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 .
Status of Claims
This office action is in response to the response filed June 4, 2026. Claims 17-22 and 27-29 are pending and stand rejected. No substantive amendment was entered; claim 22 was corrected only as to dependency. This action is made FINAL. Applicant's response entered no amendment to the substance of any claim, this action introduces no new ground of rejection. The rejections below are maintained over the same references and on the same basis as the non-final action mailed March 11, 2026. See MPEP 706.07(a).
Information Disclosure Statement
The IDS(s) has/have been considered and placed in the application file.
Response to Arguments
Applicant's arguments filed June 4, 2026 have been fully considered but are not persuasive. No substantive amendment was entered; claim 22 was corrected only as to dependency. The rejection of claims 17-22 and 27-29 under 35 U.S.C. 103 over Koivisto in view
of Zhang is maintained.
Applicant argues that Zhang concerns only the shifting of an entire image relative to a downsampling grid rather than a within-image, position-dependent confidence pattern, and that Koivisto's thresholds are general-purpose filtering criteria not tied to individual objects. Neither argument is persuasive, for the reasons set out in the rejection below. Zhang's periodic-N shift-equivariance is a property of where an object falls relative to the sampling lattice, expressed in pixels. Koivisto filters individual detected objects on the basis of each object's own detected object region and location. Correcting the confidence estimate of an individual detected object as a function of that object's position within the image was known in the art before the effective filing date, as Kuppers shows.
Applicant's remaining arguments directed to claims 19-22 rest on the same asserted
distinction and are not persuasive for the same reasons.
Double Patenting
The nonstatutory double patenting rejection is based on a judicially created doctrine grounded in public policy (a policy reflected in the statute) so as to prevent the unjustified or improper timewise extension of the “right to exclude” granted by a patent and to prevent possible harassment by multiple assignees. A nonstatutory double patenting rejection is appropriate where the conflicting claims are not identical, but at least one examined application claim is not patentably distinct from the reference claim(s) because the examined application claim is either anticipated by, or would have been obvious over, the reference claim(s). See, e.g., In re Berg, 140 F.3d 1428, 46 USPQ2d 1226 (Fed. Cir. 1998); In re Goodman, 11 F.3d 1046, 29 USPQ2d 2010 (Fed. Cir. 1993); In re Longi, 759 F.2d 887, 225 USPQ 645 (Fed. Cir. 1985); In re Van Ornum, 686 F.2d 937, 214 USPQ 761 (CCPA 1982); In re Vogel, 422 F.2d 438, 164 USPQ 619 (CCPA 1970); In re Thorington, 418 F.2d 528, 163 USPQ 644 (CCPA 1969).
A timely filed terminal disclaimer in compliance with 37 CFR 1.321(c) or 1.321(d) may be used to overcome an actual or provisional rejection based on nonstatutory double patenting provided the reference application or patent either is shown to be commonly owned with the examined application, or claims an invention made as a result of activities undertaken within the scope of a joint research agreement. See MPEP § 717.02 for applications subject to examination under the first inventor to file provisions of the AIA as explained in MPEP § 2159. See MPEP § 2146 et seq. for applications not subject to examination under the first inventor to file provisions of the AIA . A terminal disclaimer must be signed in compliance with 37 CFR 1.321(b).
The filing of a terminal disclaimer by itself is not a complete reply to a nonstatutory double patenting (NSDP) rejection. A complete reply requires that the terminal disclaimer be accompanied by a reply requesting reconsideration of the prior Office action. Even where the NSDP rejection is provisional the reply must be complete. See MPEP § 804, subsection I.B.1. For a reply to a non-final Office action, see 37 CFR 1.111(a). For a reply to final Office action, see 37 CFR 1.113(c). A request for reconsideration while not provided for in 37 CFR 1.113(c) may be filed after final for consideration. See MPEP §§ 706.07(e) and 714.13.
The USPTO Internet website contains terminal disclaimer forms which may be used. Please visit www.uspto.gov/patent/patents-forms. The actual filing date of the application in which the form is filed determines what form (e.g., PTO/SB/25, PTO/SB/26, PTO/AIA /25, or PTO/AIA /26) should be used. A web-based eTerminal Disclaimer may be filled out completely online using web-screens. An eTerminal Disclaimer that meets all requirements is auto-processed and approved immediately upon submission. For more information about eTerminal Disclaimers, refer to www.uspto.gov/patents/apply/applying-online/eterminal-disclaimer.
Claim 17-22 and 27-29 rejected on the ground of nonstatutory double patenting as being unpatentable over claims 24-31, 33-34, and 38-42 of application 18/033,226 (U.S. Patent No. 12,602,905). Although the claims at issue are not identical, they are not patentably distinct from each other because the ‘905 patent specifically recites determining risk based on a reference point within a marginal distance of a dip. These are considered minor variations that a person of ordinary kill in the art would find obvious.
