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. Information Disclosure Statement The information disclosure statement(s) (IDS) submitted on February 12, 2023 and October 19, 2023 is/are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement(s) is/are being considered by the examiner. Drawings The drawings are objected to as failing to comply with 37 CFR 1.84(p)(5) because they include the following reference character(s) not mentioned in the description: Reference element 304 in FIG. 3 and FIG. 6 is missing a description in the specification. Corrected drawing sheets in compliance with 37 CFR 1.121(d), or amendment to the specification to add the reference character(s) in the description in compliance with 37 CFR 1.121(b) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either “Replacement Sheet” or “New Sheet” pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance. In addition to Replacement Sheets containing the corrected drawing figure(s), applicant is required to submit a marked-up copy of each Replacement Sheet including annotations indicating the changes made to the previous version. The marked-up copy must be clearly labeled as “Annotated Sheets” and must be presented in the amendment or remarks section that explains the change(s) to the drawings. See 37 CFR 1.121(d)(1). Failure to timely submit the proposed drawing and marked-up copy will result in the abandonment of the application. Specification The disclosure is objected to because of the following informalities: Paragraph [0027] recites anomaly learning module 101 and also anomaly detection model 101 . However, throughout the specification the reference number 101 is associated with the anomaly learning module . 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 1-14 are rejected under 35 U.S.C.101 because the claimed invention is directed to an abstract idea without significantly more. Step 1: Claims 1-7 are directed to a process. Claims 8-14 are directed to a machine or an article of manufacture. With respect to claim(s) 1 and 8: 2A Prong 1 : The claim(s) recite(s) an abstract idea. Specifically: predicting a defect class for an input image […] ( Mental process – A person can mentally evaluate an image to predict (e.g. guess or think of) a defect class for the image. The claim does not limit the plain meaning of the term “a defect class”, which as explained by paragraph [0025] of the specification amounts to classifying data into a category. – see MPEP § 2106.04(a)(2)(III)) generating a coarse localization map for the input image using the CNN classification model ; ( Mathematical concepts – generating a coarse localization map involves mathematical calculations of using the Grad-CAM algorithm (see [0037]) – see MPEP § 2106.04(a)(2)(I)) generating an anomaly map for the input image using an anomaly detection model ; ( Mathematical concepts – generating an anomaly map involves mathematical calculations of using the Mahalanobis distance (see [0033-0034]) – see MPEP § 2106.04(a)(2)(I)) predicting a defect location in the input image via location ensemble ; ( Mathematical concepts – predicting a defect location via location ensemble involves mathematical calculations (see [0042]) – see MPEP § 2106.04(a)(2)(I)) wherein the anomaly detection model is built by learning a distribution of the real non-defect images . ( Mathematical concepts – learning a distribution, such as a normal distribution, to build a model involves mathematical calculations (see [0027]) – see MPEP § 2106.04(a)(2)(I)) If claim limitations, under their broadest reasonable interpretation, cover performance of the limitations as a mathematical or mental process, but for the recitation of generic computer components, then the claim limitations fall within the mathematical or mental process grouping of abstract ideas. Accordingly, the claim “recites” an abstract idea. 2A Prong 2 : The additional elements recited in the claim(s) do not integrate the abstract idea into a practical application, individually or in combination. Additional elements: […] using a convolutional neural network (CNN) classification model ; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) wherein the CNN classification model is built based on a CNN encoder and is finetuned with one or more labeled real defect images ; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) wherein the CNN encoder is trained by multi-task self-supervised learning with one or more real non-defect images and one or more synthetic defect images ; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) (Claim 8) an anomaly learning module having at least one processor configured to perform a learning of an anomaly detection model ; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) (Claim 8) a self-supervised pretraining module having at least one processor configured to perform multi-task self-supervised learning to train a CNN encode r; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) (Claim 8) a finetuning module having at least one processor configured to build and train a convolutional neural network (CNN) classification model ; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) (Claim 8) an inference module having at least one processor configured to perform : (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) Since the claim as a whole, looking at the additional