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 .
Specification
The disclosure is objected to because of the following informalities: typographical and grammatical errors.
Paragraph [0024] recites “illustrating an illustrating an example roof image generation system,” which appears to include duplicated wording.
Paragraph [0026] recites “synthetic images of damage roofs,” which appears to require correction to “synthetic images of damaged roofs.”
Paragraph [0036] recites “different rood material type or style,” which appears to require correction to “different roof material type or style.”
Paragraph [0050] recites “may provided to a convolutional neural network,” which appears to require correction to “may be provided.”
Paragraph [0052] recites “may influence to iterative de-noising process” and “synthetic rood damage images,” which appear to require correction.
Paragraph [0053] recites “and may a decoder,” which appears grammatically incomplete.
Appropriate correction is required.
The disclosure is objected to because of the following informalities: "DALL-E®" [0057] is properly capitalized and marked. However, "LINUX" [0071] is used as a trademark/ product name without a ® symbol, while "WINDOWS® SERVER" in the same paragraph is properly marked. This inconsistent trademark treatment is an informality, require consistent identification of trade names.
Appropriate correction is required.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. § 101 because the claimed invention is directed to a judicial exception without significantly more.
This rejection has been made in accordance with the current USPTO subject matter eligibility framework, including MPEP §§ 2103-2106.07, the 2019 Revised Patent Subject Matter Eligibility Guidance, the October 2019 Patent Eligibility Guidance Update, the 2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence, the July 2024 AI Subject Matter Eligibility Examples, the August 4, 2025 USPTO memorandum titled “Reminders on evaluating subject matter eligibility of claims under 35 U.S.C. § 101,” and the USPTO’s guidance concerning Ex parte Desjardins, Appeal No. 2024-000567. The claims have been evaluated under the broadest reasonable interpretation, and the claims have been considered as a whole.
Step 1: Statutory category
Independent claim 1 is directed to a method and therefore falls within the statutory category of a process. Independent claim 9 is directed to a computer server comprising one or more processors and memory storing computer-executable instructions and therefore falls within the statutory category of a machine. Independent claim 16 is directed to one or more non-transitory computer-readable media and therefore falls within the statutory category of a manufacture. Accordingly, the analysis proceeds to Step 2A.
Step 2A, Prong One: Judicial exception
Independent claim 1 recites, in substance: receiving text data indicating at least one of a roof surface attribute or a roof damage attribute; providing the text data as input to a generative model, wherein the generative model is a machine learning model trained to generate images of damaged roof surfaces; and generating, based on an output of the generative model, a synthetic image of a damaged roof. The claim therefore recites receiving descriptive information, using the descriptive information as input to a mathematical/ statistical generative model, and generating output image data based on the model output. The claimed workflow is shown generally in the application’s Fig. 1, which depicts receiving text input, encoding text, executing a generative ML model using text embeddings, and outputting synthetic roof-damage images.
These limitations recite an abstract idea, namely collecting descriptive information, analyzing or transforming the descriptive information using a mathematical machine-learning model, and generating synthetic image information based on the model output. The claim recites information collection, mathematical/ statistical model processing, and generation of digital content. The “text data,” “roof surface attribute,” “roof damage attribute,” “generative model,” “ML model,” “output,” and “synthetic image” are used as data objects and mathematical/ modeling constructs in a data-generation process. The claim does NOT recite an improvement to the way roof images are physically captured, sensed, stored, compressed, transmitted, or displayed. Rather, the claim uses a generative machine-learning model as a mathematical tool to produce synthetic image data corresponding to descriptive roof/ damage information. Furthermore, reading a text description of a scene (e.g., a damaged roof) and drawing an image depicting that scene is a mental process historically performed by humans (e.g., human illustrators or estimators).
The claim is similar in character to claims that courts have found abstract where the focus is collecting information, analyzing or manipulating that information using mathematical or algorithmic techniques, and presenting or acting on the result. In Electric Power Group, LLC v. Alstom S.A., the Federal Circuit recognized claims directed to collecting and analyzing information as abstract. The present claims similarly collect descriptive roof/ damage information, process that information through a mathematical/ statistical model, and generate output data in the form of a synthetic roof image.
Independent claims 9 and 16 recite substantially the same abstract idea in computer-server form and computer-readable medium form using generic computer processes. Merely implementing the same mathematical generative process using a generic server or media does not avoid the judicial exception.
