. As such, there is a need for optimizing the training of generative image models to counter sensitive biases. Some embodiments of the present disclosure provide a method for receiving a first set of training data including a group of training images and corresponding image captions, and using the first set of training data, performing a first training process to train the image generation model resulting in a trained image generation model. The method further includes defining a group of sensitive categories associated with the group of training images and further defining a group of protected attributes associated with the group of training images. The method further includes, using the group of sensitive categories and the group of protected attributes, generating a second set of training data that is balanced across at least one of the group of protected attributes with respect to at least one of the group of sensitive categories, and using the second set of training data, performing a second training process to de-bias the trained image generation model resulting in a de-biased image generation model. Some embodiments of the present disclosure provide a non-transitory computer-readable medium storing a program for training an image generation model, which when executed by a computer, configures the computer to receive a first set of training data including a group of training images and corresponding image captions, and using the first set of training data, perform a first training process to train the image generation model resulting in a trained image generation model. The program, when executed, further configures the computer to define a group of sensitive categories associated with the group of training images, define a group of protected attributes associated with the group of training images, and using the group of sensitive categories and the group of protected attributes, generate a second set of training data that is balanced across at least one of the group of protected attributes with respect to at least one of the group of sensitive categories. The program, when executed, further configures the computer to, using the second set of training data, perform a second training process to de-bias the trained image generation model resulting in a de-biased image generation model. Some embodiments of the present disclosure provide a system for training an image generation model, having a processor and a non-transitory computer-readable medium storing a set of instructions, which when executed by the processor, configure the system to receive a first set of training data including a group of training images and corresponding image captions, and using the first set of training data, perform a first training process to train the image generation model resulting in a trained image generation model. The instructions, when executed, further configure the computer to define a group of sensitive categories associated with the group of training images, define a group of protected attributes associated with the group of training images, and using the group of sensitive categories and the group o