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 .
DETAILED ACTION
The United States Patent & Trademark Office appreciates the application that is submitted by the inventor/assignee. The United States Patent & Trademark Office reviewed the following application and has made the following comments below.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 09/20/2024 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
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(a) which forms the basis for all obviousness rejections set forth in this Office action:
(a) A patent may not be obtained though the invention is not identically disclosed or described as set forth in section 102 of this title, if the differences between the subject matter sought to be patented and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art to which said subject matter pertains. Patentability shall not be negatived by the manner in which the invention was made.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103(a) 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 1-2, 9-12, 14-17, and 19-20 are rejected under 35 U.S.C. 103(a) as being unpatentable over Parisotto et al. (NPL “MORE: Simultaneous Multi-View 3D Object Recognition and Pose Estimation,” 2023, hereafter referred to as Parisotto) in view of Yang et al. (NPL “Learning Relationships for Multi-View 3D Object Recognition,” 2019, hereafter referred to as Yang) in further view of Flagg et al. (U.S. Patent Pub. No. 2022/0327851 A1, hereafter referred to as Flagg).
Regarding Claim 1, Parisotto teaches a selective multi-view method for 3D object classification (1 Introduction, Parisotto teaches an object-agnostic entropy estimation model, called “MORE” that automatically predicts the most informative viewpoints of a given 3D object directly. The predicted views are then fed into a viewpoint network to predict the category label of the target object.), comprising: inputting, by a 3D imaging sensor, a 3D data representation of a 3D object (4.5 Real-robot demonstration, Fig. 12, Parisotto teaches using an Asus Xtion RGB-D sensor to capture a partial point cloud of a bottle object. An orthographic projection technique was used to generate 15 RGB-D views of the bottle from various perspectives. Alternatively, a completed point cloud may be generated by increasing the number of available cameras and using point cloud registration techniques.);
PNG
media_image1.png
746
1284
media_image1.png
Greyscale
extracting, by processing circuitry (4.3.3, Multi-view prediction, Parisotto teaches performing classification and prediction for the single-view method on an NVIDIA RTX2070.), multiple view images from the 3D data representation of the 3D object (Fig. 12, Parisotto teaches generating 15 views of a bottle object. Multiple RGB-D views of the object are generated from 15 different perspectives distributed over a hemisphere around the object.); selecting, by the processing circuitry (4.3.3, Multi-view prediction, Parisotto teaches performing classification and prediction for the single-view method on an NVIDIA RTX2070.), a most influential view based on an assignment of importance scores (3.1, Best-views prediction model, Fig. 3, Parisotto teaches evaluating the quantity of information in each view by calculating their entropy. Higher predicted entropy values tend to denote more informative views. Figure 3 shows the five best-views of two objects (airplane and flower pot) predicted by the proposed model. For the single view object recognition task, MORE predicts the best view first then recognizes the object. The Examiner interprets entropy values to be importance scores since higher entropy values correspond to more informative views (i.e., more “important” views.)
PNG
media_image2.png
618
1111
media_image2.png
Greyscale
predicting, by the processing circuitry (4.3.3, Multi-view prediction, Parisotto teaches performing classification and prediction for the single-view method on an NVIDIA RTX2070.), a classification of the 3D object based on the selected most influential view (4.3.1, Single view prediction, 3.1, Best-views prediction model, 3.2 Multi-view classification and pose estimation, Parisotto teaches the model, “MORE” predicts the best view first and then recognizes the object. The model takes a single view as an input and it predicts the class of the originating object. The Examiner interprets the “best view” to be the “most influential view” since the claim is silent to the definition of “most influential view.”); and outputting, by the processing circuitry (4.3.3, Multi-view prediction, Parisotto teaches performing classification and prediction for the single-view method on an NVIDIA RTX2070.), a class of the 3D object (5 Conclusions, Parisotto teaches using the obtained view/views of the object to predict the category label of the target object. Results showed that the predicted views of objects are descriptive enough to achieve high accuracy scores in classification.).