Dependent claims 2-4, 7-8, 10-11, 14-15, and 17-18 further limit the method, server, and media to using a "diffusion model," using the text data as a "conditioning input" or "diffusion guidance input," performing multiple executions based on "random noise samples," and adding data identifying an object to be included in the image. These limitations recite the use of mathematical models and statistical algorithms (such as iterative noise manipulation and diffusion) as tools for performing the abstract text-to-image translation task. The claims do not recite any specific, unconventional hardware architecture or technical improvement to machine-learning hardware technology itself.
Claims 5, 12, and 19 recite determining image tags based on the text/input data and training a second machine-learning model using training data that includes the synthetic image and image tags, wherein the second model is trained to detect roof damage based on input image data. These limitations recite data labeling, creation of training data, and training a machine-learning model using generated data and tags. Such limitations involve mathematical/ statistical model training and data organization rather than a specific improvement to computer technology or roof-inspection technology.
Claims 6, 13, and 20 recite that the synthetic image depicts manufactured roof damage and that the image tags include a tag indicating that the damaged roof is manufactured damage. These limitations merely specify the semantic subject matter of the generated image and associated tag.
Accordingly, claims 1-20 recite an abstract idea under Step 2A, Prong One.
Step 2A, Prong Two: Practical application
The additional elements, considered individually and in combination, do not integrate the abstract idea into a practical application.
The recited “generative model,” “machine learning model,” “diffusion model,” “conditioning input,” “diffusion guidance input,” “random noise sample,” “image tags,” “second machine learning model,” “computer server,” “processor,” “memory,” and “non-transitory computer-readable media” amount to mathematical models, data objects, and generic computer implementation of the abstract synthetic-data generation concept.
The claims do not recite a particular improvement to computer or imaging technology. They do not improve how a roof image is physically captured, sensed, encoded, compressed, stored, transmitted, rendered, or displayed. The claims also do not improve a camera, drone, sensor, roof-inspection device, image-acquisition system, display device, or computer-server architecture. Instead, the claims generate synthetic image data corresponding to roof/ damage attributes using generic machine-learning model operations.
The claims further do not recite a particular improvement to artificial-intelligence or machine-learning technology. The claims do not recite a new diffusion architecture, a specific denoising network improvement, a particular sampler that improves inference speed, a memory-saving model structure, a parameter-update mechanism that improves model operation, a loss function that improves model training, or a specific data structure that improves computer functionality. The claimed generative model and diffusion model are recited functionally as tools for generating synthetic roof-damage images from descriptive input data.
This analysis is consistent with the USPTO’s 2024 AI subject matter eligibility guidance and AI examples, which emphasize that AI-related claims may be eligible when they recite a specific technological improvement or otherwise integrate a judicial exception into a practical application. Here, the claims do not recite such a specific technological improvement. Instead, the claims use generic computer operations and generic machine-learning model operations to receive descriptive roof/damage data, condition or guide a generative model, and output synthetic image data.
This case is distinguishable from Ex parte Desjardins. In Desjardins, the claims were found to reflect an improvement in machine-learning technology itself, including training a machine-learning model on a series of tasks while preserving prior knowledge and reducing complexity/ storage burdens. Here, the claims do not recite a particular machine-learning improvement, parameter-update mechanism, memory-saving arrangement, continual-learning structure, or reduced-complexity model operation. The claimed generative/ diffusion model merely automates the abstract task of creating synthetic image data corresponding to text/ attribute inputs.
Nor does limiting the abstract idea to the environment of roof-damage image generation or insurance/ property inspection make the claims eligible. In Recentive Analytics, Inc. v. Fox Corp., the Federal Circuit rejected the argument that applying machine learning to a new field of use was sufficient for eligibility where the claims did not recite a technical improvement to the machine-learning process itself. Similarly here, applying generative machine learning or diffusion image generation to roof-damage training data is a field-of-use limitation, not an integration of the abstract idea into a practical application.
The downstream training limitations in claims 5, 12, and 19 likewise do not integrate the abstract idea into a practical application. These claims recite using the synthetic image and image tags as training data for a second machine-learning model trained to detect roof damage. However, the claims do not recite a specific improved training method, model architecture, parameter-update technique, image-preprocessing technique, or computer-vision improvement. They merely state the intended use of the generated synthetic images and tags as training data.
Accordingly, claims 1-20 do not integrate the judicial exception into a practical application under Step 2A, Prong Two.
Step 2B: Inventive concept
The additional elements, considered both individually and as an ordered combination, do not amount to significantly more than the abstract idea.
