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 .
Priority
Acknowledgment is made of applicant’s claim for foreign priority under 35 U.S.C. 119 (a)-(d). The certified copy has been filed in parent Application No. 18884397, filed on 09/13/2024.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 09/13/2024 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 § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 6-10 and 16-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding claims 6, 7, 10, 16, 17, and 20: The terms “relatively higher” and “relatively smaller”, are relative terms which render the claims indefinite. The terms are not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. Thus, “relatively higher” and “relatively lower” is being interpreted as relating to higher or lower, respectively, by any amount.
Claims 6 (and similarly 16) recite, “wherein the weight assigning module is further configured to assign a relatively higher weight to a given cluster that is determined to have a relatively higher diversity.” The “diversity” of the cluster is unclear. There are different types of diversity that can apply here (different backgrounds, foregrounds, subjects, time of day, number of different cameras used, etc.). Accordingly, the nature of the diversity requires clarification and is being interpreted as any kind of diversity.
Claims 7 (and similarly 17) recite, “wherein the secondary training module is further configured to progressively perform training by increasing a reflection ratio of a given cluster to which a relatively higher weight is assigned.” The claimed “reflection ratio” is not described in the Specification, nor does it appear to be a standard term in the art of training. Accordingly, the reflection ratio is being interpreted as a general degree of importance/emphasis given to a cluster, which is inherent to having a higher weight.
Claims 8 (and similarly 18) recite, “wherein the instructions further enable the one or more processors to have a repetition module configured to repeat the target subset generation module, the feature vector extraction module, the feature vector clustering module, and the labeling module until set conditions are satisfied.” No condition is specified in the claim nor in the Specification, rendering the limitation indefinite. The set condition is being interpreted as any condition.
Claims 9 and 19 are rejected as dependent on claims 8 and 18.
Claims 10 (and similarly 20) recite, “wherein the target subset generation module is further configured to generate the target subset by preferentially selecting a given camera producing values with a relatively smaller difference from the source dataset from among the plurality of cameras.” Firstly, the link between target subset generation and preferential camera selection is unclear. Also, the degree of preferential selection is indefinite (only using the given camera, using the given camera 25% more than other cameras, using the given camera 30% more than other cameras, etc.) Next, the claimed “values” are indefinite and amount to any value of any type, and are being interpreted accordingly. See below for comments on the relative term “relatively smaller”. Additionally, the “difference” is indefinite as it is unclear what the values themselves are, which appear to be used for some difference metric/calculation. The claim also appears to compare values of a camera to values of the source dataset (images), which further renders the claim as indefinite. Overall the claim is generally being interpreted as using images from one camera more than others based on some similarity that relates to a source dataset.
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.
Claim(s) 1-4, 8-14, and 18-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Takeuchi (Unsupervised domain-adaptive person re-identification with multi-camera constraints) in view of Liu (Scene Recognition Mechanism for Service Robot Adapting Various Families: A CNN-Based Approach Using Multi-Type Cameras).
Regarding claim 1, Takeuchi teaches “A device for training a model for human identification, the device comprising: one or more processors; and a storage medium storing computer-readable instructions that, when executed by the one or more processors, enable the one or more processors to provide: a primary training module configured to primarily train the model with respect to a pre-prepared source dataset,” (Takeuchi, figure 2 and Section 2.1, “Figure 2 shows an overview of the proposed ECA-Net method, which is based on the mean teacher method, where the mean teacher model uses a temporally moving average of weights of the student network [4]. The frame work uses two datasets: (1) source domain data, Ds = {(xsi, ysi) | i = 1,...,Ns}; (2) target domain data, Dt = {xt i | i = 1,..., Nt}. Here, x is the person image, y is the ID label, and N is the number of samples. First, the parameters of a convolutional neural network (CNN) are trained using Ds. Next, using the obtained pre-trained model, the CNN parameters are optimized for Dt through a self-training scheme. The pseudo-labels are obtained by clustering the distance matrix data of each person. We adopt ResNet-50 as the backbone of the CNN. To improve the performance, the following techniques are used: nonlocal block [5], generalized-mean pooling [6], and batch normalization head [7]. In addition, the ID classification loss, triplet loss, and contrastive loss are adopted as the loss functions [1].” Note that, as one skilled in the art would readily understand, the structural components claimed above are required for the invention of Takeuchi. Further note that the modules recited here and throughout the claim set are being interpreted as sections of the instructions (code), which is inherent to the instructions (code) performing discrete steps.)