Parisotto does not explicitly disclose using a Cosine similarity method between visual features detected by at least one pre-trained convolutional neural network (CNN).
Yang is in the same field of art of multi-view 3D object detection. Further, Yang teaches using a (3.2, Reinforcing Block, Yang teaches computing dot-product similarity as the matching score between features.) by at least one pre-trained convolutional neural network (CNN) (3.4 Relation Network, 4.2, Implementation Details, 3.1 Overview, Yang teaches using the VGG-M network pre-trained on ImageNet as the base model. Features are extracted from the convolutional layer.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Parisotto by calculating the similarity between feature vectors of each view that is taught by Yang, to make the invention that determines the most discriminative views for 3D object classification; thus, one of ordinary skilled in the art would be motivated to combine the references since connecting corresponding regions from different views and reasoning about the relationships between them can help the views to better characterize the 3D object (Yang, 3.1 Overview).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Parisotto in view of Yang does not explicitly disclose using a Cosine similarity method between visual features.
Flagg is in the same field of art of identifying the classification of an object and identifying the presence of distinctive features. Further, Flagg teaches using a cosine similarity method between visual features detected (Paragraphs [0041-42], [0044], Flagg teaches the one or more feature vectors can be compared using cosine similarity. The one or more feature vectors can be used to assess the similarity of images. The Examiner interprets “views” to be images.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Parisotto in view of Yang by replacing the dot-product similarity calculation (Yang) with a cosine similarity method to determine the similarity between different feature vectors and thus different views that is taught by Flagg, to make the invention that determines the similarity between views to determine the most discriminative or unique views, and which contain redundant information; thus, one of ordinary skilled in the art would be motivated to combine the references since the learned weighting of the one or more feature vectors that contribute more strongly to the classification result can be weighed more highly. The learned weighting of one or more feature vectors can allow more discriminative views to contribute more to the final image analysis result (Flagg, Paragraph [0042]).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
In regards to Claim 2, Parisotto in view of Yang in further view of Flagg discloses the method of claim 1, further comprising controlling a robotic manipulator to grasp the 3D object based on the object class (4.5 Real-robot demonstration, Fig. 8, Fig. 11, Parisotto teaches for the “serve drink task,” the dual-arm robot first detects the pose and label of all objects, and then grasps and manipulates the juice bottle and the mug into the serve position.).
PNG
media_image3.png
654
1365
media_image3.png
Greyscale
PNG
media_image4.png
901
681
media_image4.png
Greyscale
In regards to Claim 9, Parisotto in view of Yang in further view of Flagg discloses the method of claim 1, wherein the extracting step extracts the multiple view images from a perspective of an arrangement of a plurality of virtual cameras arranged at positions in a circle around the 3D object (3.1 Best-views prediction model, 4.1 Dataset, Parisotto teaches the virtual cameras are positioned on 12 points on a section ring of the sphere, each one at an angle 30 degrees from the next one. 2D depth images are rendered from each virtual camera using a projection method.).
PNG
media_image5.png
671
657
media_image5.png
Greyscale
In regards to Claim 10, Parisotto in view of Yang in further view of Flagg discloses the method of claim 1, further comprising classifying, by a fully connected layer, the 3D object (3.3 Integrating Block, Yang teaches sending the feature vector f through the remaining fully-connected layers to obtain the final 3D object representation.).
