Using neural networks in laser finishing of apparel

Patent metadata

Field	Value
Publication number	US-11952693-B2
Application number	US-201816177415-A
Country	US
Kind code	B2
Filing date	Oct 31, 2018
Priority date	Oct 31, 2017
Publication date	Apr 9, 2024
Grant date	Apr 9, 2024

Abstract

Software and lasers are used in finishing apparel to produce a desired wear pattern or other design. A technique includes using machine learning to create or extract a laser input file for wear pattern from an existing garment. Machine learning can be by a generative adversarial network, having generative and discriminative neural nets. The generative adversarial network is trained and then used to create a model. This model is used generate the laser input file from an image of the existing garment with the finishing pattern. With this laser input file, a laser can re-create the wear pattern from the existing garment onto a new garment.

First claim

The invention claimed is:

1. A method comprising:
providing an assembled garment made from fabric panels of a woven first material comprising a warp comprising indigo ring-dyed cotton yarn, wherein the fabric panels are sewn together using thread;
providing a laser input file that is representative of a finishing pattern from an existing garment made from a second material, wherein the finishing pattern on the existing garment was not created by a laser, and the laser input file was obtained by
providing sample laser input files and images of sample garments with lased finishing patterns resulting from the sample laser input files to a generative adversarial network, wherein the sample laser input files comprise real laser input files for the generative adversarial network,
using a generative neural net of the generative adversarial network, generating fake laser input files for the images of sample garments with lased finishing patterns,
determining a generator loss based on the fake laser input files and the real laser input files,
inputting the real laser input files to a real discriminator of the generative adversarial network,
inputting the real laser input files and the fake laser input files to a fake discriminator of the generative adversarial network,
determining a discriminator loss based on outputs of the real discriminator and fake discriminator, and
based on outputs of the generator loss and discriminator loss, iteratively training a model to obtain final model, wherein the final model generates the laser input file for an image of the existing garment with the finishing pattern; and
using a laser to create a lased finishing pattern on an outer surface of the assembled garment based on the laser input file, wherein based on the laser input file, the laser removes selected amounts of material from the surface of the first material at different pixel locations of the assembled garment,
for lighter pixel locations of the lased finishing pattern, a greater amount of the indigo ring-dyed cotton yarn is removed, while for darker pixel locations of the lased finishing pattern, a lesser amount of the indigo ring-dyed cotton yarn is removed, and
the lased finishing pattern created is able to extend across portions of the assembled garment where two or more fabric panels are joined together by threads at least by exposing these portions to the laser.

2. The method of claim 1 wherein the first material comprises a denim and the second material comprises a denim.

3. The method of claim 1 wherein the garment comprises at least one of jeans, shirts, shorts, jackets, vests, or skirts.

4. The method of claim 1 wherein the garment and the existing garment are of the same type of garment.

5. The method of claim 1 wherein the first material comprises a denim and the second material comprises a denim, the garment and the existing garment comprise a jean, and the lased finishing pattern created by the laser on the garment includes wear patterns comprising at least one of combs or honeycombs, whiskers, stacks, or train tracks, or a combination.

6. The method of claim 1 wherein for the portions of the jean exposed to the laser where the fabric panels are joined, the fabric panels are joined together using a thread comprising cotton.

7. The method of claim 1 wherein when using the laser to create the lased finishing pattern, different laser levels are obtained by varying an output of the laser beam by altering a characteristic of the laser comprising at least one of a frequency, period, pulse width, power, duty cycle, or burning speed.

8. The method of claim 1 wherein the first material comprises a different fabric characteristic from the second material.

9. The method of claim 1 wherein the first material comprises a first surface texture characteristic which is different from a second surface texture characteristic of the second material.

10. The method of claim 1 wherein the first material comprises a first dye characteristic which is different from a second dye characteristic of the second material.

11. The method of claim 1 wherein the first material comprises a first base fabric color characteristic which is different from a second base fabric color characteristic of the second material.

12. The method of claim 1 wherein the first material comprises a first yarn characteristic which is different from a second yarn characteristic of the second material.

13. The method of claim 1 wherein the first material comprises a first yarn weight characteristic which is different from a second yarn weight characteristic of the second material.

14. The method of claim 1 wherein the first material comprises a first yarn diameter characteristic which is different from a second yarn diameter characteristic of the second material.

15. The method of claim 1 wherein the first material comprises a first yarn twist characteristic which is different from a second yarn twist characteristic of the second material.

16. The method of claim 1 wherein the discriminator loss comprises at least a real discriminator loss and a fake discriminator loss, and the discriminator loss is determined based at least in part upon an output of the real discriminator that receives real data and an output of the fake discriminator that receives fake data, and the real discriminator and the fake discriminator are trained using a loss function.

17. The method of claim 1 wherein the generator loss of the generative adversarial network comprises at least a distance loss and a generative adversarial network loss, the generator loss is determined based at least in part upon a first input of a fake laser input file, a second input of a real laser input file, a first output of the generative adversarial network, and data pertaining to the fake discriminator.

18. The method of claim 1 wherein the fake laser input files are used in determining both the generator loss and discriminator loss.

19. A method comprising:
providing an assembled garment made from fabric panels of a first material comprising a warp comprising indigo ring-dyed cotton yarn, wherein the fabric panels are sewn together using threads;
providing a laser input file that is representative of an existing finishing pattern from an existing garment made from a second material having a second fabric characteristic that is different from a first fabric characteristic of the first material, wherein the existing finishing pattern on the existing garment was not created by a laser, and the laser input file was obtained at least by a trained generative adversarial network, and a generative adversarial network is trained into the trained generative adversarial network at least by:
providing real laser input files and images of sample garments that have lased finishing patterns resulting from the real laser input files as a training dataset to the generative adversarial network, wherein at least one image of the images of the sample garments captures at least a lased finishing pattern of the lased finishing patterns on a sample garment of the sample garments, and a real laser input file of the real laser input files is used to control the laser that lases the lased finishing pattern onto the sample garment,
using a generative neural network of the generative adversarial network, generating fake laser input files for the images of the sample garments with the lased finishing patterns resulting from the real laser input files;
determining a generator loss based at least in part upon the fake laser input files and the rea

Assignees

• Strauss Levi & Co

Inventors

• Schultz Jennifer
• Bell Benjamin
• Mahanty Debdulal
• Schultz Christopher

Classifications

• G06N3/0475

  Generative networks · CPC title

• G06N3/094

  Adversarial learning · CPC title

• D06B11/0096 Primary

  to get a faded look · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• A41B1/08

  Details · CPC title