DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
Information Disclosure Statement
The information disclosure statements (IDS) submitted on July, 7th 2025, February 5th, 2026, and April 2nd 2026 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Response to Arguments
This is in response to applicant’s amendment/response filed on 02/05/2026 which have been entered and made of record.
Applicant’s arguments with respect to claim(s) 1-20 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Regarding the remaining arguments applicant argues with respect to the amended claim language, which is fully addressed in the prior art rejections set forth below.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1-8, 12-13, and 15-18 are rejected under 35 U.S.C. 103 as being unpatentable over Dundar et al. U.S. Patent Application Publication 20190244060 A1 (hereinafter Dundar) in view of Elmoznino et al. U.S. Patent Application Publication 20210150684 A1 (hereinafter Elmoznino) in further view of Yang et al. Chinese Patent Application 114612964 A (hereinafter Yang) .
Regarding claim 1, Dundar teaches: a method comprising:
constructing a set of target domain images; (Reference Style Images or Photorealistic Style Images, Para. 0132)
training an image generation model on the set of target domain images and a set of source domain images;
generating a set of paired images (Stylized Synthetic Dataset, Para. 0136) using the trained image generation model (Style Transfer Neural Network Model 610, Para. 0136) and (Stylized Synthetic Image, Para. 0136) and a source domain image (Ground Truth associated with Synthetic Image, Para. 0136 or Content Image Para. 0036 and 0132), the target domain image(Stylized Synthetic Image, Para. 0136) being paired with the source domain image(Ground Truth associated with Synthetic Image, Para. 0136 or Content Image Para. 0036 and 0132);
evaluating a quality of the set of paired images;
iteratively generating the set of paired images (Stylizing of the Synthetic Images Repeated for a Number of Iterations, Para. 0133)
and generating output target domain images (Stylized Synthetic Image, Para. 0136 and Fig. 6A) using an image translation model (Style Transfer Neural Network Model 610, Para. 0136) trained on the adjusted set of paired images (Different Iteration of Stylizing of Synthetic Images).
However, Dundar fail to explicitly teach:
training an image generation model on the set of target domain images and a set of source domain images;
noise input;
evaluating a quality of the set of paired images;
Dundar and Elmoznino are analogous to the claimed invention because both of them are in the same field of utilizing neural networks to translate images in one domain to another domain.
Elmoznino teaches:
training an image generation model (Data Pairing Model or Run-Time Model, Para. 0009-0010) on the set of target domain images (Second Subset of Images in Second Doman Space, Para. 0009)and a set of source domain images(First Domain Space Images, Para. 0009);
generating a set of paired images (Synthetic Paired Dataset of Images from Unpaired Datasets, Para. 0010) using the trained image generation model (Data Pairing Model Para. 0009 or GAN Model) and a noise input (Random Noise Input into GAN models, Para. 0037 and 0085), the noise input used to generate(Second Domain Space) image (Subset of Unpaired Images in Second Domain, Para. 0009-00010) associated with a source domain (First Doman Space) image(Subset of Unpaired Images in First Domain, Para. 0009-00010) the target domain image(Subset of Unpaired Images in Second Domain, Para. 0009-00010) being paired(Synthetic Paired Dataset of Images from Unpaired Datasets, Para. 0010) with the source domain image(Subset of Unpaired Images in First Domain, Para. 0009-00010); While Elmoznino does not explicitly state noise input, standard GAN models utilize random noise as input to generate images.
evaluating a quality of the set of paired images; (Filter pair images based on a quality measure, Para. 0038-0039)
iteratively generating the set of paired images(Filtered/Pruned Pairs of Images, Para. 0038) using the noise input(Random Noise Input into GAN models, Para. 0037 and 0085) to construct an adjusted set of paired images (Filtered/Pruned Paired Images, Para. 0039) based on the quality of the set of paired images; (Para. 0039) Pairs of images are generated using a Data Pairing Model. The filtering/pruning of paired images can be applied any time before the run-time model is used. The filtering/pruning can be automatic or manual. The filtering/pruning is not limited to one time and as such could be performed multiple times.
and generating output target domain images (Translate Image in First Domain to Second Domain, Para. 0009) using an image translation model (Run-Time Model, Para. 0009) trained on the adjusted set of paired images (Pruned Paired Images, Para. 0039).
Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Dundar’s Iterative Neural Network to incorporate Elmoznino’s Quality Evaluation and Pruning of Image Pairs. Since doing so would provide the benefit of ensuring the quality of the generated pairs as they will be used to train another model.
However Dundar and Elmoznino fail to explicitly teach noise input;
Dundar, Elmoznino, and Yang are analogous to the claimed invention because all of them are in the same field of utilizing neural networks to translate images in one domain to another domain.
Yang teaches:
generating a set of paired images (Pair of Sampled Images, Page 11 Para. 11-14 to Page 12 Para. 1-4) using the trained image generation model (GAN, Page 12, Para. 4) and a noise input (Preset Noise or Gaussian Random Noise, Page 12 Para. 2), the noise input used to generate(First Sampled Image) associated with a source domain image (Second Sampled Image) the target domain image(First Sampled Image) being paired with the source domain image(Second Sampled Image);
Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have Modified Dundar’s Iterative Neural Network altered by Elmoznino’s Quality Evaluation and Pruning of Image Pairs to incorporate Yang’s Image Generation using Noise Input. Since doing so would provide the benefit of generating images while evaluating them to ensure high quality images are generated. GAN models utilize two neural networks a generator and a discriminator. The generator generates images and the discriminator evaluates them to ensure the generated images are of sufficient quality.
Regarding claim 3, Dundar teaches the method of claim 1, wherein generating the set of paired images further comprises:
generating a second set of target domain images (Stylized Synthetic Dataset, Para. 0136) using the trained image generation model (Style Transfer Neural Network Model 610, Para. 0136);
However, Dundar fails to teach:
and generating the set of paired images using a second image generation model trained on the second set of target domain images.
Elmoznino teaches:
and generating the set of paired images using a second image generation model (Data Pairing Model, Para. 0009) trained on the second set of target domain (Second Subset of Images or Unpaired Images, Para. 0010) images. The Second Subset of Images or Set of Unpaired Images are images obtained from a server or were previously stored in memory. (Para. 0058) Thus, these images could have been generated or collected using a different model.
Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Dundar’s Iterative Neural Network to incorporate Elmoznino’s Data Pairing Model. Since doing so would provide the benefit of creating several sets of paired images using a model that has been specifically trained on the target domain. Thus, increasing the performance of the pairing model.
Regarding claim 4, Dundar teaches the method of claim 1, wherein training the image generation model further comprises:
minimizing a loss function comprising a custom loss; (Para. 0037-0039)
and computing the custom loss based on at least a set of source domain image (Ground Truth associated with Synthetic Image, Para. 0136 or Content Image Para. 0036 and 0132) features and a set of target domain image (Reference Style Images or Photorealistic Style Images, Para. 0132) features. (Para. 0037-0039 and 0149)
Regarding claim 5, Dundar teaches the method of claim 4, wherein the computing of the custom loss further comprises computing a distance between the set of source domain image features (Ground Truth associated with Synthetic Image, Para. 0136 or Content Image Para. 0036 and 0132) and the set of target domain image (Reference Style Images or Photorealistic Style Images, Para. 0132) features. (Para. 0037-0039 and 0149)
Regarding claim 6, Dundar teaches the method of claim 1, wherein constructing an adjusted set of paired images further comprises:
accessing a first image (CG-Rendered Image or Real Image, Para. 0144) and a second image (Real Image or CG-Rendered Image, Para. 0144);
computing a mask corresponding to an image attribute (Segmentation Mask, Para. 0144), the image attribute determined to be present in the first image (CG-Rendered Image or Real Image, Para. 0144);
and computing a combined image (Stylized Content Image) using the first image (Real Image or CG-Rendered Image), the second image (Real Image or CG-Rendered Image), and the mask (Segmented Mask). (Para. 0142-0145)
Regarding claim 7, Dundar teaches the method of claim 6, wherein the first image is a first target domain image (CG-Rendered Image or Real Image, Para. 0144) and the second image is a second target domain image (Real Image or CG-Rendered Image, Para. 0144). One of ordinary skill in the art would recognize that the images used could be in the same domain, closely related domains, or completely different domains. As the purpose of the models in this invention are to translate the image domain from one domain to another. Thus, a user could choose two images of the first target domain to be the source domain and the second target domain to be a different style domain.