While Takeuchi teaches a module for generating a target subset (“Target domain data” of Figure 2) by selecting cameras from a plurality of cameras (Takeuchi, Section 3.1, Paragraph 3, “Originally, the public datasets are not employed to evaluate the performance using the overlap information. For this, we additionally evaluated the performance using a private dataset (Shopping mall). It contains a video of pedestrians taken by three surveillance cameras installed inside a shopping mall in China. The person’s IDs were manually an notated, and the dataset contains 37,971 images, where 1,466 identities were used for training and 1,370 for testing. A part of each camera view overlaps, and the ratio of the number of pair images and the amount of training data is 0.13.” Note that even if the shopping mall only had three total cameras, selecting all three amounts to selecting some cameras out of a plurality of cameras.), Takeuchi does not expressly disclose cameras mounted on a service robot.
Liu discloses multiple cameras mounted on a service robot (Liu, Introduction, Paragraph 7, “To overcome the insufficient image description of a scene, multi-type cameras including fish-eye, pinhole, and depth camera are applied in this work. The fish-eye camera can offer the robot a broad view to catch more visual information of the scene (shown in Fig. 1 (b) and (c)). Hence, we have designed a model based on CNN features to recognize fish-eye scene images since the CNN features from network layers are helpful for scene recognition. Several scene recognition methods based on CNN features have been proposed without considering the choice of CNN feature extraction layers [16]–[18]. To address this issue, a selective CNN features fusion is presented for the fish-eye scene recognition model, which aims to improve the training efficiency and recognition performance. When the robot is in the right position where it can obtain a holistic horizon of the scene using a pinhole camera, the fish-eye camera will not be used since the fish-eye image may contain a redundant image description (e.g. more than one scene in the image). In this case, the images taken by the pinhole camera are similar to the images in the scene datasets. The model for the pinhole camera is built by transfer learning using a 50-layer ResNet [10] trained on the scene datasets. The depth camera is set in between the fish-eye camera and the pinhole camera, since the depth image contains range information, which shows a reflection of the position information that can help with the sufficiency determination of image description of the scene. This process can be viewed as a dichotomy problem, which is much simpler than scene recognition. Therefore, a network with fewer layers, 18-layer ResNet, is utilized to construct the model for the depth camera.”)
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to mount the plurality of cameras of Takeuchi on the service robot of Liu.
The motivation for doing so would have been to improve functionality of service robots for home security applications. Indeed, Takeuchi already describes applications in surveillance including crime detection (Takeuchi, Introduction, Paragraph 1, “Person re-identification (ReID) aims to retrieve the same per son (e.g., pedestrian) from different images. It has attracted increasing attention in several industrial fields related to intelligent surveillance. For example, in a retail situation, it is used as a base technology for analyzing consumers’ in-store behavior, such as buying and shoplifting behaviors.) Extending the surveillance from retail to home security applications using the service robot would benefit family safety. Further, one skilled in the art could have combined the elements as described above by known methods with no change in their respective functions, and the combination would have yielded nothing more than predictable results. Therefore, it would have been obvious to combine Takeuchi with the above teaching of Liu to fully disclose, "a target subset generation module configured to generate a target subset by selecting some cameras from among a plurality of cameras mounted on a service robot,”