In regards to Claim 11, Parisotto teaches a 3D object recognition subsystem for a robotic pick and place manipulator (4 Results, 4.5 Real-robot demonstration, Fig. 8, Parisotto teaches integrating MORE (Multi-View 3D Object Recognition and Pose Estimation) into a cognitive robotic system. The robotic setup consists of two UR5e arms, which are equipped with Robotiq 2F-140 gripper. The robot grasps and manipulates the objects after recognizing them in accordance with the instructed task.), comprising: a 3D imaging sensor obtaining a 3D data representation of a 3D object (4.5 Real-robot demonstration, Fig. 8, Parisotto teaches the robotic setup consists of an Asus Xtion RGB-D camera. Real point cloud data was captured.); processing circuitry (4.3.3. Multi-view prediction, Parisotto teaches using an NVIDIA RTX2070.) configured to extract multiple view images from the 3D data representation of the 3D object (Fig. 12, Parisotto teaches generating 15 views of a bottle object. Multiple RGB-D views of the object are generated from 15 different perspectives distributed over a hemisphere around the object.), select a most influential view based on an assignment of importance scores (3.1, Best-views prediction model, Fig. 3, Parisotto teaches evaluating the quantity of information in each view by calculating their entropy. Higher predicted entropy values tend to denote more informative views. Figure 3 shows the five best-views of two objects (airplane and flower pot) predicted by the proposed model. For the single view object recognition task, MORE predicts the best view first then recognizes the object. The Examiner interprets entropy values to be importance scores since higher entropy values correspond to more informative views (i.e., more “important” views.) predict a classification of the 3D object based on the selected most influential view (4.3.1, Single view prediction, 3.2 Multi-view classification and pose estimation, Parisotto teaches the model, “MORE” predicts the best view first and then recognizes the object. The model takes a single view as an input and it predicts the class of the originating object.), output a class of the 3D object (5 Conclusions, Parisotto teaches using the obtained views of the object to predict the category label of the target object. Results showed that the predicted views of objects are descriptive enough to achieve high accuracy scores in classification.), and control the robotic manipulator to grasp the 3D object based on the object class (Abstract, Fig. 1, Fig. 8, Parisotto teaches the robot may be instructed to perform the “serve drink task.” The robot first detects the label of all objects then grasps and manipulates the juice bottle and the mug into the serve position. Then the juice is poured into a cup by the robot. Finally, the robot hands the user the cup and places the juice bottle in the basket.).
Parisotto does not explicitly disclose using a Cosine similarity method between visual features detected by at least one pre-trained convolutional neural network (CNN).
Yang is in the same field of art of multi-view 3D object recognition. Further, Yang teaches using a (3.2, Reinforcing Block, Yang teaches computing dot-product similarity as the matching score between features) detected by at least one pre-trained convolutional neural network (CNN) (3.4 Relation Network, 4.2, Implementation Details, Yang teaches using the VGG-M network pre-trained on ImageNet as the base model.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Parisotto by calculating the similarity between feature vectors of each view that is taught by Yang, to make the invention that determines the most discriminative views for 3D object classification; thus, one of ordinary skilled in the art would be motivated to combine the references since connecting corresponding regions from different views and reasoning about the relationships between them can help the views to better characterize the 3D object (Yang, 3.1 Overview).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Parisotto in view of Yang does not explicitly disclose using a Cosine similarity method.
Flagg is in the same field of art of performing 3D object classification and determining distinctive features. Further, Flagg teaches using a Cosine similarity method (Paragraphs [0041-42], [0044], Flagg teaches the one or more feature vectors can be compared using cosine similarity. The one or more feature vectors can be used to assess the similarity of images.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Parisotto in view of Yang by replacing the dot-product similarity (Yang) with the Cosine similarity to calculate the importance score that is taught by Flagg, to make the invention that compares feature vectors using cosine similarity; thus, one of ordinary skilled in the art would be motivated to combine the references since features that contribute more strongly to a classification result can be weighed more heavily, therefore providing more discriminative views to contribute to the final image analysis results (Flagg, Paragraph [0042]).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
In regards to Claim 12, Parisotto in view of Yang in further view of Flagg discloses the subsystem of claim 11, wherein the processing circuitry (4.3.3, Multi-view prediction, Parisotto teaches performing classification and prediction for the single-view method on an NVIDIA RTX2070.) is further configured to extract the multiple view images from a perspective of an arrangement of a plurality of virtual cameras (3.1 Best-views prediction model, Fig. 4, Parisotto teaches taking images from 60 positions, regularly distributed on a sphere. The virtual cameras are positioned on 12 points on a section ring of the sphere, each one at an angle of 30 degrees from the next one. The sphere is circled by 5 rings, parallels to the horizontal axis of the object, which are looking at the center of the sphere from each at an angle of 30 degrees from the next one, cutting the sphere at 30, 60, 90, 120, and 150 degrees from the vertical axis of the object. The structure of the camera positions ensures a complete overview of any object can be obtained. Once the positions for the cameras are collected, grayscale depth images of the objects are taken for each of the 60 views.).