Regarding claim 8, Dundar teaches the method of claim 6, wherein the mask (Segmented Mask) is computed using an image segmentation model (Semantic Segmentation Neural Network or Recognition Neural Network, Para. 0133) or a facial landmarks extractor.
Regarding claim 12, Dundar teaches at least one processor (Para. 0033) and a memory (Para. 0073) storing instructions that, when executed by the at least one processor, configure the apparatus to perform the method of claim 1, therefore it is rejected under the same rationale as claim 1.
Regarding claim 13, has similar limitations as of claim 3, therefore it is rejected under the same rationale as claim 3.
Regarding claim 15, has similar limitations as of claim 4, therefore it is rejected under the same rationale as claim 4.
Regarding claim 16, has similar limitations as of claim 5, therefore it is rejected under the same rationale as claim 5.
Regarding claim 17, has similar limitations as of claim 6, therefore it is rejected under the same rationale as claim 6.
Regarding claim 18, Dundar teaches a non-transitory computer-readable storage medium (Para. 0073 or 0090), the computer-readable storage medium including instructions that when executed by at least one processor (Para. 0033), cause the at least one processor to perform the method of claim 1, therefore it is rejected under the same rationale as claim 1.
Claim(s) 9-11, and 19-22 are rejected under 35 U.S.C. 103 as being unpatentable over Yang et al. Chinese Patent Application 114612964 A (hereinafter Yang) in view of Karras et al. U.S. Patent 11605001 B2 (hereinafter Karras).
Regarding claim 9, Yang teaches a method comprising:
constructing an augmented set of target domain images (Preset Image, Page 4 Para. 8 or Sample Images Page 11, Para. 11-14 and Page 12 Para. 3-11, and Field Label Page 16 Para. 1) comprising condition labels (Field Labels, Page 16 Para. 1) associated with target domain images (Preset Image, Page 4 Para. 8 or Sample Images Page 11, Para. 11-14 and Page 12 Para. 3-11, and Field Label Page 16 Para. 1),the condition labels(Field Labels, Page 16 Para. 1) corresponding to one or more image characteristic levels;
training an image producing model (cGAN Page 3, Section: Specific Implementation Examples Para. 6, Image Conversion Network, Page 4 Para. 8, and Countermeasure Network, Page 12 Para. 3-4) on the augmented set of target domain images (Preset Image, Page 4 Para. 8), the image producing model being a conditional image producing model (cGAN Page 3, Section: Specific Implementation Examples Para. 6, Image Conversion Network, Page 4 Para. 8, and Countermeasure Network, Page 12 Para. 3-4); The Image Conversion Network Model is trained using two sample images in different domains, with corresponding labels associated to the domain field. Page 11, Para. 11-14 and Page 12 Para. 3-11) cGAN models and other neural network models are trained through large image datasets.
generating a first feature map (Feature Mapping Relationship and Style Features, Page 5 Para. 6-12) using the trained image producing model (cGAN Page 3, Section: Specific Implementation Examples Para. 6, Image Conversion Network, Page 4 Para. 8, and Countermeasure Network, Page 12 Para. 3-4) and a first input set (First Image, Page 4 Para. 6 and Field Label Page 16 Para. 1) including a source image (First Image, Page 4 Para. 6 or ) and a first condition label (Field Labels including High-level Content Feature for Image 41, Page 16 Para. 1) indicative of a first image characteristic (Field Labels including High-level Content Feature for Image 41, Page 16 Para. 1)level, the first feature map (Feature Mapping Relationship and Style Features, Page 5 Para. 6-12) associated with a layer of the trained image producing model (cGAN Page 3, Section: Specific Implementation Examples Para. 6, Image Conversion Network, Page 4 Para. 8, and Countermeasure Network, Page 12 Para. 3-4); (Page 5 Para. 6-12)
generate a second feature map (Feature Mapping Relationship and Style Features, Page 5 Para. 6-12) using the trained image producing model (cGAN Page 3, Section: Specific Implementation Examples Para. 6, Image Conversion Network, Page 4 Para. 8, and Countermeasure Network, Page 12 Para. 3-4) and a second input set (Preset Image, Page 4 Para. 8 and Field Label Page 16 Para. 1) including the source image(Preset Image, Page 4 Para. 8) and a second condition label (Field Labels including High-level Content Feature for Image 41, Page 16 Para. 1) indicative of a second image characteristic(Field Labels including High-level Content Feature for Image 42, Page 16 Para. 1) level, the second feature map(Feature Mapping Relationship and Style Features, Page 5 Para. 6-12) associated with the layer of the trained image producing model (cGAN Page 3, Section: Specific Implementation Examples Para. 6, Image Conversion Network, Page 4 Para. 8, and Countermeasure Network, Page 12 Para. 3-4); (Page 5 Para. 6-12)