Takeuchi in view of Liu further disclose, “a feature vector extraction module configured to extract feature vectors of the target subset by using the model, a labeling module configured to perform labeling on the feature vectors, and a secondary training module configured to secondarily train the model with respect to a target dataset by using results of the labeling.” (Takeuchi, Figure 2 and Sections 2.1-2.2, “Figure 2 shows an overview of the proposed ECA-Net method, which is based on the mean teacher method, where the mean teacher model uses a temporally moving average of weights of the student network [4]. The frame work uses two datasets: (1) source domain data, Ds = {(xsi, ysi) | i = 1,...,Ns}; (2) target domain data, Dt = {xt i | i = 1,..., Nt}. Here, x is the person image, y is the ID label, and N is the number of samples. First, the parameters of a convolutional neural network (CNN) are trained using Ds. Next, using the obtained pre-trained model, the CNN parameters are optimized for Dt through a self-training scheme. The pseudo-labels are obtained by clustering the distance matrix data of each person. We adopt ResNet-50 as the backbone of the CNN. To improve the performance, the following techniques are used: nonlocal block [5], generalized-mean pooling [6], and batch normalization head [7]. In addition, the ID classification loss, triplet loss, and contrastive loss are adopted as the loss functions [1].”; “On the UDA ReID framework, the distance between each person is calculated based on the CNN features f for each sample. To calculate the distance matrix, we apply the widely used k-reciprocal re-ranking [8]. The k-reciprocal re-ranking uses the Jaccard distance, which is estimated based on the combination of neighboring images. It is given by
PNG
media_image1.png
194
602
media_image1.png
Greyscale
d is the cosine distance between two CNN features, and R(i,k) is the k-reciprocal nearest neighbors of a sample i. The pseudo-labels are extracted by performing clustering on the obtained distance matrix data, ˜yt = Fclust(dJ). In the proposed method, the same person-pair list in the target domain is used to improve the performance for the target domain,
PNG
media_image2.png
59
608
media_image2.png
Greyscale
Note that Pt does not have the ID labels yt. The list denotes a set of person-pair images in each video frame and the identities between different pairs are not given. The ID labels are essential for training a high-performance ReID model [1]. Thus, an approach that incorporates this incomplete data without the ID labels into the model is required. We propose an approach to refine the distance matrix using the same person-pair list Pt. As shown in Equations (1) and (2), the input variable for distance calculation is the CNN feature f. The closer the cosine distance of the CNN feature pairs, the closer is the Jaccard distance. Using this property, we impose the following constraint,
PNG
media_image3.png
39
587
media_image3.png
Greyscale
on the two-person pairs corresponding to Pt. In conventional UDA ReID methods, when the image characteristics are different even for the same person, the Jaccard distance becomes large, and the same pseudo-label is not obtained. In contrast, the proposed method can be guided to appropriate pseudo labels by modifying the values of the CNN features based on the same person-pair list obtained from the multi-camera environment. Thus, based on the training data with the refined pseudo-labels, Dt = {(xt i, ˜yt i) |i = 1,...,Nt}, the model adapted to the target domain is trained.” Note that “CNN features f for each sample” extracted by the model of Takeuchi are mapped to the claimed feature vectors extracted by the model; “pseudo-labels are extracted by performing clustering on the obtained distance matrix” maps to the claimed labeling based on the feature vectors, as the pseudo labels are obtained in accordance with the distance matrix, which itself is based on the feature vectors; using training data with refined pseudo-labels, Dt, amounts to the “secondarily” training step using the labeling results, also described by “CNN parameters are optimized for Dt through a self-training scheme”.)