In regards to Claim 14, Parisotto in view of Yang in further view of Flagg discloses the subsystem of claim 11, wherein the processing circuitry (4.3.3, Multi-view prediction, Parisotto teaches performing classification and prediction for the single-view method on an NVIDIA RTX2070.) is further configured to compare feature vectors obtained by the feature extraction based on their similarity using the Cosine Similarity (Paragraph [0040], Flagg teaches one or more feature vectors can encapsulate common features from the classifications or distinctive features so the one or more feature vectors may be compared using cosine similarity.), and assign an importance score to each feature vector (Paragraph [0042], Flagg teaches assigning a weighting to each feature to allow the one or more feature vectors that contribute more strongly to a final answer to be weighted more highly. The learned weighting of the one or more feature vectors can allow more discriminative views to contribute more to the final image analysis results. The Examiner interprets “weighting” to be an assigned importance score since the claim is silent to the specific definition of “importance.”).
In regards to Claim 15, Parisotto in view of Yang in further view of Flagg discloses the subsystem of claim 14, wherein the processing circuitry (4.3.3, Multi-view prediction, Parisotto teaches performing classification and prediction for the single-view method on an NVIDIA RTX2070.) is further configured to determine the importance score I of each feature vector fv as a Cosine distance between current fvi and all other feature vectors (Paragraph [0044], Flagg teaches the one or more feature vectors can be used as a query vector, with all other one or more vectors in the feature vector database sorted by cosine similarity, Euclidean distance, or other similarity measure.).
In regards to Claim 16, Parisotto in view of Yang in further view of Flagg discloses the subsystem of claim 15, wherein the processing circuitry (4.3.3, Multi-view prediction, Parisotto teaches performing classification and prediction for the single-view method on an NVIDIA RTX2070.) is further configured to select the feature vector with highest importance score, Most Similar View (MSV), as the most influential view (Section 1 Introduction, 3.1 Overview, Yang teaches generating an importance score for each feature vector view. Then, each view is assigned an importance score, which denotes the relative discriminative power of that view. The Examiner interprets the most informative view to be the view with the greatest importance score since it is the view with the highest relative discriminative power.).
PNG
media_image6.png
397
912
media_image6.png
Greyscale
In regards to Claim 17, Parisotto in view of Yang in further view of Flagg discloses the subsystem of claim 15, wherein the processing circuitry (4.3.3, Multi-view prediction, Parisotto teaches performing classification and prediction for the single-view method on an NVIDIA RTX2070.) is further configured to select the feature vector with lowest importance score, Most Dissimilar View (MDV), as the most influential view (Section 1 Introduction, 3.1 Overview, Fig. 4, Yang teaches generating an importance score for each feature vector view. Then, each view is assigned an importance score, which denotes the relative discriminative power of that view. The Examiner interprets the view(s) having the lowest importance scores (0.00) to be the “Most Dissimilar View” since these views have the lowest similarity score compared to other views. See Fig 4, the view(s) having importance scores of 0.00.).
In regards to Claim 19, Parisotto in view of Yang in further view of Flagg discloses the subsystem of claim 12, wherein the arrangement of the plurality of virtual cameras is virtual cameras arranged at positions in a circle around the 3D object (3.1 Best-views prediction model, Fig. 4, Parisotto teaches the virtual cameras are positioned on 12 points on a section ring of the sphere, each one at an angle 30 degrees from the next one.).