elements individually and in combination, does not contain any other additional elements that are indicative of integration into a practical application, the claim is directed to an abstract idea.” 2B : The claim(s) do(es) not include additional elements that are sufficient to amount to significantly more than the judicial exception. Additional elements: […] using a convolutional neural network (CNN) classification model ; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) wherein the CNN classification model is built based on a CNN encoder and is finetuned with one or more labeled real defect images ; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) wherein the CNN encoder is trained by multi-task self-supervised learning with one or more real non-defect images and one or more synthetic defect images ; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) (Claim 8) an anomaly learning module having at least one processor configured to perform a learning of an anomaly detection model ; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) (Claim 8) a self-supervised pretraining module having at least one processor configured to perform multi-task self-supervised learning to train a CNN encode r; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) (Claim 8) a finetuning module having at least one processor configured to build and train a convolutional neural network (CNN) classification model ; (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) (Claim 8) an inference module having at least one processor configured to perform : (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) Considering the additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible . With respect to claim(s) 2 and 9: 2A Prong 1 : The claim(s) recite(s) an abstract idea. Specifically: aligning the input image before the generation of the anomaly map . ( Mathematical concepts – aligning the input image involves mathematical calculations such as transformations by a Spatial Transformer Network (see [0046]) – see MPEP § 2106.04(a)(2)(I)) 2A Prong 2 : The additional elements recited in the claim(s) do not integrate the abstract idea into a practical application, individually or in combination. Additional elements: (Claim 9) wherein the inference module is further configured to (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) 2B : The claim(s) do(es) not include additional elements that are sufficient to amount to significantly more than the judicial exception. Additional elements: (Claim 9) wherein the inference module is further configured to (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible. Therefore, the claim is not patent eligible . With respect to claim(s) 3 and 10: 2A Prong 1 : The claim(s) recite(s) an abstract idea. Specifically: wherein the input image is aligned by transforming a main part of the input image to match a pre-selected golden sample . ( Mathematical concepts – aligning the input image involves mathematical calculations such as transformations by a Spatial Transformer Network (see [0046]) – see MPEP § 2106.04(a)(2)(I)) Additionally, the claim(s) do not recite any new additional elements that would amount to an integration of the abstract idea into a practical application (individually or in combination) or significantly more than the judicial exception. Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible. Therefore, the claim is not patent eligible . With respect to claim(s) 4 and 11: 2A Prong 1 : The claim(s) recite(s) an abstract idea. Specifically: […] identifying one or more anomaly regions in one or more unlabeled real images […] ( Mental process – A person of ordinary skill van mentally evaluate unlabeled real images to identify anomaly regions in one or more of the unlabeled real images – see MPEP § 2106.04(a)(2)(III)) 2A Prong 2 : The additional elements recited in the claim(s) do not integrate the abstract idea into a practical application, individually or in combination. Additional elements: wherein the synthetic defect images are generated using the anomaly detection model […] (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) […] and overlaying the anomaly regions on the real non-defect images. (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) 2B : The claim(s) do(es) not include additional elements that are sufficient to amount to significantly more than the judicial exception. Additional elements: wherein the synthetic defect images are generated using the anomaly detection model […] (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) […] and overlaying the anomaly regions on the real non-defect images. (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible. Therefore, the claim is not patent eligible . With respect to claim(s) 5 and 12: 2A Prong 2 : The additional elements recited in the claim(s) do not integrate the abstract idea into a practical application, individually or in combination. Additional elements: (Claim 5) wherein the CNN classification model is built and trained by : (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) (Claim 12) wherein at least one processing unit of the finetuning module further performs the following for building and training the CNN classification model : (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) adding the CNN encoder with a randomly initialized linear