Regarding claim 2, Takeuchi in view of Liu disclose, “The device of claim 1,”
“wherein the instructions further enable the one or more processors to have a feature vector clustering module configured to determine similarities of the feature vectors and cluster the feature vectors according to the similarities.” (Takeuchi, Figure 2 and Sections 2.2, “On the UDA ReID framework, the distance between each person is calculated based on the CNN features f for each sample. To calculate the distance matrix, we apply the widely used k-reciprocal re-ranking [8]. The k-reciprocal re-ranking uses the Jaccard distance, which is estimated based on the combination of neighboring images. It is given by
PNG
media_image1.png
194
602
media_image1.png
Greyscale
d is the cosine distance between two CNN features, and R(i,k) is the k-reciprocal nearest neighbors of a sample i. The pseudo-labels are extracted by performing clustering on the obtained distance matrix data, ˜yt = Fclust(dJ). In the proposed method, the same person-pair list in the target domain is used to improve the performance for the target domain,
PNG
media_image2.png
59
608
media_image2.png
Greyscale
Note that Pt does not have the ID labels yt. The list denotes a set of person-pair images in each video frame and the identities between different pairs are not given. The ID labels are essential for training a high-performance ReID model [1]. Thus, an approach that incorporates this incomplete data without the ID labels into the model is required. We propose an approach to refine the distance matrix using the same person-pair list Pt. As shown in Equations (1) and (2), the input variable for distance calculation is the CNN feature f. The closer the cosine distance of the CNN feature pairs, the closer is the Jaccard distance. Using this property, we impose the following constraint,
PNG
media_image3.png
39
587
media_image3.png
Greyscale
on the two-person pairs corresponding to Pt. In conventional UDA ReID methods, when the image characteristics are different even for the same person, the Jaccard distance becomes large, and the same pseudo-label is not obtained. In contrast, the proposed method can be guided to appropriate pseudo labels by modifying the values of the CNN features based on the same person-pair list obtained from the multi-camera environment. Thus, based on the training data with the refined pseudo-labels, Dt = {(xt i, ˜yt i) |i = 1,...,Nt}, the model adapted to the target domain is trained.”)
Regarding claim 3, Takeuchi in view of Liu disclose, “The device of claim 2,”
“wherein the feature vector clustering module is further configured to cluster the feature vectors by using a Density-Based Spatial Clustering of Applications with Noise (DBSCAN) technique. (Takeuchi, Section 3.2., “For data preprocessing, we performed data augmentation via random erasing, random cropping, and flipping. Then, Adam optimizer was applied [15]. The warm-up learning rate was used to prevent the overfitting of the CNN parameters [16]. DBSCAN was used as the clustering algorithm for estimating pseudo-labels [17]. It does not require the specification of the number of clusters in advance and is widely used in the conventional UDA ReID methods.”)
Regarding claim 4, Takeuchi in view of Liu disclose, “The device of claim 2,”
“wherein the labeling module is further configured to perform pseudo-labeling for each cluster clustered by the feature vector clustering module.” (Takeuchi, Figure 2 and Sections 2.2, “On the UDA ReID framework, the distance between each person is calculated based on the CNN features f for each sample. To calculate the distance matrix, we apply the widely used k-reciprocal re-ranking [8]. The k-reciprocal re-ranking uses the Jaccard distance, which is estimated based on the combination of neighboring images. It is given by
PNG
media_image1.png
194
602
media_image1.png
Greyscale
d is the cosine distance between two CNN features, and R(i,k) is the k-reciprocal nearest neighbors of a sample i. The pseudo-labels are extracted by performing clustering on the obtained distance matrix data, ˜yt = Fclust(dJ). In the proposed method, the same person-pair list in the target domain is used to improve the performance for the target domain,
PNG
media_image2.png
59
608
media_image2.png
Greyscale
Note that Pt does not have the ID labels yt. The list denotes a set of person-pair images in each video frame and the identities between different pairs are not given. The ID labels are essential for training a high-performance ReID model [1]. Thus, an approach that incorporates this incomplete data without the ID labels into the model is required. We propose an approach to refine the distance matrix using the same person-pair list Pt. As shown in Equations (1) and (2), the input variable for distance calculation is the CNN feature f. The closer the cosine distance of the CNN feature pairs, the closer is the Jaccard distance. Using this property, we impose the following constraint,
PNG
media_image3.png
39
587
media_image3.png
Greyscale
on the two-person pairs corresponding to Pt. In conventional UDA ReID methods, when the image characteristics are different even for the same person, the Jaccard distance becomes large, and the same pseudo-label is not obtained. In contrast, the proposed method can be guided to appropriate pseudo labels by modifying the values of the CNN features based on the same person-pair list obtained from the multi-camera environment. Thus, based on the training data with the refined pseudo-labels, Dt = {(xt i, ˜yt i) |i = 1,...,Nt}, the model adapted to the target domain is trained.”)