PNG
media_image7.png
657
657
media_image7.png
Greyscale
In regards to Claim 20, Parisotto in view of Yang in further view of Flagg discloses the subsystem of claim 11, wherein the processing circuitry (4.3.3, Multi-view prediction, Parisotto teaches performing classification and prediction for the single-view method on an NVIDIA RTX2070.) is further configured to classify, by a fully connected layer, the 3D object (3.3 Integrating Block, Yang teaches sending the feature vector f through the remaining fully-connected layers to obtain the final 3D object representation.).
Claims 3-7 and 13 are rejected under 35 U.S.C. 103(a) as being unpatentable over Parisotto et al. (NPL “MORE: Simultaneous Multi-View 3D Object Recognition and Pose Estimation,” 2023, hereafter referred to as Parisotto) in view of Yang et al. (NPL “Learning Relationships for Multi-View 3D Object Recognition,” 2019, hereafter referred to as Yang) in further view of Flagg et al. (U.S. Patent Pub. No. 2022/0327851 A1, hereafter referred to as Flagg) in further view of Han et al. (U.S. Patent Pub No. 2025/0166395 A1).
Regarding Claim 3, Parisotto in view of Yang in further view of Flagg teaches The method of claim 1, further comprising feature extracting, by the at least one pre-trained CNN ψ (3.1 Best-views prediction model, Parisotto teaches a CNN for identifying high level features of different scales. Both MobileNextV2 and VGG-16 were evaluated.).
Parisotto in view of Yang in further view of Flagg does not explicitly disclose a stack of feature maps fmi of a detected visual feature from the extracted multiple view images Vi (before the predicting by a classification model.
Han is in the same field of art of 3D object detection based on multiple views corresponding to different viewpoints. Further, Han teaches a stack of feature maps fmi of a detected visual feature (Paragraph [0056], Han teaches the view encoders apply CNNs to the 2D inputs. The CNN may hierarchically extract features for each view. For a given view, the encoder CNN architecture filters and combines the 2D view into successively abstract feature maps. The Examiner interprets “successively abstract feature maps” to be a “stack” of feature maps since the claim is silent to the meaning of “stack”.) from the extracted multiple view images Vi (Paragraph [0056], Han teaches the view encoders analyze each 2D view independently.) before the predicting by a classification model (Paragraphs [0056], [0061], Fig. 3A, Han teaches the multiple encoded view representations become input to subsequent cross-attention interactions and/or fusion stages. The feature maps are extracted prior to generating the 3D object detections (324) by the detection head (322) as shown in Fig. 3A.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Parisotto in view of Yang in further view of Flagg by extracting feature maps from multiple views that is taught by Han, to make the invention that captures informative characteristics about the 3D representation of the object from each viewpoint; thus, one of ordinary skilled in the art would be motivated to combine the references since analyzing each view independently allows the model to extract salient features from each view and to further analyze and exchange information between the encoded features from multiple views output by the view encoders. Therefore, complementary information from the distinct views can be consolidated, providing a more comprehensive feature description compared to individual views (Han, Paragraphs [0056-57]).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
In regards to Claim 4, Parisotto in view of Yang in view of Flagg in further view of Han discloses the method of claim 3, further comprising comparing feature vectors obtained by the feature extracting based on their similarity using the Cosine Similarity (Paragraph [0040], Flagg teaches one or more feature vectors can encapsulate common features from the classifications or distinctive features so the one or more feature vectors may be compared using cosine similarity.), and assigning an importance score to each feature vector (Paragraph [0042], Flagg teaches assigning a weighting to each feature to allow the one or more feature vectors that contribute more strongly to a final answer to be weighted more highly. The learned weighting of the one or more feature vectors can allow more discriminative views to contribute more to the final image analysis results. The Examiner interprets a weight to be an importance score since the claim is silent to the definition of “importance score.”).
In regards to Claim 5, Parisotto in view of Yang in view of Flagg in further view of Han discloses the method of claim 4, further comprising determining the importance score I of each feature vector fv as a Cosine distance between current fvi and all other feature vectors (Paragraph [0044], Flagg teaches the one or more feature vectors can be used as a query vector, with all other one or more vectors in the feature vector database sorted by cosine similarity, Euclidean distance, or other similarity measure.).
