DETAILED ACTION
Remarks
Claims 1-7 and 11-14 have been examined and rejected. This Office Action is responsive to the amendment filed on 12/19/2025, which has been entered in the above identified application.
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 .
Claims 1-7 and 11-14 are presented for examination.
Respond to Amendment
The amendment filed 12/19/2025 has been entered. Claims 1-7, 13 and 14 have been amended. Claims 11 and 12 were previously presented. Claims 8-10 was previously canceled. Claims 1-7 and 11-14 are pending in the application.
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, 3, 4, 5, 13 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Kim et al (US 10095977 B1) hereafter Kim, in view of Kakishita et al (US 20220172341 A1) hereafter Kakishita, and further in view of Yang et al (US 20210019602 A1) hereafter Yang.
With respect to claim 1, Kim teaches a data processing apparatus comprising at least one processor; and at least one memory, including instructions stored thereon (the apparatus used in the invention has a processor (and/or a memory) to hold the data for performing convolutional operations [col. 11, lines 60-65]), which, when executed by the at least one processor, cause the apparatus to:
hold, in a first holding unit, filter coefficients to be used for convolution (the pixels of the training images or the feature maps derived from the test image configured to be the coefficients of the filter. The Field of View (FOV) configured to be an element to hold the pixels to generate a new pixel region for an outputted feature map [col. 8, lines 25-65]),
generate a dilated filter (In the dilation operation, a convolution filter is dilated or extended. A part of weights is filled with zeros except for the weights obtained from the original convolution filter [col. 8, lines 25-45 and FIGS. 3A-C]).
However, Kim does not disclose generate, in a case where a signal is received, a dilated filter obtained by adding a number of new filter coefficients being based on a predetermined dilation rate between adjacent filter coefficients read from the first holding unit, wherein the filter coefficients are read from the first holding unit prior to generating the dilated filter, and wherein data amount of the filter coefficients read from the first holding unit is smaller than data amount of the dilated filter obtained by adding the new filter coefficient to the filter coefficients read from the first holding unit; and perform the convolution by using filter coefficients of the dilated filter and image feature data, wherein, in the convolution, a plurality of multipliers are used to multiply one common filter coefficient of the generated dilated filter with each of a plurality of pieces of image feature data.
In the same field of endeavor, Kakishita teaches generate, in a case where a signal is received, a dilated filter obtained by adding a number of new filter coefficients being based on a predetermined dilation rate between adjacent filter coefficients read from the first holding unit (the kernel used for convolution processing and pooling processing of the fully convolutional networks (FCN) is changed by the dilate processing. The dilate processing is processing of inserting a predetermined number of zeros between elements of a kernel and expanding the kernel [par. 0091-0100]),
wherein the filter coefficients are read from the first holding unit prior to generating the dilated filter, and wherein data amount of the filter coefficients read from the first holding unit is smaller than data amount of the dilated filter obtained by adding the new filter coefficient to the filter coefficients read from the first holding unit (a DC kernel 1102 is an example of a kernel used for the dilated convolution processing and is constructed by inserting one zero into adjacent elements of the C kernel which is an example of a kernel used for the convolution processing. The C kernel is basically used in a convolution processing and a pooling processing, and the DC kernel is then used by inserting a predetermined number of zeros between elements of a C kernel [par. 0009, 0091-0100]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated the concept of determining a state of a container or a collection target based on an image acquired by a single camera as suggested by Kakishita into the concept of generating dilated filter and convolution calculation using the filter as suggested by Kim because both of these systems addressing the process of applying convolution operations on the filters to generate optimal results. Doing so would be desirable because the system of Kim would be more efficient by dividing an image into a plurality of regions by creating a dilate kernel by inserting a predetermined number of zeros between elements of a kernel used in a convolution processing and a pooling processing, and executing the convolution processing and the pooling processing by using the dilate kernel (Kakishita, [par. 0006-0010]).
