GAN IMAGE GENERATION FROM FEATURE REGULARIZATION

Min Jin Chong; Krishna Kumar Singh; Yijun Li; Jingwan Lu

Patent

Atlas literature

Patent

US 12,657,882 B2

GAN IMAGE GENERATION FROM FEATURE REGULARIZATION

Min Jin Chong, Krishna Kumar Singh, Yijun Li, Jingwan Lu

ADOBE INC., San Jose, CA (US)·Jun. 16, 2026·US

Drawings

Patent drawings and their descriptions. Click a drawing to enlarge it.

FIG. 1

FIG. 1 shows an example of an image generation system according to aspects of the present disclosure.

FIG. 2

FIG. 2 shows an example of an image generation appa- ratus according to aspects of the present disclosure.

FIG. 3

FIG. 3 shows an example of a generator network accord- ing to aspects of the present disclosure.

FIG. 4

FIG. 4 shows an example of a discriminator network according to aspects of the present disclosure.

FIG. 5

FIG. 5 shows an example of a discriminator network during training according to aspects of the present disclo- sure.

FIG. 6

FIG. 6 shows an example of a method for training a GAN with feature regularization according to aspects of the pres- ent disclosure.

FIG. 7

FIG. 7 shows an example of a pipeline for using a partially pretrained discriminator network with other GANs accord- ing to aspects of the present disclosure.

FIG. 8

FIG. 8 shows an example of a method for generating an image using a GAN according to aspects of the present disclosure.

FIG. 9

FIG. 9 shows an example of a method for providing an image to a user according to aspects of the present disclo- sure.

FIG. 10

FIG. 10 shows an example of a computing device accord- ing to aspects of the present disclosure.

Claims

Claims define the patent's legal scope. Independent claims stand alone; dependent claims (nested) narrow them. Click a claim to expand its dependents.

3 independent · 17 dependent

1
IndependentGenerative Adversarial Network (GAN) for image generation
A method comprising: obtaining an input vector; and generating an image based on the input vector using a generative adversarial network (GAN) that includes at least one encoder layer and at least one classiﬁer layer, wherein the at least one classiﬁer layer is trained by computing a regularization loss that includes a gradient with respect to encoded features, wherein the at least one encoder layer generates the encoded features and the at least one classiﬁer layer generates a classiﬁcation output based on the encoded features, and wherein the gradient has a same dimension as the encoded features.
1. 2
  Dependent← claim 1Generative Adversarial Network (GAN) for image generation
  The method of claim 1, wherein: the gradient is computed based on the classiﬁcation output independently of the at least one encoder layer.
2. 3
  Dependent← claim 1Generative Adversarial Network (GAN) for image generation
  The method of claim 1, wherein: the input comprises a text prompt.
3. 4
  Dependent← claim 1Generative Adversarial Network (GAN) for image generation
  The method of claim 1, wherein: the image comprises a face image, and a discriminator network of the GAN is trained to classify face images as real or synthetic.
5
IndependentGenerative Adversarial Network (GAN) for image generation
An apparatus comprising: at least one processor; at least one memory including instructions executable by the processor; and the apparatus further comprising a GAN comprising parameters stored in the at least one memory, wherein the GAN includes at least one encoder layer and at least one classiﬁer layer, wherein the at least one classiﬁer is trained to generate images by computing a regulariza-tion loss that incudes a gradient with respect to encoded features generated by the GAN, wherein the at least one encoder layer generates the encoded features and the at least one classiﬁer layer generates a classiﬁcation out-put based on the encoded features, and wherein the gradient has a same dimension as the encoded features.
1. 6
  Dependent← claim 5Generative Adversarial Network (GAN) for image generation
  The apparatus of claim 5, wherein: the GAN comprises a generator network conﬁgured to generate the images.
2. 7
  Dependent← claim 5Generative Adversarial Network (GAN) for image generation
  The apparatus of claim 5, further comprising: a training component conﬁgured to compute a regular-ization loss, wherein the training is based on the regularization loss.
3. 9
  Dependent← claim 5Generative Adversarial Network (GAN) for image generation
  The apparatus of claim 5, wherein: the GAN comprises a discriminator network conﬁgured to classify the images, wherein the training is based on the classifying of the images.
13
IndependentGenerative Adversarial Network (GAN) for image generation
A non-transitory computer readable medium storing code, the code comprising instructions that, when executed by at least one processor, cause the at least one processor to perform operations comprising: obtaining an input vector; and generating an image based on the input vector using a generative adversarial network (GAN) that includes at least one encoder layer and at least one classiﬁer layer, wherein the at least one classiﬁer layer is trained by computing a regularization loss that includes a gradient with respect to encoded features, wherein the at least one encoder layer generates the encoded features and the at least one classiﬁer layer generates a classiﬁcation output based on the encoded features, and wherein the gradient has a same dimension as the encoded features.
1. 14
  Dependent← claim 13Generative Adversarial Network (GAN) for image generation
  The non-transitory computer readable medium of claim 13, wherein: the gradient is computed based on the classiﬁcation output independently of the at least one encoder layer.
2. 15
  Dependent← claim 13Generative Adversarial Network (GAN) for image generation
  The non-transitory computer readable medium of claim 13, wherein: the input comprises a text prompt.
3. 16
  Dependent← claim 13Generative Adversarial Network (GAN) for image generation
  The non-transitory computer readable medium of claim 13, wherein: the image comprises a face image, and a discriminator network of the GAN is trained to classify face images as real or synthetic.
4. 18
  Dependent← claim 13Generative Adversarial Network (GAN) for image generation
  The non-transitory computer readable medium of claim 13, wherein: the discriminator network of the GAN comprises a plu-rality of classiﬁers.
5. 20
  Dependent← claim 13Generative Adversarial Network (GAN) for image generation
  The non-transitory computer readable medium of claim 13, the code further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising: computing a regularization loss, wherein the regulariza-tion loss comprises an R₁ regularization loss. ∗ ∗ ∗ ∗ ∗

Device structures

Layer stacks claimed or described, ordered top of device to substrate.

Generative Adversarial Network (GAN) for image generation

classifier_layerclassifier layer

encoder_layerencoder layer

GAN discriminator network with pretrained encoder and feature regularization

classifierclassifier

pretrained_encoderpretrained encoder

Cited prior art

Patents and literature cited by this patent (applicant and examiner references).

Cited patents · 31

US 10,636,141 B210,636,141 B2 * 4/2020 Zhou......................... G06T 7/11examiner
US 11,995,803 B111,995,803 B1 * 5/2024 Karpman.................. G06T 5/70examiner
US 12,045,315 B212,045,315 B2 * 7/2024 Takeda.................. G06T 3/4046examiner
US 2019/0122072 A12019/0122072 A1 * 4/2019 Cricrı`..................... G06N 3/045examiner

GAN IMAGE GENERATION FROM FEATURE REGULARIZATION | Matter42 Literature