The claims use generic computer components to perform ordinary computer functions, including receiving data, providing data as model input, executing a generative or diffusion model, receiving model output, generating image data, determining tags, and training another machine-learning model using image data and tags. These are conventional data-processing and mathematical/ modeling operations performed using generic computer technology.
The ordered combination also does not provide an inventive concept. The ordered combination follows the abstract idea itself: receive descriptive roof/ damage information, use that information to condition or guide a generative/ diffusion model, generate synthetic roof-damage image data, optionally generate tags, and optionally use the generated image data and tags to train a second model. This ordered combination is no more than the abstract synthetic-data generation and training-data preparation concept implemented on generic computer components.
Dependent claims 2-4, 10-11, and 17-18 recite diffusion-model execution, conditioning inputs, diffusion guidance inputs, random noise samples, and repeated executions. These limitations merely specify known mathematical/statistical tools for generating different synthetic images and do not add significantly more than the abstract idea.
Dependent claims 5-8, 12-15, and 19-20 recite image tags, downstream model training, manufactured roof damage, roof material type, roof pitch, roof age, damage location, damage cause, damage severity, and objects to be included in the synthetic image. These limitations merely specify the content of the input data, output data, labels, and field-of-use context. They do not recite an unconventional computer component or a technological improvement that transforms the abstract idea into patent-eligible subject matter.
Independent claims 9 and 16 recite system and computer-readable-medium counterparts using generic processors, memory, and executable instructions to perform substantially the same operations. The recitation of generic computer components does not transform the abstract idea into patent-eligible subject matter.
Accordingly, claims 1-20 are directed to a judicial exception without significantly more and are therefore rejected under 35 U.S.C. § 101.
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 (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.
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 1-5, 8-12, and 15-19 are rejected under 35 U.S.C. §103 as being unpatentable over Asami (Asami et al, "Data Augmentation with Synthesized Damaged Roof Images Generated by GAN", 2022) in view of Rombach (Rombach et al, "High-Resolution Image Synthesis with Latent Diffusion Models". arXiv:2112.10752 [Cs]., 2022).
Regarding claim 1, Asami teaches a method for generating synthetic roof images ( [Abstract], [Sec. “Generation of synthesized images”], [Figs. 3-4]: Asami teaches GAN-based data augmentation that generates realistic synthesized disaster data; Asami further teaches that StyleGAN2 was used to generate high-quality synthetic damaged-roof images, and that examples of real and generated damaged-roof images are shown in Figs. 3 and 4. ), the method comprising:
receiving
( [Page 260 > Sec. “Generation of synthesized images”], [Figs. 3-4]: Asami teaches using StyleGAN2 to generate synthesized damaged-roof images from a trained damaged-roof image dataset. Asami teaches that each module of the synthesis network receives input data including a variable and random noise, wherein the variable determines global roof-image features, such as the shape of the roof [roof surface attribute], and the random noise determines local roof-image features, such as the texture of the roof [roof surface attribute]. Asami further teaches training StyleGAN2 using 714 damaged-roof images and generating synthesized damaged-roof images. Accordingly, under the broadest reasonable interpretation, Asami teaches receiving data indicating at least one roof surface attribute, such as roof shape or roof texture, for generation of synthesized damaged-roof images. )
providing the
( [Page 260 > Sec. “Generation of synthesized images”], [Figs. 3-4]: Asami teaches using StyleGAN2 as a machine-learning generative model to generate synthesized damaged-roof images. Asami teaches that the StyleGAN2 model was trained using a damaged-roof image dataset and that, after training, the model generated synthesized damaged-roof images. Asami further teaches that the StyleGAN synthesis network receives input data including a variable/latent input and random noise, where the variable determines global image features, such as roof shape, and the random noise determines local image features, such as roof texture. Accordingly, Asami teaches providing data as input to a generative ML model trained to generate images of damaged roof surfaces. )
generating, based on an output of the generative model, a synthetic image of a damaged roof.