Claim 17-22 and 27-29 rejected on the ground of nonstatutory double patenting as being unpatentable over the patented claims 15-21 and 23-25 of application 18/033,385 (U.S. Patent No. 12,573,190). Although the claims at issue are not identical, they are not patentably distinct from each other because both characterize a multi-scale CNN that exhibits periodic confidence dips defined by a scaling factor and recite the specific technical solution of adapting object detection parameters, such as lowering a class threshold or increasing a confidence metric, based on the distance between an object’ reference point and those dips.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 17-22 and 27-29 is/are rejected under 35 U.S.C. 103 as being unpatentable over Koivisto et al. (US 2024/0192320 A1 – hereinafter “Koivisto”) in view of Zhang (Making Convolutional Networks Shift-Invariant Again – hereinafter “Zhang”)
Claims 17 and 29.
Koivisto teaches an object detection arrangement comprising (Koivisto ¶52, Fig. 1A: Object detection system 100)
a memory storing computer program instructions (Koivisto Fig. 16; ¶312: “the memory 1604 may store computer-readable instructions (e.g., that represent a program(s) and/or a program element(s)”); and
a controller configured to execute the computer program instructions, whereby the controller is configured to (Koivisto ¶314: “The CPU(s) 1606 may be configured to execute the computer-readable instructions to control one or more components of the computing device 1600 to perform one or more of the methods and/or processes described herein”):
receive image data representing an image containing a candidate object (Koivisto, ¶51: “The object detector 106 may be configured to analyze sensor data, such as image data, received from the communications manager 104 and generate detected object data that is representative of detected objects captured in the sensor data.”)
determine that the candidate object is at
convolutional neural network (CNN) (Koivisto ¶79: "the object detection system 100
may employ multi-scale inferencing using the object detector 106. In multi-scale
inferencing, the object detector 106 may infer the same images multiple times at
different scales." Koivisto ,¶90: Multi-headed architecture, "A first head may be
larger (e.g., have a larger stride such as 16) than a second head (e.g., having a stride of
8).")
(Koivisto ¶91: "16x16 pixel blocks" and "8x8 pixel blocks" are the pixel areas correspond to the stride. Koivisto ¶103: "each output cell (e.g., grid cell) may correspond to a pixel area (e.g., 16x16) in the input image. This pixel area may correspond to a spatial element region described herein, and may be based on the stride of the DNN." This follows from Zhang’s own definition that the network’s shift-equivariance holds only for shifts that are integer multiples of N (Zhang § 3.1). For a spatially localized object, a translation of that object by k pixels is, at the object’s support, indistinguishable from a whole-image translation by k pixels. A person of ordinary skill in the art would therefore understand that the classification confidence assigned to a localized candidate object is a function of that object’s position modulo N, that is, its phase relative to the sampling grid, so that the multi-scale CNN exhibits periodic dips in confidence spaced according to a confidence distance defined by the scaling factor.); and
compensate classification of the candidate object by the multi-scale CNN, to reduce the
risk of the candidate object being misclassified (Koivisto ¶78: "an aggregated detection may be retained based at least in part on the confidence score exceeding a threshold (e.g., an adjustable value). This filtering may be performed to reduce false positives." Koivisto performs this determination for a particular detected object, not by a class-wide criterion alone, Koivisto generates a confidence score for the individual detected object from features of that object. Koivisto ¶105: “The confidence score generator 112 may generate a confidence score for a cluster or aggregated detection based at least in part on features associated with at least the aggregated detection.” Koivisto ¶107: “the confidence score generator 112 comprises an MLP.” Those features include the object’s own position in the image. Koivisto ¶110: “To determine a confidence score for an aggregated detection, the feature determiner 110 may extract one or more features from … the aggregated detection” where “a feature for an aggregated detection may be based at least in part on the aggregated detected object data, such as a location and/or detected object region assigned to the aggregated detection” and “a feature may correspond to a height, a width, an area, a center point or midpoint (e.g., x and y, which may correspond to separate features in some examples)”; Koivisto ¶111: statistics over “the locations ( e.g., detected object region coordinates) and/or detected object regions.” Koivisto also filters individual detected objects on the basis of each object’s own region ¶¶69, 72-73.).
Koivisto discloses all of the subject matter as described above except for specifically teaching “risk of being misclassified … that exhibits periodic dips in classification confidence along an axis of the image, the periodic dips spaced according to a confidence distance defined by a scaling factor.” However, Zhang in the same field of endeavor teaches risk of being misclassified … that exhibits periodic dips in classification confidence along an axis of the image (Zhang §1, Fig. 1: "small input shifts or translations can cause drastic changes in the output." Figure 1 explicitly plots classification confidence (Prob of correct class) vs. diagonal shift in pixels. The baseline network shows periodic oscillations/dips in confidence as the image is shifted along the axis i.e. risk of being misclassified when combined with Koivisto’s adjustable threshold (¶0080). Zhang §4.3, Fig. 5: Feature distance heatmaps throughout VGG architecture show periodic stippling patterns at each downsampling layer. "On the baseline network, shift-equivariance is reduced each time down sampling takes place. Periodic-N shift-equivariance holds, with N doubling with each downsampling."), the periodic dips spaced according to a confidence distance defined by a scaling factor (Zhang §3.1: "In some cases, the definitions in Eqns. 1, 2 may hold only when shifts (Δh, Δw) are integer multiples of N. We refer to such scenarios as periodic shift-equivariance/invariance of N. For example, periodic-2 shift-invariance means that even-pixel shifts produce an identical output, but odd-pixel shifts may not." Zhang §4.3: "Periodic-N shift-equivariance still holds, as indicated by the stippling pattern in 'pool 1' , and each subsequent subsampling doubles the factor N." This means at stride 2 pooling the periodicity will be 2 pixels; likewise, after the second pooling the periodicity will be 4 pixels; and after third pooling the periodicity will be 8 pixels. The periodic spacing is defined by the accumulated scaling/stride factor. Koivisto ¶0093: Stride 16 (first head) and stride 8 (second head).