However, the combination of Kim and Kakishita does not explicitly teach perform the convolution by using filter coefficients of the dilated filter and image feature data, wherein, in the convolution, a plurality of multipliers are used to multiply one common filter coefficient of the generated dilated filter with each of a plurality of pieces of image feature data.
In the same field of endeavor, Yang teaches perform the convolution by using filter coefficients of the dilated filter and image feature data (examples of data patterns of imagery data and filter coefficients for performing convolution operations over multiple layers using multiple CNN processing engines are described in FIGS 17A-C. [par. 0023, 0024]), wherein, in the convolution, a plurality of multipliers are used to multiply one common filter coefficient of the generated dilated filter with each of a plurality of pieces of image feature data (a CNN include multiple convolutional layers, each of which may include multiple filter coefficients, such as weights and other parameters. A multiplier is simply a hardware unit that multiplies a filter coefficient (weight) by a data value pixel. In an example of an AI chip, the AI tasks includes performing multiple layers of convolutions, wherein performing convolutions may require imagery data in digital form and corresponding filter coefficients [par. 0039-0041, 0044]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated the concept of CNN processing engine includes multiple convolution layers where a first memory buffer stores imagery data and a second memory buffer stores filter coefficients as suggested by Yang into the combination of Kim and Kakishita because all of these systems addressing the process of generating results (imagery data) by using convolution based on the filter coefficients/weights. Doing so would be desirable because the combination of Kim and Kakishita would be more efficient by performing convolutions based on the imagery data and filter coefficients, wherein the AI chip may be extendable and scalable, and the filter coefficients may be retrieved from the second memory buffer to the CNN processing block (Yang, [par. 0045, 0046]).
With respect to claim 3, the combination of Kim, Kakishita and Yang teaches wherein the generatingdilated filter based on the filter coefficients which are read from the first holding unit and is stored in a third holding unit (Kakishita, a DC kernel 1102 is an example of a kernel used for the dilated convolution processing and is constructed by inserting one zero into adjacent elements of the C kernel which is an example of a kernel used for the convolution processing. The C kernel is basically used in a convolution processing and a pooling processing, and the DC kernel is then used by inserting a predetermined number of zeros between elements of a C kernel [par. 0009, 0091-0100]).
With respect to claim 4, the combination of Kim, Kakishita and Yang teaches wherein the generating generates the dilated filter by adding a coefficient of 0 to the filter coefficients read from the first holding unit (Kim, besides the weights (or filter coefficients) held in the original convolution filter, the dilating process adds more zero weights to the convolution filter, and hence dilate the filter size [col. 8, lines 25-45 and FIGS. 3A-C]).
With respect to claim 5, the combination of Kim, Kakishita and Yang teaches wherein the generating generates the dilated filter by inserting (extension dilation rate - 1) Os as coefficients between filter coefficients adjacent to each other in vertical and horizontal directions in the filter coefficients read from the first holding unit (Kim, the convolution filter is dilated (or extended) by inserting the zero weights besides the original weights of the convolution filter. The 0s are inserted between the 1s coefficients those are adjacent to each other. The size of the dilated matrix depends on the dilation rate, for example, FIG. 3B applies dilation rate of 2 to obtain a 7x7 matrix [col. 8, lines 25-60 and FIGS. 3A-D]).
With respect to claim 13, it is a data processing method that corresponding to the data processing apparatus of claim 1. Therefore, it is rejected for the same reason as claim 1 above.
With respect to claim 14, it is a non-transitory computer-readable storage medium storing a computer program that corresponding to the data processing apparatus of claim 1. Therefore, it is rejected for the same reason as claim 1 above.
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over Kim et al (US 10095977 B1) hereafter Kim, in view of Kakishita et al (US 20220172341 A1) hereafter Kakishita, and further in view of Yang et al (US 20210019602 A1) hereafter Yang, as applied in claim 1 above, and further in view of Mishima et al (US 20150221099 A1) hereafter Mishima.
With respect to claim 2, the combination of Kim, Kakishita and Yang teaches the invention as claimed in claim 1 above.
