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 .
Drawings
The drawings were received on 10/31/2024. These drawings are accepted.
Claim Rejections - 35 USC § 103
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 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.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1, 3, 7-9, 11, 15-17, and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Bowman et al. (US20260057647, hereinafter “Bowman”) and in view of Preechakul et al (NPL “Diffusion Autoencoders: Toward a Meaningful and Decodable Representation”, hereinafter “Preechakul”)
(Claims 1, 3, 7-8 are listed after Claims 9-16)
Claim 9. Bowman teaches A system comprising:
one or more processors;
one or more computer memory in data communication with the one or more processors, the one or more computer memory having computer readable data stored thereon, the computer readable data including instruction that, when executed by one or more processors, causes the one or more processors to perform a method ([0031] “a client device can include at least one processor 208 (e.g., a CPU or GPU) and a memory 210 to execute application 207 and/or perform tasks on behalf of application 207.”) of a machine learning system ([0046] “train the models” when a model is trained it is understood to be a machine learning system) that includes an image encoder, ([0046] “The training manager 308 may train a model that has an encoder-decoder structure… training manager 308 may train the models using the synthetic images” When an encoder-decoder structure is trained by images, it is understood by the examiner to be an image encoder) a text encoder, ([0046] “The training manager 308 may train a model that has an encoder-decoder structure…The training manager 308 may train the models using the synthetic images and their corresponding defect labels or annotations.” When an encoder-decoder structure is trained by images and their corresponding defect labels/annotations, it is understood by the examiner to be a text encoder) and a diffusion model, ([0043] “generative model” and [0050] “the generative model may be implemented as… Score- or Probability-based Generative Models like diffusion models,”) the method including
receiving a training dataset with data pairs, the data pairs include at least a first data pair that has at least (i) image data that displays an anomaly ([0029] “images that exhibit specific appearances of defects from varying perspectives and under different environmental conditions. These generated images come with annotations and can be utilized in the training process of an AI model, which significantly enhances the efficiency and effectiveness of training AI systems to detect and distinguish various defect types and conditions.”) and (ii) text data describing the corresponding image data including the anomaly; ([0023] “annotated images, each clearly illustrating the specific defect including annotations for scratches or reflections.”)
Bowman does not explicitly teach generating, via the image encoder, image embeddings using pixels of the image data;
generating, via the text encoder, text embeddings using the text data;
generating semantic subcode using the image embeddings and the text embeddings;
generating, via the diffusion model, stochastic subcode using the pixels of the image data;
generating, via the diffusion model, reconstructed image data using the stochastic subcode and the semantic subcode; and
optimizing a loss based at least on an expected value of a difference between a predicted noise of a noisy image at a particular time and an actual noise of the noisy image at the particular time during the generation of the reconstructed image data; and
updating parameters of the diffusion model using the loss.
Preechakul teaches generating, via the image encoder, image embeddings using pixels of the image data; (pg3PDF “The goal of the semantic encoder Enc(x0) is to summarize an input image into a descriptive vector zsem =Enc(x0) with necessary information to help the decoder pθ(xt−1|xt,zsem) denoise and predict the output image.” a descriptive vector is understood to be the same as the claimed embeddings which includes image and text as defined in the following website: https://www.ibm.com/think/topics/vector-embedding)
generating, via the text encoder, text embeddings using the text data; (pg5PDF “We trained linear classifiers using images and attribute labels” The text embeddings and the image embeddings are generated by the same semantic encoder cited above in pg3PDF, it is a descriptive vector (same as the instant claimed embedding) used to guide the autoencoder to pg3PDF “predict the output image.” )
generating semantic subcode using the image embeddings and the text embeddings; (pg3PDF “semantic encoder zsem = Encφ(x0) that learns to map an
input image x0 to a semantically meaningful zsem. Here, the conditional DDIM decoder takes as input a latent variable z = (zsem,xT), which consists of the high-level “se
mantic” subcode zsem”)
generating, via the diffusion model, stochastic subcode using the pixels of the image data; (pg3PDF “semantic encoder zsem = Encφ(x0) that learns to map an
input image x0 to a semantically meaningful zsem. Here, the conditional DDIM decoder takes as input a latent variable z = (zsem,xT), which consists of… a low-level “stochastic” subcode xT”)
generating, via the diffusion model, reconstructed image data using the stochastic subcode and the semantic subcode; (Fig. 2 shows the output image is a reconstructed image using the stochastic and semantic subcode.)