( [Abstract], [Sec. “Generation of synthesized images”], [Figs. 3-4]: Asami teaches “Data Augmentation with Synthesized Damaged Roof Images Generated by GAN” and teaches using a trained StyleGAN2 generative model to generate synthesized damaged-roof images. Asami teaches that, after training StyleGAN2 using a damaged-roof image dataset, synthesized damaged-roof images are generated from the output of the trained generative model, and Figs. 3-4 show examples of real damaged-roof images and generated synthesized damaged-roof images. )
Asami teaches generating synthesized damaged-roof images using a trained StyleGAN2 generative model, but does not expressly disclose controlling the generated roof-image content using text data, where Rombach teaches:
receiving text data indicating at least one of an object attribute or a scene attribute;
( [Abstract], [Sec. 3.3], [Fig. 3-4]: Rombach teaches latent diffusion models for high-resolution image synthesis and teaches that introducing cross-attention layers makes diffusion models flexible generators for conditioning inputs, including text and bounding boxes. Rombach further teaches that text conditioning may be used to control the image-generation process, and that latent diffusion models perform text-conditioned synthesis based on such conditioning input. Accordingly, Rombach teaches receiving text data indicating image content to be generated, such as an object attribute or a scene attribute, for use in conditioned image generation. )
providing the text data as input to a generative model, wherein the generative model is a machine learning (ML) model trained to generate images of objects or scenes;
( [Abstract], [Sec. 3.3], [Fig. 3-4]: Rombach teaches latent diffusion models for high-resolution image synthesis, wherein the neural-network diffusion model is implemented as a machine-learning generative model. Rombach further teaches using cross-attention layers to provide conditioning inputs, including text, to the latent diffusion model so that image generation is controlled by the conditioning input. Accordingly, Rombach teaches providing text data as input/ conditioning input to a generative ML model trained to generate images of objects or scenes. )
It would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Asami’s synthesized damaged-roof image generation method to use Rombach’s text-conditioned latent diffusion generation technique, thereby allowing damaged-roof images to be generated based on user-specified text attributes. Such a modification would have predictably provided controllable and diverse synthetic damaged-roof training images, reduced the need to collect scarce real-world damaged-roof examples, and amounted to applying Rombach’s known text-conditioning technique to Asami’s known synthetic damaged-roof image-generation task according to its established function, with a reasonable expectation of success.
Regarding claim 2, Asami [as modified by Rombach] teaches the method of claim 1, wherein the generative model is a diffusion model configured to generate the synthetic image by iteratively performing diffusion inference operations.
( Rombach, [Abstract], [Sec. 2], [Sec. 3.1], [Sec. 3.3], [Fig. 3]: Rombach teaches latent diffusion models for high-resolution image synthesis. Rombach explains that diffusion models synthesize images by starting from noise and performing a learned reverse/ denoising process through multiple sequential steps. Rombach further teaches applying the diffusion model in a latent space and using a denoising UNet architecture to generate image samples. Therefore, Rombach teaches a diffusion model configured to generate a synthetic image by iteratively performing diffusion inference operations. )
Regarding claim 3, Asami [as modified by Rombach] teaches the method of claim 2, wherein the diffusion model is configured to use the text data as at least one of:
a conditioning input to the diffusion model; or
( Rombach, [Abstract], [Sec. 3.3], [Fig. 3]: Rombach teaches that latent diffusion models may be conditioned on inputs such as text or bounding boxes, and that cross-attention layers are introduced into the diffusion model architecture to make the model a flexible generator for such conditioning inputs. Rombach further teaches that text conditioning is used to control the image-generation process. )
a diffusion guidance input during the diffusion inference operations.
( Rombach, [Sec. 3.3], [Fig. 3]: Rombach teaches conditioning the latent diffusion model during the denoising/ image-synthesis process using conditioning information, including text. The text-conditioning information is supplied to the diffusion model through cross-attention during generation so that the iterative denoising/ inference process is guided toward image content corresponding to the text input. )
Regarding claim 4, Asami [as modified by Rombach] teaches the method of claim 2, further comprising:
performing a first execution of the diffusion model based on a first random noise sample and a conditioning input based on the text data;
( Rombach, [Abstract], [Sec. 2], [Sec. 3.3], [Fig. 3]: Rombach teaches latent diffusion models for high-resolution image synthesis, wherein diffusion models generate image data through a sequential denoising process. Rombach further teaches that diffusion generation may be controlled using conditioning inputs, including text, by introducing cross-attention layers into the diffusion model architecture; (a text condition
y
is encoded via
τ
θ
y
and supplied to cross-attention layers of the denoising network, so each sampling run executes the diffusion model based on a random noise sample and a conditioning input derived from the text). )
receiving the synthetic image based on the first execution of the diffusion model; and
( Rombach, [Abstract], [Sec. 2-3], [Fig. 2-3]: Rombach explains that, after completing the denoising process from the initial noise latent to a clean latent representation, the model decodes the latent back to pixel space using a VAE decoder, thereby outputting a synthesized image corresponding to the text condition. )
performing a second execution of the diffusion model based on a second random noise sample and the conditioning input; and
( Rombach, [Abstract], [Sec. 2-4], [Fig. 3]: ombach teaches that multiple samples can be drawn by re-running the diffusion process from different random noise seeds while keeping the same conditioning input fixed (e.g., the same text prompt). Each run starts from a different Gaussian noise latent and applies the same text-conditioned denoising, thereby performing a second execution of the diffusion model based on a second random noise sample and the same conditioning input. )
receiving a second synthetic image, different from the synthetic image, based on the second execution of the diffusion model.