Regarding claim 8, Takeuchi in view of Liu disclose, “The device of claim 2,”
“wherein the instructions further enable the one or more processors to have a repetition module configured to repeat the target subset generation module, the feature vector extraction module, the feature vector clustering module, and the labeling module until set conditions are satisfied.” (Takeuchi, Section 2.1, last Paragraph, “We adopt ResNet-50 as the backbone of the CNN. To improve the performance, the following techniques are used: nonlocal block [5], generalized-mean pooling [6], and batch normalization head [7]. In addition, the ID classification loss, triplet loss, and contrastive loss are adopted as the loss functions [1].” As one skilled in the art would understand, minimizing a loss function amounts to iterating until a set condition relating to the loss is met. Accordingly, all steps of the invention of Takeuchi in view of Liu, including the modules are repeated until a criteria is met.)
Regarding claim 9, Takeuchi in view of Liu disclose, “The device of claim 8,”
“wherein the instructions further enable the one or more processors to have a curriculum sequence generation module configured to generate a curriculum sequence for training the model by using results of labeling repeatedly generated by the repetition module, and wherein the secondary training module is further configured to perform curriculum learning on the model according to the curriculum sequence.” (A curriculum sequence is interpreted as any training data configuration. As described in the rejection of claim 8, repetition (iteration) is performed in the invention of Takeuchi in view of Liu. This results in the repeated generation of labels in each iteration, which are subsequently used as a “curriculum” for use by the secondary training model in each iteration.)
Regarding claim 10, Takeuchi in view of Liu disclose, “The device of claim 1,”
“wherein the target subset generation module is further configured to generate the target subset by preferentially selecting a given camera producing values with a relatively smaller difference from the source dataset from among the plurality of cameras.” (Liu, Introduction, Paragraph 7, “To overcome the insufficient image description of a scene, multi-type cameras including fish-eye, pinhole, and depth camera are applied in this work. The fish-eye camera can offer the robot a broad view to catch more visual information of the scene (shown in Fig. 1 (b) and (c)). Hence, we have designed a model based on CNN features to recognize fish-eye scene images since the CNN features from network layers are helpful for scene recognition. Several scene recognition methods based on CNN features have been proposed without considering the choice of CNN feature extraction layers [16]–[18]. To address this issue, a selective CNN features fusion is presented for the fish-eye scene recognition model, which aims to improve the training efficiency and recognition performance. When the robot is in the right position where it can obtain a holistic horizon of the scene using a pinhole camera, the fish-eye camera will not be used since the fish-eye image may contain a redundant image description (e.g. more than one scene in the image). In this case, the images taken by the pinhole camera are similar to the images in the scene datasets. The model for the pinhole camera is built by transfer learning using a 50-layer ResNet [10] trained on the scene datasets. The depth camera is set in between the fish-eye camera and the pinhole camera, since the depth image contains range information, which shows a reflection of the position information that can help with the sufficiency determination of image description of the scene. This process can be viewed as a dichotomy problem, which is much simpler than scene recognition. Therefore, a network with fewer layers, 18-layer ResNet, is utilized to construct the model for the depth camera.” Accordingly, in the multi-camera service robot of Takeuchi in view of Liu, cameras are preferentially selected for target image collection according to similarity to the source data. Note that the above passage was incorporated with rationale and motivation in the rejection of claim 1.)
Regarding claims 11-14 and 18-20, these claims recite a method with steps corresponding to the elements of the system recited in Claims 1-4 and 8-10. Therefore, the recited steps of this claim are mapped to the analogous elements in the corresponding system claim. Additionally, the rationale and motivation to combine the references apply here.
Claim(s) 5-7 and 15-17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Takeuchi in view of Liu further in view of Xie (Sampling and Re-Weighting: Towards Diverse Frame Aware Unsupervised Video Person Re-Identification).