PNG
media_image1.png
687
613
media_image1.png
Greyscale
And optimizing a loss (pg2PDF “The model is trained with a loss function”) based at least on an expected value of a difference between a predicted noise of a noisy image at a particular time (pg2PDF “a function ϵ θ(xt, t) that takes a noisy image xt and predicts its noise”) and an actual noise of the noisy image at the particular time (pg2PDF “
PNG
media_image2.png
39
198
media_image2.png
Greyscale
where ϵ is the actual noise added to x0 to produce xt…. we define a Gaussian diffusion process at time t (out of T)” the subtraction of the epsilon is the difference between the predicted noise and the actual noise) during the generation of the reconstructed image data; (this occurs to generate the output reconstructed image data as previously shown in figure 2) and
updating parameters of the diffusion model using the loss. (pg2PDF “Diffusion-based (DPMs) …model is trained with a loss function” when a model is trained with a loss function it is understood to be the same as the instant claimed updating parameters of the diffusion model using the loss )
It would have been obvious to persons of ordinary skill in the art before the effective filing date of the claimed invention to modify Bowman to have generating image and text embeddings using an encoder and generating semantic and stochastic subcode based on that to generate a reconstructed image, optimizing a loss based on a difference between predicted and actual noise of a noisy image and updating parameters of the diffusion model using the loss as taught by Preechakul to arrive at the claimed invention discussed above. The motivation for the proposed modification would have been to (Preechakul et al pg1PDF “encode any image into a two-part latent code that captures both semantics and stochastic variations and allows near-exact reconstruction.”)
Claim 11. Bowman and Preechakul The system of claim 9, wherein:
Bowman teaches the image data displays an object; and
the anomaly is a defect on the object. ([0003] “FIGS. 1A-1B illustrate synthetic images of an object with one or more defects”)
Claim 15. Bowman and Preechakul The system of claim 9, wherein the method further comprises:
Bowman teaches receiving a source image with source image data that is non-anomalous; ([0026] “The image data for the vehicle may be provided using a captured image of a physical vehicle, or a synthetic image generated to provide a realistic representation of the vehicle. In some embodiments, a generative model might be trained to take a type of object as input” is understood to be the same as an image without defect which is the synonymous with the instant claimed non-anomalous)
receiving text input that describes (i) a desired anomaly to be generated with respect to the source image and (ii) at least one attribute of the anomaly; ([0022] “a user (or control application, etc.) can specify one or more types of defects to be represented as being present in an object in an image to be generated, and can specify aspects such as an extent, size, shape, or other aspect of the defect.”) and
generating, via the machine learning system, a synthetic image using the source image and the text input, wherein the synthetic image displays the source image data with the desired anomaly as described by the text input. ([0026] “a user may have specified a type of defect, here a scratch 104 in the paint, and the generative model generated the image 100 of the vehicle 102 having a realistic looking scratch. The user may have specified other aspects as well, such as a number of defects to include, a size or extent of each defect, a location of the defect, and so on.”)
Claim 16. Bowman and Preechakul The system of claim 15, wherein the method further comprises:
Bowman teaches creating a new dataset that include at least the source image ([0026] “images may be used as training data for a defect detection model…. The image data for the vehicle may be provided using a captured image of a physical vehicle,”) and the synthetic image; ([0028] “generate another image of the vehicle having different defects and/or under different lighting conditions. FIG. 1B illustrates a synthetic image 110 with generated realistic defects produced by a synthetic defect generation system.”) and
training an anomaly detector using the new dataset, the anomaly detector including at least one machine learning model. ([0028] “enables a model to be trained on specific defects, as well as combinations of defects of similar or different types.” And [0050] “The generative model may be a machine learning model… training data to train a defect detection model.”)
Claim 1. The method herein has been executed and performed by the system of claim 9 and is likewise rejected
Claim 3. The method herein has been executed and performed by the system of claim 11 and is likewise rejected
Claim 7. The method herein has been executed and performed by the system of claim 15 and is likewise rejected
Claim 8. The method herein has been executed and performed by the system of claim 16 and is likewise rejected
Claim 17. Bowman teaches A computer implemented method of generating a dataset for training a machine learning model, the method comprises:
receiving a source image with source image data that is non-anomalous; ([0026] “The image data for the vehicle may be provided using a captured image of a physical vehicle, or a synthetic image generated to provide a realistic representation of the vehicle. In some embodiments, a generative model might be trained to take a type of object as input” is understood to be the same as an image without defect which is the synonymous with the instant claimed non-anomalous)
receiving text input that describes (i) an anomaly to be generated with respect to the source image data and (ii) one or more attributes of the anomaly; ([0022] “a user (or control application, etc.) can specify one or more types of defects to be represented as being present in an object in an image to be generated, and can specify aspects such as an extent, size, shape, or other aspect of the defect.”)
the synthetic image displaying the source image data with the anomaly as described by the text input, ([0026] “a user may have specified a type of defect, here a scratch 104 in the paint, and the generative model generated the image 100 of the vehicle 102 having a realistic looking scratch. The user may have specified other aspects as well, such as a number of defects to include, a size or extent of each defect, a location of the defect, and so on.”)
wherein, the dataset includes at least the source image ([0026] “images may be used as training data for a defect detection model…. The image data for the vehicle may be provided using a captured image of a physical vehicle,”) and the synthetic image, ([0028] “generate another image of the vehicle having different defects and/or under different lighting conditions. FIG. 1B illustrates a synthetic image 110 with generated realistic defects produced by a synthetic defect generation system.”)and the dataset is configured to train the machine learning model to perform an anomaly detection task. ([0028] “enables a model to be trained on specific defects, as well as combinations of defects of similar or different types.” And [0050] “The generative model may be a machine learning model… training data to train a defect detection model.”)
