Granted Patent B1

US 12,499,521 · App. 18/296,251 · Granted Dec 16, 2025

Inventors: Pratik Mohan Ramdasi (Sunnyvale, CA); Christopher Brossman (San Francisco, CA); Sowmya Tatavarty (Redwood City, CA); Agustin Mautone (Montevideo, UY); Fabrizio Albertoni (Montevideo, UY)

Assignee: The RealReal, Inc.

G06T5/77G06T5/50G06T2207/20084G06T2207/20221

Abstract

Claims (31)

1 . A method comprising:

receiving an image of an item and a prop, wherein the prop supports the item, wherein the prop includes a joint, and wherein the joint includes a seam that is visible in the image; and

modifying the image such that the seam of the joint is not visible in the image without altering portions of the image showing the item, wherein the prop is a mannequin, wherein the joint is a shoulder joint, and wherein the modifying of shoulder joint portions of the image comprises:

using a trained convolutional neural network to identify the shoulder joint portions in the image;

using an image segmentation model to identify the item in the image;

removing the joint portions using a generative adversarial network, wherein the generative adversarial network (GAN) generates a second image with the shoulder joint portions modified; and

combining the second image with the image.

2 . The method of claim 1 , further comprising:

modifying the image to remove a background.

3 . The method of claim 1 , further comprising:

modifying the image to align the item.

4 . The method of claim 1 , further comprising:

removing a base from the image, wherein the base is attached to the prop that supports the item.

5 . The method of claim 1 , further comprising:

cropping the image from a bottom of the image upwards, wherein the cropping is performed using a trained convolutional neural network.

6 . The method of claim 1 , wherein the removing of the background comprises:

detecting item pixels corresponding to the item and detecting prop pixels corresponding to the prop; and

replacing pixels other than the item pixels or the prop pixels in the image with white pixels.

7 . A system comprising:

a memory; and

a controller configured to receive an image of an item and a prop, wherein the prop supports the item, wherein the prop includes a joint, wherein the joint includes a seam that is visible in the image, wherein the controller is configured to modify the image such that the seam of the joint is not visible in the image without altering portions of the image showing the item, wherein the prop is a mannequin, wherein the joint is a shoulder joint, and wherein the controller modifies shoulder joint portions of the image by:

using a trained convolutional neural network to identify the shoulder joint portions in the image;

using an image segmentation model to identify the item in the image;

removing the joint portions using a generative adversarial network that generates a second image with the shoulder joint portions modified; and

combining the second image with the image.

8 . The system of claim 7 , wherein the controller modifies the image to remove a background.

9 . The system of claim 7 , wherein the controller modifies the image to align the item.

10 . The system of claim 7 , wherein the controller removes a base from the image, and wherein the base is attached to the prop that supports the item.

11 . The system of claim 9 , wherein the controller modifies the image to align the item using a trained convolutional neural network.

12 . The system of claim 7 , wherein the controller crops the image from a bottom of the image upwards, and wherein the cropping is performed using a trained convolutional neural network.

13 . The system of claim 7 , wherein the controller removes the background by detecting item pixels corresponding to the item and prop pixels corresponding to the prop, and by replacing pixels other than the item pixels or the prop pixels in the image with white pixels.

Assignments (4)

References Cited (19)