However, the combination of Kim, Kakishita and Yang does not disclose wherein the generating storesdilated filter generated by the generating in a second holding unit.
In the same field of endeavor, Mishima teaches wherein the generating storesdilated filter generated by the generating in a second holding unit (Each imaging device has a line memory and a memory. The filter coefficients are stored in the memory, and the program is stored in the processing circuit, such that the processing unit processes the values stored in the line memory by referring to the filter coefficients [par. 0181-0187 and FIG. 16]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated the concept of storing the filter coefficients in a memory as suggested by Mishima into the combination of Kim, Kakishita and Yang because all of these systems addressing the process of applying convolution operations using filter coefficients to generate optimal results. Doing so would be desirable because the combination of Kim, Kakishita and Yang would be more efficient by storing filter coefficients into the memory in order to combine with the convolution operations to generate extended (dilated) filters, and hence to improve the process of image segmentation (Mishima, [par. 0181-0187]).
Claims 6 and 7 are rejected under 35 U.S.C. 103 as being unpatentable over Kim et al (US 10095977 B1) hereafter Kim, in view of Kakishita et al (US 20220172341 A1) hereafter Kakishita, further in view of Yang et al (US 20210019602 A1) hereafter Yang, as applied in claim 1 above, and further in view of Ikeda (US 20210233231 A1) hereafter Ikeda.
With respect to claim 6, the combination of Kim, Kakishita and Yang teaches the invention as claimed in claim 1 above.
The combination of Kim, Kakishita and Yang teaches wherein the generating generates the dilated filter (Kim, there are 3 layers included in the invention: encoding layer, intermediate layer and decoding layer. The intermediate layer includes a number of convolution filters configured to be dilated in the dilating process, for example, the intermediate layer includes the (3-1)th filter to the (3-K)th filter where K is the number of filters in this layer [col. 7, lines 30-50 and FIGS. 2A-B]).
However, the combination of Kim, Kakishita and Yang does not disclose the generating generates the dilated filter.
In the same field of endeavor, Ikeda teaches the generating generates the dilated filter for each layer (a convolution process configured to apply a convolution filter to the inspection image, whereas the convolutional neural network may perform multiple convolution processes multiple times on multiple intermediate layers. [par. 0029-0030 and FIG. 1]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated the concept of applying convolution filter on multiple intermediate layers as suggested by Ikeda into the combination of Kim, Kakishita and Yang because all of these systems addressing the process of applying convolution operations on the filters to obtain new feature maps. Doing so would be desirable because the combination of Kim, Kakishita and Yang would be more efficient by performing an intermediate process on an intermediate layer more than once by using the extended (dilated) filters in order to extract a specific feature image/audio, to reduce the amount of computations, and hence to increase the performance of the image/audio segmentation (Ikeda, [par. 0029, 0030]).
With respect to claim 7, the combination of Kim, Kakishita, Yang and Ikeda teaches the invention as claimed in claim 6 above.
The combination of Kim, Kakishita, Yang and Ikeda teaches wherein the performing performs, for each layer (Ikeda, the convolution process is performed on each layer of multiple intermediate layers [par. 0029-0030]), convolution (Kim, the communication unit receives an input image and the processor performs the convolutional operations multiple times on the input image in the encoding layer and deconvolutional operations in the decoding layer [col. 6, lines 35-65 and col. 7, lines 1-15]) corresponding to the layer by using the filter coefficients of the dilated filter generated for the layer by the generating and the image feature data (Kim, all of the dilation convolution filters included in the intermediate layer are distributed between the (1-1)th filter – (1-K)th filter and the (2-1)th filter – (2-3)th filter of the encoding and decoding layers. [col. 7, lines 5-50 and FIGS. 2A-B]).
Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Kim et al (US 10095977 B1) hereafter Kim, in view of Kakishita et al (US 20220172341 A1) hereafter Kakishita, and in view of Ikeda (US 20210233231 A1) hereafter Ikeda, further in view of Yang et al (US 20210019602 A1) hereafter Yang, as applied in claim 6 above, and further in view of Oosake et al (US 20210133473 A1) hereafter Oosake.