Bowman does not explicitly teach generating, via an image encoder, source image embeddings using pixels of the source image;
generating, via a text encoder, text input embeddings using the text input;
generating a semantic subcode using the source image embeddings and the text input embeddings;
generating, via a diffusion model, a stochastic subcode using the pixels of the source image; and
generating, via the diffusion model, a synthetic image using the stochastic subcode and the semantic subcode,
Preechakul teaches generating, via an image encoder, source image embeddings using pixels of the source image; (pg3PDF “The goal of the semantic encoder Enc(x0) is to summarize an input image into a descriptive vector zsem =Enc(x0) with necessary information to help the decoder pθ(xt−1|xt,zsem) denoise and predict the output image.” a descriptive vector is understood to be the same as the claimed embeddings which includes image and text: https://www.ibm.com/think/topics/vector-embedding)
generating, via a text encoder, text input embeddings using the text input; (pg5PDF “We trained linear classifiers using images and attribute labels” The text embeddings and the image embeddings are generated by the same semantic encoder cited above in pg3PDF, it is a descriptive vector (instant claimed embedding) used to guide the autoencoder to pg3PDF “predict the output image.” )
generating a semantic subcode using the source image embeddings and the text input embeddings; (pg3PDF “semantic encoder zsem = Encφ(x0) that learns to map an input image x0 to a semantically meaningful zsem. Here, the conditional DDIM decoder takes as input a latent variable z = (zsem,xT), which consists of the high-level “semantic” subcode zsem”)
generating, via a diffusion model, a stochastic subcode using the pixels of the source image; (pg3PDF “semantic encoder zsem = Encφ(x0) that learns to map an
input image x0 to a semantically meaningful zsem. Here, the conditional DDIM decoder takes as input a latent variable z = (zsem,xT), which consists of… a low-level “stochastic” subcode xT”)
and generating, via the diffusion model, a synthetic image using the stochastic subcode and the semantic subcode, (Fig. 2 shows the output image is a reconstructed image using the stochastic and semantic subcode.)
PNG
media_image1.png
687
613
media_image1.png
Greyscale
It would have been obvious to persons of ordinary skill in the art before the effective filing date of the claimed invention to modify Bowman to have generating image and text embeddings using an encoder and generating semantic and stochastic subcode based on that to generate a reconstructed image, optimizing a loss based on a difference between predicted and actual noise of a noisy image and updating parameters of the diffusion model using the loss as taught by Preechakul to arrive at the claimed invention discussed above. The motivation for the proposed modification would have been to (Preechakul et al pg1PDF “encode any image into a two-part latent code that captures both semantics and stochastic variations and allows near-exact reconstruction.”)
Claim 19. The method herein has been executed and performed by the system of claim 11 and is likewise rejected
Claim 20. Bowman and Preechakul The computer-implemented method of claim 19,
Bowman teaches wherein the one or more attributes of the anomaly include (i) a size of the defect and (ii) a location of the defect. ([0026] “a user may have specified a type of defect, here a scratch 104 in the paint, and the generative model generated the image 100 of the vehicle 102 having a realistic looking scratch. The user may have specified other aspects as well, such as a number of defects to include, a size or extent of each defect, a location of the defect, and so on.”)
Allowable Subject Matter
Claims 2, 4-6, 10, 12-14 and 18 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Bowman et al US20260057647 teaches a diffusion autoencoder which is trained on annotated images containing defects and source images without defects and produces reconstructed images based on the diffusion model but does not render obvious the claimed combination as a whole
Preechakul et al NPL “Diffusion Autoencoders: Toward a Meaningful and Decodable Representation” teaches a diffusion autoencoder which generates semantic and stochastic subcode based on the annotated training images and generates a reconstructed image by optimizing a loss based on the difference between predicted and actual noise but does not render obvious the claimed combination as a whole
Yu et al NPL “PromptFix: You Prompt and We Fix the Photo” teaches a VLM trained on source images and degraded images but does not render obvious the claimed combination as a whole
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure:
Pan et al US20250166266 teaches synthetic defect image generation utilizing a diffusion model
Any inquiry concerning this communication or earlier communications from the examiner should be directed to OWAIS MEMON whose telephone number is (571)272-2168. The examiner can normally be reached M-F (7:00am - 4:00pm) CST.
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, Gregory Morse can be reached at (571) 272-3838. 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.
/OWAIS I MEMON/Examiner, Art Unit 2663
Prosecution Insights