In regards to Claim 6, Parisotto in view of Yang in view of Flagg in further view of Han discloses the method of claim 5, further comprising selecting the feature vector with highest importance score, Most Similar View (MSV), as the most influential view (Section 1 Introduction, 3.1 Overview, Fig. 4, Yang teaches generating an importance score for each feature vector view. Then, each view is assigned an importance score, which denotes the relative discriminative power of that view. The Examiner interprets the view with the highest importance score to by the “Most Similar View” since it is the most similar to the other views and therefore, received the highest importance score. See Fig. 4 below, the view having an importance score of 0.30.).
PNG
media_image8.png
288
772
media_image8.png
Greyscale
In regards to Claim 7, Parisotto in view of Yang in view of Flagg in further view of Han discloses the method of claim 5, further comprising selecting the feature vector with lowest importance score, Most Dissimilar View (MDV), as the most influential view (Section 1 Introduction, 3.1 Overview, Fig. 4, Yang teaches generating an importance score for each feature vector view. Then, each view is assigned an importance score, which denotes the relative discriminative power of that view. The Examiner interprets the view(s) having the lowest importance scores (0.00) to be the “Most Dissimilar View.” See Fig 4 above, the view(s) having importance scores of 0.00.).
In regards to Claim 13, Parisotto in view of Yang in view of Flagg discloses the subsystem of claim 11, wherein the processing circuitry (4.3.3, Multi-view prediction, Parisotto teaches performing classification and prediction for the single-view method on an NVIDIA RTX2070.) is further configured to extract by the at least one pre-trained CNN ψ (3.1 Best-views prediction model, Parisotto teaches a CNN for identifying high level features of different scales. Both MobileNextV2 and VGG-16 were evaluated.).
Parisotto in view of Yang in further view of Flagg does not explicitly disclose a stack of feature maps fmi of the detected visual feature from the extracted multiple view images Vi before the predicting by a classification model.
Han is in the same field of art of 3D object detection based on multiple views corresponding to different viewpoints. Further, Han teaches a stack of feature maps fmi of the detected visual feature (Paragraph [0056], Han teaches the view encoders apply CNNs to the 2D inputs. The CNN may hierarchically extract features for each view. For a given view, the encoder CNN architecture filters and combines the 2D view into successively abstract feature maps. The Examiner interprets “successively abstract feature maps” to be a “stack” of feature maps since the claim is silent to the definition of “stack.”) from the extracted multiple view images Vi (Paragraph [0056], Han teaches the view encoders analyze each 2D view independently.) before the predicting by a classification model (Paragraphs [0056], [0061], Fig. 3A, Han teaches the multiple encoded view representations become input to subsequent cross-attention interactions and/or fusion stages. The feature maps are extracted prior to generating the 3D object detections (324) by the detection head (322) as shown in Fig. 3A.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Parisotto in view of Yang in further view of Flagg by extracting feature maps from multiple views that is taught by Han, to make the invention that captures informative characteristics about the 3D representation of the object from each viewpoint; thus, one of ordinary skilled in the art would be motivated to combine the references since analyzing each view independently allows the model to extract salient features from each view and to further analyze and exchange information between the encoded features from multiple views output by the view encoders. Therefore, complementary information from the distinct views can be consolidated, providing a more comprehensive feature description compared to individual views (Han, Paragraphs [0056-57]).
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claims 8 and 18 are rejected under 35 U.S.C. 103(a) as being unpatentable over Parisotto et al. (NPL “MORE: Simultaneous Multi-View 3D Object Recognition and Pose Estimation,” 2023, hereafter referred to as Parisotto) in view of Yang et al. (NPL “Learning Relationships for Multi-View 3D Object Recognition,” 2019, hereafter referred to as Yang) in view of Flagg et al. (U.S. Patent Pub. No. 2022/0327851 A1, hereafter referred to as Flagg) in further view of Wei et al. (NPL “View-GCN: View-based Graph Convolutional Neural Network for 3D Shape Analysis,” 2020, hereafter referred to as Wei).