computing a combined feature map (Feature Group, Page 5 Lines 9) using the first feature map (Feature Mapping Relationship and Style Features, Page 5 Para. 6-12), the second map (Feature Mapping Relationship and Style Features, Page 5 Para. 6-12), and a mask (Page 10 Para. 5);
and using the combined feature map (Feature Group, Page 5 Lines 9) in generating output target images (Second Image, Page 4 Para. 8) using the trained image producing model (cGAN Page 3, Section: Specific Implementation Examples Para. 6, Image Conversion Network, Page 4 Para. 8, and Countermeasure Network, Page 12 Para. 3-4).
Although Yang fails to explicitly disclose training on a set of target domain images. It would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Yang’s Preset image used to train a cGAN to be a set of target domain images, since it is known in the art to use sets/multiple images to train GAN models. Doing so would provide the benefit of training Yang’s model in a robust way increasing the accuracy of the model, as training on multiple images enhances the quality of model.
However Yang fails to explicitly teach:
training an image producing model on the augmented set of target domain images;
the condition labels corresponding to one or more image characteristic levels;
a first conditional label indicative of a first image characteristic level;
a second conditional label indicative of a second image characteristic level;
Yang and Karras are analogous to the claimed invention because both of them are in the same field of utilizing a style-based network to generate images and feature mapping image attributes.
Karras teaches:
training an image producing model(Style-Based Generator System, Col 4 Lines 8-24 and Col 6 Lines 3-25) on the augmented set of target domain images (Generated Images with Two or one latent code, Col. 7 Lines 1-18);
the condition labels(Classification Label, Col. 4 Lines 40-51 or Style Signals Col.12 Lines 27-33) corresponding to one or more image characteristic levels(Style Signal is the Strength of Image Attributes, Col. 12 Lines 27-33);
a first conditional label (Classification Label, Col. 4 Lines 40-51 or Style Signals Col.12 Lines 27-33) indicative of a first image characteristic level (First Style Signal, Col. 41 Lines 57-67 to Col. 42 Lines 1-6); At each convolution layer the style is adjusted according to the image attributes detailed in the style signal which can be either from the source or destination data, Col. 6 Lines 3-35.
a second conditional label(Classification Label, Col. 4 Lines 40-51 or Style Signals Col.12 Lines 27-33) indicative of a second image characteristic level(Second or Third Style Signal, Col. 42 Lines 7-37);
Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Yang’s cGAN/Image Conversion Network to incorporate Karras’s Style-Based Generator System that Modifies the Style Based on Attributes of the Images. Since doing so would provide the benefit of increasing the flexibility and control of the generated images, by adjusting/controlling the strength of the image attributes at each layer in the system (Karras et al. Col. 6 Lines 3-25).
Regarding claim 10, Yang teaches the method of claim 9, further comprising:
generating a first output image (Second Image, Page 4 Para. 8) by running the trained image producing model (cGAN Page 3, Section: Specific Implementation Examples Para. 6, Image Conversion Network, Page 4 Para. 8, and Countermeasure Network, Page 12 Para. 3-4) using an input set (First Image, Page 4 Para. 6 and Preset Image, Page 4 Para. 8) including the first condition label (Field Labels, Page 16 Para. 1);
generating a second output image (Different Second Image using Different Input Set) by running the trained image producing model (cGAN Page 3, Section: Specific Implementation Examples Para. 6, Image Conversion Network, Page 4 Para. 8, and Countermeasure Network, Page 12 Para. 3-4) using an additional input set using the second condition label (Field Labels, Page 16 Para. 1);
computing the mask (Page 10 Para. 5 based on the first output image (Second Image, Page 4 Para. 8) and the second output image (Different Second Image using Different Input Set). The cGAN or Image Conversion Network can be used multiple times with various image input sets in different domains to obtain desired results. One of ordinary skill could generate two output images and reuse them as input into the model to compute a mask of the two images. As well as use the mask to generate a third image which is a combination of both previous images.
