Generating synthetic images as training dataset for a machine learning network

Patent metadata

Field	Value
Publication number	US-12073565-B2
Application number	US-202218145880-A
Country	US
Kind code	B2
Filing date	Dec 23, 2022
Priority date	Apr 26, 2018
Publication date	Aug 27, 2024
Grant date	Aug 27, 2024

Abstract

A method may include identifying a first image for training a deep learning network, wherein the first image includes at least one target object associated with at least one location in the first image, and wherein the first image is associated with a mask image; determining a set of deformations to create a training set of deformed images, wherein the training set is to be used to train the deep learning network; generating the training set of deformed images by applying the set of deformations to the first image; and generating a set of deformed mask images by applying the set of deformations to the mask image, wherein each deformed image of the training set of deformed images is associated with a respective mask image to identify the location of the at least one target object in each deformed image.

First claim

What is claimed is:

1. A method, comprising:
applying, by a device, a deformation to an image;
generating, by the device, a deformed image based on applying the deformation to the image;
applying, by the device, the deformation to a mask image that identifies a target location in the image;
generating, by the device, a deformed mask image based on applying the deformation to the mask image; and
training, by the device, a model using the deformed image and the deformed mask image.

2. The method of claim 1 , wherein the deformation is a first deformation and the deformed image is a first deformed image;
wherein the method further comprises:
applying a second deformation to the first deformed image; and
generating a second deformed image based on applying the second deformation to the first deformed image; and
wherein training the model comprises:
training the model using the second deformed image.

3. The method of claim 1 , wherein the deformation is a first deformation and the deformed image is a first deformed image;
wherein the method further comprises:
applying a second deformation to the image;
generating a second deformed image based on applying the second deformation to the image; and
refraining from applying the second deformation to the mask image; and
wherein training the model comprises:
training the model using the second deformed image.

4. The method of claim 1 , further comprising:
selecting the deformation based on a type of the image; and
wherein generating the deformed image comprises:
generating the deformed image based on selecting the deformation.

5. The method of claim 1 , further comprising:
generating the mask image based on the target location in the image.

6. The method of claim 1 , further comprising:
randomly or pseudorandomly selecting the deformation; and
wherein generating the deformed image comprises:
generating the deformed image based on randomly or pseudorandomly selecting the deformation.

7. The method of claim 1 , further comprising:
performing segmentation of images using the model.

8. A device, comprising:
one or more memories; and
one or more processors, coupled to the one or more memories, configured to:
apply a deformation to an image;
generate a deformed image based on an application of the deformation to the image;
apply the deformation to a mask image that identifies a target location in the image;
generate a deformed mask image based on an application of the deformation to the mask image; and
train a model using the deformed image and the deformed mask image.

9. The device of claim 8 , wherein the deformation is a first deformation and the deformed image is a first deformed image;
wherein the one or more processors are further configured to:
apply a second deformation to the first deformed image; and
generate a second deformed image based on applying an application of the second deformation to the first deformed image; and

wherein the one or more processors, to train the model, are configured to:
train the model using the second deformed image.

10. The device of claim 8 , wherein the deformation is a first deformation and the deformed image is a first deformed image;
wherein the one or more processors are further configured to:
apply a second deformation to the image;
generate a second deformed image based on an application of the second deformation to the image; and
refrain from applying the second deformation to the mask image; and

wherein the one or more processors, to train the model, are configured to:
train the model using the second deformed image.

11. The device of claim 8 , wherein the one or more processors are further configured to:
select the deformation based on a type of the image; and
wherein the one or more processors, to generate the deformed image, are configured to:
generate the deformed image based on a selection of the deformation.

12. The device of claim 8 , wherein the one or more processors are further configured to:
generate the mask image based on the target location in the image.

13. The device of claim 8 , wherein the one or more processors are further configured to:
randomly or pseudorandomly select the deformation; and
wherein the one or more processors, to generate the deformed image, are configured to:
generate the deformed image based on a random or pseudorandom selection of the deformation.

14. The device of claim 8 , wherein the one or more processors are further configured to:
perform segmentation of images using the model.

15. A non-transitory computer-readable medium storing a set of instructions, the set of instructions comprising:
one or more instructions that, when executed by one or more processors of a device, cause the device to:
apply a deformation to an image;
generate a deformed image based on an application of the deformation to the image;
apply the deformation to a mask image that identifies a target location in the image;
generate a deformed mask image based on an application of the deformation to the mask image; and
train a model using the deformed image and the deformed mask image.

16. The non-transitory computer-readable medium of claim 15 , wherein the deformation is a first deformation and the deformed image is a first deformed image;
wherein the one or more instructions further cause the device to:
apply a second deformation to the first deformed image; and
generate a second deformed image based on an application of the second deformation to the first deformed image; and

wherein the one or more instructions, that cause the device to train the model, cause the device to:
train the model using the second deformed image.

17. The non-transitory computer-readable medium of claim 15 , wherein the deformation is a first deformation and the deformed image is a first deformed image;
wherein the one or more instructions further cause the device to:
apply a second deformation to the image;
generate a second deformed image based on an application of the second deformation to the image; and
refrain from applying the second deformation to the mask image; and

wherein the one or more instructions, that cause the device to train the model, cause the device to:
train the model using the second deformed image.

18. The non-transitory computer-readable medium of claim 15 , wherein the one or more instructions further cause the device to:
select the deformation based on a type of the image; and
wherein the one or more instructions, that cause the device to generate the deformed image, cause the device to:
generate the deformed image based on a selection of the deformation.

19. The non-transitory computer-readable medium of claim 15 , wherein the one or more instructions further cause the device to:
generate the mask image based on the target location in the image.

20. The non-transitory computer-readable medium of claim 15 , wherein the one or more instructions further cause the device to:
perform segmentation of images using the model.

Assignees

• Capital One Services Llc

Inventors

• Farivar Reza
• Taylor Kenneth

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06T2207/30176

  Document · CPC title

• G06T7/194

  involving foreground-background segmentation · CPC title

• G06T7/11

  Region-based segmentation · CPC title