( Rombach, [Abstract], [Sec. 2-4], [Fig. 5]: Rombach shows that sampling from different noise latents under the same text condition produces different synthesized images that all match the textual description but vary in appearance and layout, demonstrating diversity in samples. )
Regarding claim 5, Asami [as modified by Rombach] teaches the method of claim 1, further comprising:
determining one or more image tags based on the text data; and
( Asami, [Dataset], [Sec. “Generation of synthesized images”], [Sec. “Training of the classification model”], [Table 2]: Asami teaches using damaged-roof image data in roof-damage-rate classes, including no damage, -25 damage, 25-50 damage, 50-75 damage, and 75-damage, and teaches that generated synthesized damaged-roof images are divided into roof segments and annotated for use in training a damage-rate classification model. Rombach, [Sec. 3.3], [Fig. 3]: Rombach teaches text-conditioned image generation using conditioning inputs, including text. Accordingly, where Rombach’s text data indicates the desired roof/ damage image content and Asami teaches annotating generated damaged-roof images for classification, Asami [as modified by Rombach] teaches determining one or more image tags based on the text data used to condition generation of the synthetic damaged-roof image. )
training a second machine learning model using training data including the synthetic image and the image tags, wherein the second machine learning model is trained to detect roof damage based on input image data.
( Asami, [Sec. “Generation of synthesized images”], [Sec. “Training of the classification model”], [Fig. 2]: Asami teaches that generated synthesized damaged-roof images are divided into roof segments, annotated, and added to the original dataset. Asami further teaches verifying the StyleGAN2-based data augmentation using a roof-damage-rate classification model, such as a ResNet-based classification model, trained using the dataset augmented with synthesized damaged-roof images. )
Regarding claim 8, Asami [as modified by Rombach] teaches the method of claim 1, further comprising:
receiving additional data identifying an object to be included in the synthetic image of the damaged roof; and
( Rombach, [Abstract], [Sec. 3.3], [Sec. 4.3], [Fig. 3]: Rombach teaches that latent diffusion models are highly flexible conditional generators capable of receiving complex text prompts that identify multiple specific objects or layout arrangements to be included in the synthesized image. Because latent diffusion models can incorporate various conditioning inputs beyond text, including spatial annotations such as bounding boxes or segmentation maps, which identify specific objects to be included or emphasized in the generated image. The paper explains that cross-attention layers make diffusion models “powerful and flexible generators for general conditioning,” including object-level and scene-level attributes. )
providing the additional data as a conditioning input to the generative model.
( Rombach, [Abstract], [Sec. 3.3], [Fig. 3]: Rombach teaches providing the entire text prompt (including any additional data identifying objects) to domain-specific text encoders, which project the text into intermediate representations. These representations are then mapped as conditioning inputs into the cross-attention layers of the generative diffusion model to synthesize the final image containing the requested objects. )
Regarding claims 9-12, and 15-19 the rationale provided in the rejection of claims 1-5 and 8 is incorporated herein. Further, the method of claims 1-5 and 8 corresponds to the computer server of claims 9-12 and 15, as well as the non-transitory computer-readable media of claims 16-19 ( Asami, [Sec. “Generation of synthesized images”]: Asami teaches computer system with an NVIDIA GeForce GTX 2080 8 GB GPU. ), and performs the steps disclosed herein.
Claims 6-7, 13-14, and 20 are rejected under 35 U.S.C. §103 as being unpatentable over Asami [as modified by Rombach] in view of Howe (Howe et al, US 2017/0270650 A1, 2017).