classifier; and finetuning the CNN classification model with the labeled real defect images . (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) 2B : The claim(s) do(es) not include additional elements that are sufficient to amount to significantly more than the judicial exception. Additional elements: (Claim 5) wherein the CNN classification model is built and trained by : (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) (Claim 12) wherein at least one processing unit of the finetuning module further performs the following for building and training the CNN classification model : (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) adding the CNN encoder with a randomly initialized linear classifier; and finetuning the CNN classification model with the labeled real defect images . (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible. Therefore, the claim is not patent eligible . With respect to claim(s) 6 and 13: 2A Prong 1 : The claim(s) recite(s) an abstract idea. Specifically: taking a weighted sum of the coarse localization map generated […] and the anomaly map generated […] to obtain an intermediate defect location map; and applying a binary threshold to the intermediate defect location map to obtain the final defect location . ( Mathematical concepts – these steps recite mathematical calculations (see [0057-0060]) – see MPEP § 2106.04(a)(2)(I)) 2A Prong 2 : The additional elements recited in the claim(s) do not integrate the abstract idea into a practical application, individually or in combination. Additional elements: by the CNN classification model (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) by the anomaly detection model (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) 2B : The claim(s) do(es) not include additional elements that are sufficient to amount to significantly more than the judicial exception. Additional elements: by the CNN classification model (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) by the anomaly detection model (Mere instructions to apply an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).) Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible. Therefore, the claim is not patent eligible . With respect to claim(s) 7 and 14: 2A Prong 1 : The claim(s) recite(s) an abstract idea. Specifically: performing a binary classification task with the real non-defect and the synthetic defect images using a classification head to calculate a cross-entropy loss; performing a contrastive learning task with the real non-defect and the synthetic defect images using a contrastive head to calculate a contrastive loss; and updating the weights of the CNN encoder and two heads to minimize a weighted sum of the cross-entropy loss and the contrastive loss. ( Mathematical concepts – these steps recite mathematical calculations (see [0057-0060]) – see MPEP § 2106.04(a)(2)(I)) Additionally, the claim(s) do not recite any new additional elements that would amount to an integration of the abstract idea into a practical application (individually or in combination) or significantly more than the judicial exception. Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible. Therefore, the claim is not patent eligible . 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. Claims 1, 6, 8, and 13 are rejected under 35 U.S.C. 103 as being unpatentable over MEMO (US 20180322623 A1) in view of LI (" CutPaste : Self-Supervised Learning for Anomaly Detection and Localization"), HUANG ("Registration based Few-Shot Anomaly Detection"), KR 20210141060 A, and JEZEQUEL ("Efficient Anomaly Detection Using Self-Supervised Multi-Cue Tasks"), hereafter MEMO, LI, HUANG, KR 20210141060 A, and JEZEQUEL respectively. Regarding Claim 1 , MEMO teaches: predicting a defect class for an input image using a convolutional neural network (CNN) classification model; (MEMO [0056] teaches: "A defect analysis system 300 can use the trained CNN 310 to classify target objects as having one or more defects based on captured 3-D images (i.e., an input image ) 14t of those target objects. [...] A defect detection module 370 may then classify the objects as belonging to one or more classes (i.e., predicting a defect class ) (shown in FIG. 1A as 18A, 18B, and 18C) corresponding to the absence of defects or the presence of particular types of defects." Furthermore, defect detection module 370 is part of system 300 that uses the CNN (i.e., using a convolutional neural network ) in the process of classifying the object as belonging to one or more classes. MEMO [0159] teaches: “Accordingly, in some embodiments, different convolutional neural networks 310 are trained to detect defects in different parts of the object, and, in some embodiments, different convolutional neural networks 310 are trained to detect different classes or types of defects.”) wherein the CNN classification model is built based on a CNN encoder and is finetuned with one or more labeled real defect images ; (MEMO [0158] teaches starting from a previously trained CNN (i.e., CNN classification model is built based on a CNN encoder ) and performing retraining (i.e., and is finetuned ) using a different set (i.e., with one or more ) with data from the specific application of interest. MEMO [0159] teaches: “These embodiments allow the resulting convolutional neural networks to be fine-tuned to