With respect to claim 11, the combination of Kim, Kakishita, Yang and Ikeda teaches the invention as claimed in claim 6 above.
However, the combination of Kim, Kakishita, Yang and Ikeda does not specifically disclose wherein the each layer is each layer of a hierarchical neural network, and the filter coefficients are weighting factors corresponding to the layer.
In the same field of endeavor, Oosake teaches wherein the each layer is each layer of a hierarchical neural network (the intermediate layers, those are shared by the first and second layer, configured to receive the feature quantities. This architecture is known as a hierarchical neural network that includes first layer, second layer and multiple intermediate layers [par. 0008, 0009]),
and the filter coefficients are weighting factors corresponding to the layer (the layers included in the CNN configuration having a structure of a plurality of nodes those connected by edges and corresponding to a plurality of weight parameters. The CNN is the network that uses a filter to extract features such as extracting edges from an image, hence the filter coefficients are weight parameters that corresponding to the layers [par. 0048, 0049 and FIGS. 1, 2]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated a brief explanation of feature data, layers and filter coefficients in CNN as suggested by Oosake into the combination of Kim, Kakishita, Yang and Ikeda because all of these systems addressing the process of applying convolution operations on the filters to obtain new feature elements. Doing so would be desirable because the combination of Kim, Kakishita, Yang and Ikeda would be more efficient by using the feature data and the layer that corresponding to the data and the filter coefficients to enhance the process of using hierarchical CNN, so as to improve the performance of image/audio segmentation (Oosake, [par. 0008, 0009]).
Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Kim et al (US 10095977 B1) hereafter Kim, in view of Kakishita et al (US 20220172341 A1) hereafter Kakishita, further in view of Yang et al (US 20210019602 A1) hereafter Yang, as applied in claim 1 above, and further in view of Oosake et al (US 20210133473 A1) hereafter Oosake.
With respect to claim 12, the combination of Kim, Kakishita and Yang teaches the invention as claimed in claim 1 above.
However, the combination of Kim, Kakishita and Yang does not disclose wherein the performing performs an activation process and/or a pooling process on a result of the convolution.
In the same field of endeavor, Oosake teaches wherein the performing performs an activation process and/or a pooling process on a result of the convolution (After the convolution is performed at each layer, a feature map is obtained with a size equal to the input image, and the pooling process is a process that reduces the feature map output as a result of the convolutional operation to obtain a new feature map [par. 0049]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated the concept of using the pooling process as suggested by Oosake into the combination of Kim, Kakishita and Yang because all of these systems addressing the process of applying convolution operations on the filters to obtain new feature maps. Doing so would be desirable because the combination of Kim, Kakishita and Yang would be more efficient by combining an additional process such as the pooling process to extract features and to reduce (or enlarge) the feature map output, so as to increase the performance of the image/audio segmentation (Oosake, [par. 0018-0021]).
Response to Arguments
The examiner respectfully acknowledges the applicant’s amendments to claims 1-7, 13 and 14.
Applicant’s arguments filed on 12/19/2025 regarding the rejections to claims 1-7 and 11-14 under 35 USC 101 have been considered and are consequently withdrawn.
Applicant’s arguments filed on 12/19/2025 regarding the rejections to claims 1-7 and 11-14 under 35 USC 103 have been fully considered and moot in view of new ground of rejection (see rejection above).
Conclusion
Applicant’s amendment necessitated the new grounds of rejection presented in this Office Action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP 706.07(a). Applicant is remined of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filled within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Quoc Phung whose telephone number is (703) 756 1330. The examiner can normally be reached on Monday through Friday from 9am to 5pm PT.
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) athttp://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Jennifer Welch can be reached on 571-272-7212. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Q.L.P./Examiner, Art Unit 2143
/JENNIFER N WELCH/Supervisory Patent Examiner, Art Unit 2143
Prosecution Insights