Regarding claim 5, Takeuchi in view of Liu disclose, “The device of claim 2,”
Takeuchi in view of Liu do not expressly disclose “wherein the instructions further enable the one or more processors to have a weight assigning module configured to assign a weight for each cluster clustered by the feature vector clustering module.”
Xie teaches a diversity-based weighing for applications in clustering and Re-ID, (Xie, Introduction, Paragraph 4, “To address the above problems, this paper proposes a SRC method to optimize the distribution of feature space and improve clustering accuracy for unsupervised video person re-ID. To cope with the bias of tracklet features caused by noisy frames, a dynamic noise trimming module is presented for sampling to improve the accuracy of tracklet similarity. This module uses the diversity between frame features to identify and trim noisy frames dynamically. Then, a stable tracklet structure is obtained to enable an accurate representation of tracklet features. Aiming to learn the rich information from hard frames, a diverse frame re-weighting module is designed for re-weighting to obtain a discriminative tracklet feature space. By calculating the amount of information at each frame, this module explores the dispersion of tracklets to mine the hard frames. Then, the rich information of tracklets is learned from both the input data and feature embedding levels. Our contributions in this paper are summarized in two folds: We propose a clustering method with sampling and re-weighting strategies for unsupervised video person re-ID. It is built on the application-specific characteristics existing in the video person re-ID task. We design a dynamic noise trimming module to avoid feature bias caused by noisy frames and a diverse frame re-weighting module to enhance the training of hard frames.”)
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to use the weighing module taught by Xie to assign weights to the clusters of Takeuchi in view of Liu.
The motivation for doing so would have been, as described above by Xie, to emphasize richer information for better training. Further, one skilled in the art could have combined the elements as described above by known methods with no change in their respective functions, and the combination would have yielded nothing more than predictable results. Therefore, it would have been obvious to combine Takeuchi in view of Liu with the above teaching of Xie to fully disclose, “wherein the instructions further enable the one or more processors to have a weight assigning module configured to assign a weight for each cluster clustered by the feature vector clustering module.”
Regarding claim 6, Takeuchi in view of Liu further in view of Xie disclose, “The device of claim 5,”
“wherein the weight assigning module is further configured to assign a relatively higher weight to a given cluster that is determined to have a relatively higher diversity.” (Xie, Introduction, Paragraph 4, “To address the above problems, this paper proposes a SRC method to optimize the distribution of feature space and improve clustering accuracy for unsupervised video person re-ID. To cope with the bias of tracklet features caused by noisy frames, a dynamic noise trimming module is presented for sampling to improve the accuracy of tracklet similarity. This module uses the diversity between frame features to identify and trim noisy frames dynamically. Then, a stable tracklet structure is obtained to enable an accurate representation of tracklet features. Aiming to learn the rich information from hard frames, a diverse frame re-weighting module is designed for re-weighting to obtain a discriminative tracklet feature space. By calculating the amount of information at each frame, this module explores the dispersion of tracklets to mine the hard frames. Then, the rich information of tracklets is learned from both the input data and feature embedding levels. Our contributions in this paper are summarized in two folds: We propose a clustering method with sampling and re-weighting strategies for unsupervised video person re-ID. It is built on the application-specific characteristics existing in the video person re-ID task. We design a dynamic noise trimming module to avoid feature bias caused by noisy frames and a diverse frame re-weighting module to enhance the training of hard frames.” Note that this was incorporated with motivation and rationale in the rejection of claim 5.)
Regarding claims 15-17, these claims recite a method with steps corresponding to the elements of the system recited in Claims 5-7. Therefore, the recited steps of this claim are mapped to the analogous elements in the corresponding system claim. Additionally, the rationale and motivation to combine the references apply here.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to AARON JOSEPH SORRIN whose telephone number is (703)756-1565. The examiner can normally be reached Monday - Friday 9am - 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, Sumati Lefkowitz can be reached at (571) 272-3638. 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.
/AARON JOSEPH SORRIN/
Examiner, Art Unit 2672
/SUMATI LEFKOWITZ/Supervisory Patent Examiner, Art Unit 2672
Prosecution Insights