detect particular types of defects and/or to detect defects in particular parts.” MEMO [0054-0056] teaches that the training data may include captured images of defective objects with labels (i.e., labeled real defect images ) indicating locations and types (or classifications) of defects found on the labeled objects.”) MEMO is not relied upon for teaching: generating a coarse localization map for the input image using the CNN classification model; generating an anomaly map for the input image using an anomaly detection model; and predicting a defect location in the input image via location ensemble; wherein the CNN encoder is trained by multi-task self-supervised learning with one or more real non-defect images and one or more synthetic defect images; and wherein the anomaly detection model is built by learning a distribution of the real non-defect images. However, LI teaches: generating a coarse localization map for the input image using the CNN classification model; (LI [pg. 2, Figure 1] teaches: "An image-level representation makes a holistic decision for anomaly detection and is used to localize defect via GradCAM [51]." Furthermore, LI [pg. 2, Figure 1] also teaches localizing defects via GradCAM using a CNN (i.e., using the CNN classification model ).” GradCAM stands for Gradient-weighted Class Activation Mapping, and thus the result of performing defect localization via GradCAM is a coarse localization map .) wherein the CNN encoder is trained by […] self-supervised learning with one or more real non-defect images and one or more synthetic defect images ; (LI [pg. 1, 1. Introduction] teaches: "Our innovation is at designing a novel proxy task for self-supervised learning of representations. Specifically, we formulate a proxy classification task between normal training data and the ones augmented by the CutPaste , the proposed data augmentation strategy that cuts an image patch and pastes at a random location of an image." LI [pg. 2, Figure 1] teaches training a CNN (i.e., CNN encoder ) by self-supervised learning using normal data with no defects (i.e., real non-defect images ) and CutPaste augmented data with defects (i.e., synthetic defect images ). Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of MEMO and LI before them, to include LI’s defect localization via GradCAM and self-supervised learning using normal and defect augmented training data in MEMO’s method for inspection and defect detection. One would have been motivated to make such a combination in order to encourage the model to find local irregularities (LI [pg. 8, 6. Conclusion]). MEMO in view of LI is not relied upon for teaching: generating an anomaly map for the input image using an anomaly detection model; and predicting a defect location in the input image via location ensemble; […] multi-task […] wherein the anomaly detection model is built by learning a distribution of the real non-defect images. However, HUANG teaches: generating an anomaly map for the input image using an anomaly detection model ; (HUANG [pg. 3, 1. Introduction] Given the support set, the normal distribution (i.e., anomaly detection model ) of registered features for the target category is estimated with a statistical-based distribution estimator [8] (i.e., anomaly learning module). Test samples that are out of the statistical normal distribution are considered anomalies.” HUANG [pg. 7-8, 4.3 Inference] teaches: " During inference, test samples that are out of the normal distribution are considered anomalies. For each test image (i.e., input image ) in T test , we use the Mahalanobis distance M( f ij ) to give an anomaly score to the patch in position (i,j) , where (eq. (5)) M f ij = f ij - μ ij T - ij -1 f ij - μ ij The matrix of Mahalanobis distances M= M f ij 1≤i≤W, 1≤j≤H forms an anomaly map .") wherein the anomaly detection model is built by learning a distribution of the real non-defect images. (HUANG [pg. 7, 4.2 Normal Distribution Estimation] teaches: "After achieving the registered features, a statistical-based estimator [8] is used to estimate the normal distribution of target category features, which uses multivariate Gaussian distributions to get a probabilistic representation (i.e., anomaly detection model is built by learning a distribution ) of the normal class (i.e., real non-defect images )." HUANG [pg. 8, 5.1 Experimental Setups] teaches: "The test set contains both images with various kinds of defects (anomaly) and defect-free images (normal).") Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of MEMO, LI, and HUANG before them, to include HUANG’s estimating the normal distribution of registered features for the target category to get a probabilistic representation of the normal class and use of the Mahalanobis distance to compute an anomaly map for each test image in MEMO and LI’s method for inspection and defect detection. One would have been motivated to make such a combination in order to achieve a model that is directly generalizable and with a high potential to be applicable in real-world anomaly detection environments (HUANG [pg. 14, 6. Conclusion]). MEMO in view of LI and HUANG is not relied upon for teaching: predicting a defect location in the input image via location ensemble; […] multi-task […] However, KR 20210141060 A teaches: predicting a defect location in the input image via location ensemble; (KR 20210141060 A [47] teaches: "The weighted summation (i.e., via location ensemble ) image is an average map that reflects spatial information (i.e., predicting a defect location ) common to the differential image that reflects the structural similarity between the original image and the reconstructed image, and the attention map that reflects the location information that should be focused when detecting anomalies according to the coordinate information in physical space.") Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of MEMO, LI, HUANG, and KR 20210141060 A before them, to KR 20210141060 A’s weighted summation that reflecting location information of anomalies in MEMO, LI, and HUANG’s method for inspection and defect detection. One would have been motivated to make such a combination in order to improve the discriminating power of abnormal values and effectively detect defects (KR 20210141060 A [56-58]). MEMO in view of LI, HUANG, and KR 20210141060 A is not relied upon for teaching, but JEZEQUEL teaches: […] multi-task […] (JEZEQUEL [pg. 3, D. SSL Anomaly Detection] teaches: "In this section, we first present how to apply SSL (i.e., self- supervised learning ) for AD (i.e., anomaly detection) and then discuss some state-of-the-art methods exploiting SSL for AD." JEZEQUEL [pg. 4, III. Novel Pretext Tasks] teaches: "We gradually detail the proposed pretext tasks (i.e., multi-task ) for anomaly detection which focus on different visual cues: structure, colorimetry and texture." JEZEQUEL [pg. 10, B. Implementation Details] teaches using a WideResNet (i.e., CNN encoder ) for the feature extractor network ϕ and training the CNN using stochastic gradient descent (SGD) optimizer.) Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of MEMO, LI, HUANG, KR 20210141060 A, and JEZEQUEL before them, to include JEZEQUEL’s self-supervised multi-cue tasks for training a CNN in MEMO, LI, HUANG and KR 20210141060 A’s method for inspection and defect detection. One would have been motivated to make such a combination in order to add tasks for improving the performance on fine-grained anomaly detection problems (JEZEQUEL [pg. 1, Abstract]). Regarding Claim 6 , MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL teaches the elements of claim 1 as outlined above. KR 20210141060 A further teaches: taking a weighted sum of the coarse localization map […] and the anomaly map […] to obtain an intermediate defect location map; and (KR 20210141060 A [47] teaches: "The weighted summation (i.e., taking a weighted sum) image (i.e., intermediate defect location map ) is an average map that reflects spatial information common to the differential image (i.e., anomaly map ) that reflects the structural similarity between the original image and the reconstructed image, and the attention map (i.e., coarse localization map ) that reflects the location information that should be focused when detecting anomalies according to the coordinate information in physical space." The weighted summation image (i.e., intermediate defect location map ) is the result of the weighted sum (i.e., taking a weighted sum ) of an average map, the differential image (i.e., anomaly map ) and the attention map (i.e., coarse localization map ).) applying a binary threshold to the intermediate defect location map to obtain the final defect location . (KR 20210141060 A [55] The abnormal image generating unit 250 receives a calculated image, for example, a weighted sum image (i.e., intermediate defect location map ), and compares the first reference value preset in pixel units with the weighted sum image to generate an abnormal image in which only the abnormal part is emphasized. In this case, the abnormal image generating unit 250 may generate an abnormal image (i.e., to obtain the final defect location ) in which only the abnormal portion is emphasized by changing (i.e., applying ) the pixel value of the pixel that is less than or equal to the first reference value (i.e., a binary threshold) in the weighted sum image (i.e., to the intermediate defect location map ) to a maximum value or a minimum value. That is, only an abnormal portion of the abnormal image output from the abnormal image generating unit 250 may be shaded.) HUANG teaches: […] the anomaly map generated by the anomaly detection model […] (HUANG [pg. 3, 1. Introduction] Given the support set, the normal distribution (i.e., anomaly detection model ) of registered features for the target category is estimated with a statistical-based distribution estimator [8] (i.e., anomaly learning module). Test samples that are out of the statistical normal distribution are considered anomalies.” HUANG [pg. 7-8, 4.3 Inference] teaches: " During inference, test samples that are out of the normal distribution are considered anomalies. For each test image (i.e., input image ) in T test , we use the Mahalanobis distance M( f ij ) to give an anomaly score to the patch in position (i,j) , where (eq. (5)) M f ij = f ij - μ ij T - ij -1 f ij - μ ij The matrix of Mahalanobis distances M= M f ij 1≤i≤W, 1≤j≤H forms an anomaly map .") LI further teaches: […] the coarse localization map generated by the CNN classification model […] (LI [pg. 2, Figure 1] teaches: "An image-level representation makes a holistic decision for anomaly detection and is used to localize defect via GradCAM [51]." Furthermore, LI [pg. 2, Figure 1] also teaches localizing defects via GradCAM using a CNN.”) Regarding Claim 8 , the claim recites similar limitations as corresponding claim 1 and is rejected for similar reasons as claim 1 using similar teachings and rationale. Additionally, MEMO teaches: a finetuning module having at least one processor configured to build and train a convolutional neural network (CNN) classification model; and an inference module having at least one processor configured to perform: (MEMO [0020] teaches: “a processor configured to control the one or more depth cameras; a memory storing instructions that, when executed by the processor, cause the processor to: control the one or more depth cameras to capture the plurality of depth images of the target object; compute a three-dimensional (3-D) model of the target object using the depth images; render one or more views of the 3-D model; compute a descriptor by supplying the one or more views of the 3-D model to a convolutional stage of a convolutional neural network; supply the descriptor to a defect detector to compute one or more defect classifications of the target object; and output the one or more defect classifications of the target object.” MEMO [0055] teaches: "[…] a convolutional neural network (CNN) training module 20, which is configured to train a convolutional neural network 310 for detecting the defects in the training data. The CNN training module 20 may use a pre-trained network […].” MEMO [0158] teaches starting form a previously trained CNN (i.e., based on a CNN encoder) and performing retraining (i.e., fine-tuning ) using a different set (i.e., one or more) with data from the specific application of interest.”) […] having at least one processor […] (MEMO [0020] teaches a processor.) MEMO is not relied upon for teaching: an anomaly learning module […] configured to perform a learning of an anomaly detection model; a self-supervised pretraining module […] configured to perform multi-task self-supervised learning to train a CNN encoder ; However, HUANG teaches: an anomaly learning module […] configured to perform a learning of an anomaly detection model; (HUANG [pg. 7, 4.2 Normal Distribution Estimation] teaches: "After achieving the registered features, a statistical-based estimator [8] (i.e., anomaly learning module ) is used to estimate (i.e., learning ) the normal distribution (i.e., anomaly detection model ) of target category features, which uses multivariate Gaussian distributions to get (i.e., built by ) a probabilistic representation of the normal class (i.e., real non-defect images )." The statistical-based estimator can be implemented in MEMO’s processor.) Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of MEMO and HUANG before them, to include HUANG’s statistical-based estimator for estimating the normal distribution of registered features for the target category in MEMO’s method for inspection and defect detection. One would have been motivated to make such a combination in order to achieve a model that is directly generalizable and with a high potential to be applicable in real-world anomaly detection environments (HUANG [pg. 14, 6. Conclusion]). MEMO in view of HUANG is not relied upon for teaching, but JEZEQUEL teaches: a self-supervised pretraining module […] configured to perform multi-task self-supervised learning to train a CNN encoder ; (JEZEQUEL [pg. 8, Algorithm 1] teaches an algorithm (i.e., pretraining module ) for implementing self-supervised multi-cue tasks training of a WideResNet encoder ϕ (i.e., CNN encoder ) for anomaly detection.) Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of MEMO, HUANG, and JEZEQUEL before them, to include JEZEQUEL’s self-supervised multi-cue tasks for training a CNN in MEMO and HUANG’s method for inspection and defect detection. One would have been motivated to make such a combination in order to add tasks for improving the performance on fine-grained anomaly detection problems (JEZEQUEL [pg. 1, Abstract]). Regarding Claim 13 , MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL teaches the elements of claim 6 as outlined above. Additionally, the claim recites similar limitations as corresponding claim 6 and is rejected for similar reasons as claim 6 using similar teachings and rationale. Claims 2-3 and 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL as applied respectively above to claims 1 and 8, and further in view of XIN (US 20220036525 A1), hereafter XIN. Regarding Claim 2 , MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL teaches the elements of claim 1 as outlined above. However, MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL is not relied upon for teaching, but XIN teaches: aligning the input image before the generation of the anomaly map . (XIN [0023] teaches: "Therefore, it may be advantageous to, among other things, orientate the test image in a manner that aligns it (i.e., aligning the input image ) with the template image, determine any meaningful difference between the registered test image and template image to generate a differential image (i.e., before the generation of the anomaly map ), use the registered test image, template image, and differential image to generate a synthetic image, use the synthetic image as the input for a multi-scale detection network, and generate a defect map.") Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of MEMO, LI, HUANG, KR 20210141060 A, JEZEQUEL, and XIN before them, to include XIN’s image alignment prior to generating a differential image in MEMO, LI, HUANG, KR 20210141060 A, and JEZEQUEL’s method for inspection and defect detection. One would have been motivated to make such a combination in order to decrease false detections by orientating the test image in a manner that aligns with the template image (XIN [0024]). Regarding Claim 3 , MEMO in view of LI, HUANG, KR 20210141060 A, JEZEQUEL, and XIN teaches the elements of claim 2 as outlined above. XIN further teaches: wherein the input image is aligned by transforming a main part of the input image to match a pre-selected golden sample . (XIN [0023] teaches: "Therefore, it may be advantageous to, among other things, orientate the test image in a manner that aligns it (i.e., input image is aligned by transforming ) with the template image (i.e., to match a [...] golden sample ), determine any meaningful difference between the registered test image and template image to generate a differential image [...]." XIN [0002] teaches: "The template image is a label without defects and the standard (i.e., golden sample ) by which all subsequent labels are compared." Furthermore, XIN [0004] teaches generating a template image, and under broadest reasonable interpretation, a pre-selected golden sample can be interpreted as generating the template image without defects for comparison.) Regarding Claim 9 , MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL teaches the elements of claim 8 as outlined above. Additionally, the claim recites similar limitations as corresponding claim 2 and is rejected for similar reasons as claim 2 using similar teachings and rationale. Regarding Claim 10 , MEMO in view of LI, HUANG, KR 20210141060 A, JEZEQUEL, and XIN teaches the elements of claim 9 as outlined above. Additionally, the claim recites similar limitations as corresponding claim 3 and is rejected for similar reasons as claim 3 using similar teachings and rationale. Claims 4 and 11 are rejected under 35 U.S.C. 103 as being unpatentable over MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL as applied respectively above to claims 1 and 8, and further in view of LIN ("Few-Shot Defect Segmentation Leveraging Abundant Normal Training Samples Through Normal Background Regularization and Crop-and-Paste Operation") and BERGMANN ("Uninformed Students: Student–Teacher Anomaly Detection with Discriminative Latent Embeddings"), hereafter LIN and BERGMANN respectively. Regarding Claim 4 , MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL teaches the elements of claim 1 as outlined above. However, MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL is not relied upon for teaching, but LIN teaches: wherein the synthetic defect images are generated using […] one or more anomaly regions in one or more unlabeled real images and overlaying the anomaly regions on the real non-defect images . (LIN [pg. 1, I. Introduction] teaches: “In general, it is tedious and time-consuming to manually identify and annotate the defect regions on real-captured images. Therefore, automating the process is in demand.” LIN [pg. 4-5, E. Crop-and-Paste ( CaP )] teaches cropping out the defect region in an anomalous image (i.e., one or more anomaly regions ) in and paste it upon a normal image (i.e., overlaying the anomaly regions on the real non-defect images ), resulting in a augmented anomalous image (i.e., wherein the synthetic defect images are generated ).” LIN [pg. 9, G. Comparison on Few-Shot Anomaly Detection] teaches: “We achieve such high classification accuracy at very low annotation cost by means of leveraging sufficient normal training data which is annotation-free (i.e., unlabeled real images ).”) Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of MEMO, LI, HUANG, KR 20210141060 A, JEZEQUEL, and LIN before them, to include LIN’s crop-and-paste method of defect regions upon a normal image to generate an augmented anomalous image in MEMO, LI, HUANG, KR 20210141060 A, and JEZEQUEL’s method for inspection and defect detection. One would have been motivated to make such a combination MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL in order to improve the accuracy of defect segmentation (LIN [pg. 2, I. Introduction]). MEMO in view of LI, HUANG, KR 20210141060 A, JEZEQUEL, and LIN is not relied upon for teaching, but BERGMANN teaches: […] using the anomaly detection model identifying one or more anomaly regions in one or more unlabeled real images […] (BERGMANN [pg. 2, I. Introduction] Our models (i.e., anomaly detection model ) can be trained end-to-end on large unlabeled image datasets (i.e., unlabeled [...] images ) and make use of all available training data. We introduce scoring functions based on the students’ predictive variance and regression error to obtain dense anomaly maps for the segmentation of anomalous regions (i.e., identifying one or more anomaly regions) in natural images (i.e., in one or more [...] real images ).) Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of MEMO, LI, HUANG, KR 20210141060 A, JEZEQUEL, LIN, and BERGMANN before them, to include BERGMANN’s model segmentation of anomalous regions of large unlabeled image datasets of natural images in MEMO, LI, HUANG, KR 20210141060 A, JEZEQUEL, and LIN’s method for inspection and defect detection. One would have been motivated to make such a combination in order to detect previously unknown defective regions and include them in training to improve the model’s performance (BERGMANN [pg. 1, 1. Introduction]). Regarding Claim 11 , MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL teaches the elements of claim 8 as outlined above. Additionally, the claim recites similar limitations as corresponding claim 4 and is rejected for similar reasons as claim 4 using similar teachings and rationale. Claims 5 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL as applied respectively above to claims 1 and 8, and further in view of CN 114972334 A. Regarding Claim 5 , MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL teaches the elements of claim 1 as outlined above. MEMO further teaches: finetuning the CNN classification model with the labeled real defect images . (MEMO [0158] teaches starting form a previously trained CNN (i.e., based on a CNN encoder ) and performing retraining (i.e., fine-tuning ) using a different set (i.e., one or more ) with data from the specific application of interest. Furthermore, MEMO [0054-0056] teaches that the training data may include captured images of defective objects (i.e., r eal defect images ) with labels (i.e., labeled ) indicating locations and types (or classifications) of defects found on the labeled objects.) MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL is not relied upon for teaching, but CN 114972334 A teaches: adding the CNN encoder with a randomly initialized linear classifier ; (CN 114972334 A [59] teaches: “Each layer is composed of a convolutional network.” CN 114972334 A [89] teaches: "The classifier is replaced with a linear classification layer to perform downstream classification tasks. The linear classification layer weights are randomly initialized.") Accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of MEMO in view of LI, HUANG, KR 20210141060 A, JEZEQUEL, and CN 114972334 A before them, to include CN 114972334 A’s replacing the classifier with a linear classification layer that is randomly initialized in MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL’s method for inspection and defect detection. One would have been motivated to make such a combination in order to not require manual labeling of data during the classification process, further reducing the waste of manpower and material resources (CN 114972334 A [96]). Regarding Claim 12 , MMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL teaches the elements of claim 8 as outlined above. Additionally, the claim recites similar limitations as corresponding claim 5 and is rejected for similar reasons as claim 5 using similar teachings and rationale. Claims 7 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL as applied respectively above to claims 1 and 8, and further in view of HE (“HE, " TransFG : A Transformer Architecture for Fine-Grained Recognition”), hereafter HE. Regarding Claim 7 , MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL teaches the elements of claim 1 as outlined above. LI further teaches: performing a binary classification task with the real non-defect and the synthetic defect images using a classification head to calculate a cross-entropy loss ; (LI [pg. 1, 1. Introduction] teaches: "Our innovation is at designing a novel proxy task for self-supervised learning of representations. Specifically, we formulate a proxy classification task between normal training data and the ones augmented by the CutPaste , the proposed data augmentation strategy that cuts an image patch and pastes at a random location of an image." LI [pg. 2, Figure 1] teaches training a CNN using normal data with no defects (i.e., real non-defect images) and CutPaste augmented data with defects (i.e., synthetic defect images). LI [pg. 3, 2.1. Self-Supervised Learning with CutPaste ] teaches calculating a cross-entropy loss using equation (1): where g is a binary classifier (i.e., for performing a binary classification task). LI [pg. 5, 4. Experiments] teaches a WideResNet (i.e., CNN ) plus an MLP projection head followed by the last linear layer (i.e., classification head ).) MEMO in view of LI, HUANG, KR 20210141060 A, and JEZEQUEL is not relied upon for teaching, but HE teaches: performing a contrastive learning task with the real non-defect and the synthetic defect images using a contrastive head to calculate a contrastive loss; and updating the weights of the CNN encoder and two heads to minimize a weighted sum of the cross-entropy loss and the contrastive loss . (HE [pg. 1, Abstract] teaches: "A contrastive loss is applied to enlarge the distance between feature representations of confusing classes." Furthermore, paragraph [0060] of the specification uses the exact same contrastive loss disclosed in equation 9) in HE [pg. 4, Contrastive Feature Learning]. HE [pg. 4, Contrastive Feature Learning] A simple cross-entropy loss is not enough to fully supervise the learning of features since the differences between subcategories might be small. HE [pg. 4, Contrastive Feature Learning] also teaches training the model with sum of the cross-entropy loss and contrastive loss, as shown in equation (10): Equation (10) includes a constant margin \alpha to