Moderated generative image model training
AssigneeShutterstock, Inc.
InventorsRaúl Gómez Bruballa, Alessandra Sala
Life of this patent
Estimating the term…
Abstract
A method for training a generative image model includes receiving training data having training images and image captions and annotating the training data resulting in annotated training data having the training images, the image captions, and moderation labels, such that each training image is associated in the annotated training data with a corresponding image caption and at least one corresponding moderation label. The method further includes performing a training process to train a generative image model using the training images, the image captions, and the moderation labels, resulting in a trained model that generates images conditioned to both the image captions and the moderation labels.
Claims20
A method for training a generative image model, comprising:
- receiving training data comprising a plurality of training images and a plurality of image captions;
- annotating the training data resulting in annotated training data comprising the plurality of training images, the plurality of image captions, and a plurality of moderation labels, wherein each training image is associated in the annotated training data with a corresponding image caption and at least one corresponding moderation label;
- and
- performing a training process to train a generative image model using the plurality of training images, the plurality of image captions, and the plurality of moderation labels, resulting in a trained model that generates images conditioned to both the plurality of image captions and the plurality of moderation labels,
- wherein each moderation label in the plurality of moderation labels is a word or a phrase.
A non-transitory computer-readable medium storing a program for training a generative image model, which when executed by a computer, configures the computer to:
- receive training data comprising a plurality of training images and a plurality of image captions;
- annotate the training data resulting in annotated training data comprising the plurality of training images, the plurality of image captions, and a plurality of moderation labels, wherein each training image is associated in the annotated training data with a corresponding image caption and at least one corresponding moderation label;
- and
- perform a training process to train a generative image model using the plurality of training images, the plurality of image captions, and the plurality of moderation labels, resulting in a trained model that generates images conditioned to both the plurality of image captions and the plurality of moderation labels,
- wherein each moderation label in the plurality of moderation labels is a word or a phrase.
A system for training a generative image model, comprising:
- a processor;
- and
- a non-transitory computer readable medium storing a set of instructions, which when executed by the processor, configure the processor to:
- receive training data comprising a plurality of training images and a plurality of image captions;
- annotate the training data resulting in annotated training data comprising the plurality of training images, the plurality of image captions, and a plurality of moderation labels, wherein each training image is associated in the annotated training data with a corresponding image caption and at least one corresponding moderation label;
- perform a training process to train a generative image model using the plurality of training images, the plurality of image captions, and the plurality of moderation labels, resulting in a trained model that generates images conditioned to both the plurality of image captions and the plurality of moderation labels;
- provide a first input to the trained model, the first input comprising a description of a desired image;
- provide a second input to the trained model, the second input comprising a moderation label;
- and
- receive as an output from the trained model in response to the first input and the second input, an output image that (1) comprises image content that matches at least part of the description of the desired image and (2) omits image content that matches the moderation label,
- wherein each moderation label in the plurality of moderation labels is a word or a phrase.
Dependencies are inferred from the claim text.