However, Yang fails to explicitly teach:
generating a second output image by running the trained image producing model using an additional input set using the second condition label;
Karras teaches:
generating a second output image (Col. 5 Lines 50-61) by running the trained image producing model (Style-Based Generator System, Col 4 Lines 8-24 and Col 6 Lines 3-25) using an additional input set (Additional Data, Col.4 Lines 40-51) using the second condition label (Classification Label, Col. 4 Lines 40-51 or Style Signals Col.12 Lines 27-33);
Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Yang’s cGAN/Image Conversion Network to incorporate Karras’s Style-Based Generator System that can receive additional input. Since doing so would provide the benefit of enhancing the models to accept additional input that can tailor the image generated to specific requirements. Which increases the flexibility and images the model can generate.
Regarding claim 11, Yang teaches the method of claim 9, wherein the trained image producing model (cGAN Page 3, Section: Specific Implementation Examples Para. 6, Image Conversion Network, Page 4 Para. 8, and Countermeasure Network, Page 12 Para. 3-4) is a conditional image translation model or a conditional image generation label (cGAN Page 3, Section: Specific Implementation Examples Para. 6, Image Conversion Network, Page 4 Para. 8, and Countermeasure Network, Page 12 Para. 3-4).
Regarding claim 19, Yang teaches at least one processor (Page 22, Para. 10) and a memory (Page 22, Para. 9) storing instructions that, when executed by the at least one processor, configure the apparatus to perform the method of claim 9, therefore it is rejected under the same rationale as claim 9.
Regarding claim 20, Yang teaches a non-transitory computer-readable storage medium (Page 22, Para. 9), the computer-readable storage medium including instructions that when executed by a computer, cause the computer to perform the method of claim 9, therefore it is rejected under the same rationale as claim 9.
Regarding claim 21, Yang fails to teach the method of claim 9, further comprising:
determining the one or more image characteristic levels based on a level of stylized strength.
However, Karras teaches the method of claim 9, further comprising:
determining the one or more image characteristic levels based on a level of stylized strength(Style Signal is the Strength of Image Attributes, Col. 12 Lines 27-33).
Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Yang’s cGAN/Image Conversion Network to incorporate Karras’s Style-Based Generator System that Modifies the Style Based on Attributes of the Images. Since doing so would provide the benefit of increasing the flexibility and control of the generated images, by adjusting/controlling the strength of the image attributes at each layer in the system (Karras et al. Col. 6 Lines 3-25).
Regarding claim 22, Yang fails to teach the method of claim 9, wherein the first image characteristic level corresponds to a high identity preservation and the second image characteristic level a low identity preservation.
However, Karras teaches the method of claim 9, wherein the first image characteristic level(First Style Signal, Col. 41 Lines 57-67 to Col. 42 Lines 1-6) corresponds to a high identity preservation (Preserve the Style of the Layer) and the second image characteristic level(Second or Third Style Signal, Col. 42 Lines 7-37) a low identity preservation(Preserve the Style of the Layer). The style at each layer can be at different scales and strengths. As well as the style can be either from the source or destination data Col. 6 Lines 3-56.
Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Yang’s cGAN/Image Conversion Network to incorporate Karras’s Style-Based Generator System that Modifies the Style Based on Attributes of the Images. Since doing so would provide the benefit of increasing the flexibility and control of the generated images, by adjusting/controlling the strength of the image attributes at each layer in the system (Karras et al. Col. 6 Lines 3-25).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BRIANNA R COCHRAN whose telephone number is (571)272-4671. The examiner can normally be reached Mon-Fri. 7:30am - 5:00pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Alicia Harrington can be reached at (571) 272-2330. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/BRIANNA RENAE COCHRAN/Examiner, Art Unit 2615
/DANIEL F HAJNIK/Supervisory Patent Examiner, Art Unit 2616
Prosecution Insights