In regards to Claim 8, Parisotto in view of Yang in further view of Flagg disclose the method of claim 1.
Parisotto in view of Yang in further view of Flagg does not explicitly disclose wherein the extracting step extracts the multiple view images from a perspective of an arrangement of a plurality of virtual cameras arranged at positions irregularly spherical around the 3D object.
Wei is in the same field of art of selective multi-view-based 3D object recognition. Further, Wei teaches wherein the extracting step extracts the multiple view images from a perspective of an arrangement of a plurality of virtual cameras arranged at positions irregularly spherical around the 3D object (Section 1 Introduction, Section 5 View-Based Graph Convolutional Network, Fig. 4, Wei teaches a Graph Convolutional Network (GCN) which can flexibly model different view configurations, including circular, dodecahedral, and irregular configurations. Fig. 4(b) shows a dodecahedral configuration. Fig. 4(c) shows an irregular configuration.).
PNG
media_image9.png
213
645
media_image9.png
Greyscale
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Parisotto in view of Yang in further view of Flagg by flexibly modeling different view configurations, such as the dodecahedron or irregular configurations that is taught by Wei, to make the invention that flexibly models different view configurations for 3D object classification; thus, one of ordinary skilled in the art would be motivated to combine the references since it enables flexible modeling of different view configurations (Wei, 1 Introduction), which improves robustness of the model. Additionally, as shown in Fig. 4, the dodecahedral view configuration provides additional coverage compared to the circular configuration.
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
In regards to Claim 18, Parisotto in view of Yang in further view of Flagg disclose the subsystem of claim 12.
Parisotto in view of Yang in further view of Flagg does not explicitly disclose wherein the arrangement of the plurality of virtual cameras is virtual cameras arranged at positions irregularly spherical around the 3D object.
Wei is in the same field of art of selective multi-view-based 3D object recognition. Further, Wei teaches wherein the arrangement of the plurality of virtual cameras is virtual cameras arranged at positions irregularly spherical around the 3D object (Section 1 Introduction, Section 5 View-Based Graph Convolutional Network, Fig. 4, Wei teaches a Graph Convolutional Network (GCN) which can flexibly model different view configurations, including circular, dodecahedral, and irregular configurations.).
Therefore, it would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Parisotto in view of Yang in further view of Flagg by flexibly modeling different view configurations, such as the dodecahedron or irregular configurations that is taught by Wei, to make the invention that flexibly models different view configurations for 3D object classification; thus, one of ordinary skilled in the art would be motivated to combine the references since it enables flexible modeling of different view configurations (Wei, 1 Introduction), which improves robustness of the model. Additionally, as shown in Fig. 4, the dodecahedral view configuration provides additional coverage compared to the circular configuration.
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Pertinent Prior Art
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Kasaei et al. (NPL “Simultaneous Multi-View Object Recognition and Grasping in Open-Ended Domains,” April 2024) teaches a multi-view object recognition/grasping approach in which the view with the highest entropy is identified as the “best view” for grasping an object.
Savvides et al. (U.S. Patent Pub. No. 2025/0182444) teaches a system and method for multi-view product detection and recognition. Feature vectors are extracted from multiple views of the objects and a matching process is performed which calculates a cosine similarity between the feature vectors extracted from the images and feature vectors of objects in a database. The matched object for each feature vector may be the object having the highest cosine similarity between the extracted feature vector and the feature vectors of the enrolled objects.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SYDNEY L BLACKSTEN whose telephone number is (571)272-7120. The examiner can normally be reached 8:30am-4:30pm.
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, Oneal Mistry can be reached at 313-446-4912. 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.
/SYDNEY L BLACKSTEN/Examiner, Art Unit 2674
/ONEAL R MISTRY/Supervisory Patent Examiner, Art Unit 2674
Prosecution Insights