Regarding claim 6, Asami [as modified by Rombach] teaches the method of claim 5, wherein:
Asami [as modified by Rombach] teaches generating synthetic damaged-roof images and using corresponding annotations/ tags for roof-damage classification/ training, but fails to expressly disclose where Howe teaches:
the synthetic image comprises a representation of manufactured roof damage; and
( [0052], [0057], [0075], [Fig. 3]: Howe teaches that a corpus of labeled imagery may include images showing legitimate hail damage, fraudulent damage [manufactured roof damage], damage by other causes, and no damage. Howe further teaches damage-cause metadata for roof/ property images, including hail, manufacturing defect, installation defect, people walking on the roof, fallen limb damage, and fraudulent damage such as by a hammer. Howe also teaches fraud detection for determining whether a homeowner intentionally damaged the roof, such as with a ball-peen hammer. )
the image tags include a tag indicating that the damaged roof in the synthetic image is manufactured damage.
( [0055]-[0057], [0069]: Howe teaches that roof damage can be labeled with metadata, including damage location, damage severity, and damage cause. Howe further teaches that damage labels can correspond to images and identify an image as containing a particular type of damage, and that such labeled data is used to train a classifier. )
It would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to include manufactured/ fraudulent roof damage as one of the damage classes or tags in Asami [as modified by Rombach]’s synthetic damaged-roof image generation and training-data workflow, because Howe teaches that fraudulent or intentionally man-made roof damage was a known roof-damage cause to be detected and labeled in roof/ property damage assessment. Such a modification would have predictably generated training examples for a known damage class, improved training-data coverage for rare or high-value insurance assessment scenarios, and amounted to applying Howe’s known roof-damage classification/ tagging information to Asami [as modified by Rombach]’s known synthetic damaged-roof training-data generation workflow according to its established function, with a reasonable expectation of success.
Regarding claim 7, Asami [as modified by Rombach] teaches the method of claim 1, wherein the text data indicates a roof surface attribute comprising at least one of:
a roof material type;
a roof pitch; or
a roof age, and
( [0040], [0058], [0063], [0067]: Howe teaches receiving and analyzing voice and/or text commentary related to property-damage images, wherein the commentary may describe what is contained in the images and the specific part of the property captured in the frame. Howe further teaches property characteristics including roof and shingle characteristics such as pitch, orientation, dimensions of each section of the roof, shingle manufacturer, model, quality, age, color, and number of shingle layers. Howe also teaches roof material examples, including asphalt shingle composition roofs, traditional three-tab shingles, and architectural or laminated shingles. )
wherein the text data indicates a roof damage attribute comprising at least one of:
a damage location;
a damage cause; or
a damage severity.
( [0040], [0055-0057], [0067]: Howe teaches receiving and analyzing voice and/or text commentary related to property-damage images, wherein the commentary may include commentary about the severity of the damage. Howe further teaches image metadata including damage location labeling resolution, damage severity, and damage cause. Howe teaches that damage location may be labeled at different resolutions, including pixel level, shingle-tab level, roof-section level, whole-roof/ property level, or whole-image level. Howe teaches that damage severity may indicate whether the damage is cosmetic, dents the surface, or fractures the shingle mat. Howe also teaches that damage cause may include hail, manufacturing defect, installation defect, people walking on the roof, fallen limb damage, and fraudulent damage such as by a hammer. )
It would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to include Howe’s known roof surface attributes and roof damage attributes in the text/conditioning data used in Asami [as modified by Rombach]’s synthetic damaged-roof image generation process, because Howe teaches that such attributes were known and useful metadata for roof/ property damage assessment and classifier training. Such a modification would have predictably allowed generation of synthetic damaged-roof images reflecting known roof-assessment conditions, such as roof pitch, age, material, damage location, cause, and severity, thereby improving the diversity and usefulness of training images for roof-damage detection/ classification.
Regarding claims 13-14, and 20 the rationale provided in the rejection of claims 6-7 is incorporated herein. Further, the method of claims 6-7 corresponds to the computer server of claims 13 and 20, as well as the non-transitory computer-readable media of claim 14 ( Asami, [Sec. “Generation of synthesized images”]: Asami teaches computer system with an NVIDIA GeForce GTX 2080 8 GB GPU. ), and performs the steps disclosed herein.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KEN KUDO whose telephone number is (571)272-4498. The examiner can normally be reached M-F 8am - 5pm.
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, Vincent Rudolph can be reached at 571-272-8243. 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.
KEN KUDO
Examiner
Art Unit 2671
/KEN KUDO/ Examiner, Art Unit 2671
/VINCENT RUDOLPH/ Supervisory Patent Examiner, Art Unit 2671
Prosecution Insights