It would have been obvious to one of ordinary skill in the art to combine the teachings of Koivisto and Zhang because Koivisto's multi-scale, stride-based CNN detection system
relies on confidence scores to filter detections (¶78) and provides adjustable per-class threshold (¶¶70, 78, 89). Zhang expressly identifies that strided CNNs of this type produce periodic confidence oscillations at intervals defined by the network's scaling factor (§ 3.1, § 4.3, Fig. 1). This a documented deficiency that a skilled artisan would have been motivated to detect and compensate for the periodic confidence suppression that Zhang documents, in order to improve the reliability of Koivisto's confidence-based filtering, with reasonable expectation of success, because Koivisto’s thresholds are already adjustable and Zhang quantifies the artifact’s periodicity, so that the correction is deterministic. This motivation is supported by KSR rationale (A), combining known elements by known methods to yield predictable results, and rationale (G), as Zhang's analysis constitutes a suggestion in the prior art to address a known artifact of the precise CNN architecture Koivisto employs. MPEP § 2141 (III).
That the confidence estimate of an individual detected object can be corrected as a function of that object's own position within the image was known in the art before the effective filing date (Kuppers, Abstract: “Our approach allows, for the first time, to obtain calibrated confidence estimates with respect to image location and box scale"; Kuppers, Fig. 1: "The miscalibration score highly depends on the object location (the center coordinates cx and cy) and increases as the prediction gets close to the image boundaries"). A person of ordinary skill in the art would therefore have had a reasonable expectation of success in correcting the confidence of a detected object according to where that object falls in the image. Kuppers is of record and is cited as evidence of the state of the art and of a reasonable expectation of success. Kuppers is not relied upon to supply any limitation of the claims.
Claim 18.
Koivisto in view of Zhang discloses the object detection arrangement of claim 17, wherein the multiscale CNN classifies the candidate object by comparing a classification confidence determined for the candidate object for each of one or more object classes with a corresponding class threshold (Koivisto ¶78: Filtering based on "confidence score exceeding a threshold (e.g., an adjustable value)." ¶89: "different thresholds may be used for different classes." ¶90: "the detected object filter 116A may retain at least some of the detected objects ... but may use a higher threshold for the associated coverage values.").
Claim 19.
Koivisto in view of Zhang discloses the object detection arrangement of claim 18, wherein the controller is configured to compensate the classification of the candidate object by increasing the classification confidence and/or lowering the classification threshold, for at least one of the one or more object classes (Koivisto ¶ 70: the threshold "may be based at least in part on a class of the detected object"; ¶ 78: the threshold is "an adjustable value"; ¶ 89: "different thresholds may be used for different classes"; ¶93: Different heads use different thresholds; filter may "use a higher threshold for the associated coverage values" for certain detections, this implies adjustability in both directions.).
Koivisto teaches adjustable confidence thresholds. One of ordinary skill, knowing from Zhang that objects near periodic dips have artificially suppressed confidence, would have found it obvious to either increase the confidence score or lower the classification threshold for such at-risk objects. This is the natural corrective action when the cause of suppressed confidence is a known CNN architecture artifact rather than genuine low detection quality. The threshold so adjusted is applied to a particular detected object on the basis of that object's own detected object region (Koivisto, ¶¶ 69, 105, 110).
Claim 20.
Koivisto in view of Zhang discloses the object detection arrangement of claim 17, wherein the controller is configured to determine that the candidate object is at risk of being misclassified by a multi-scale convolutional neural network (CNN) by determining that a reference point of the candidate object lies within a defined marginal distance of one of the periodic dips, the defined marginal distance measured in pixels (Zhang §3.1: “Periodic-N shift-equivariance/invariance … In some cases, the definitions in Eqns. 1, 2 may hold only when shifts … are integer multiples of N.” Objects at positions that are NOT multiples of N are at risk. Fig. 1: Shows confidence as continuous function of pixel shift-dips occur at specific pixel locations. §4.3: "each subsequent subsampling doubles the factor N" - the periodic dip locations are at defined pixel intervals. Koivisto determines the confidence for a detected object from that object’s own position, ¶110.).
The rationale provided for the rejection of claim(s) 17 is applicable to claim 20, mutatis mutandis. Accordingly, claim 20 is rendered obvious by the combination of Koivisto and Zhang.
Claim 21.
Koivisto in view of Zhang discloses the object detection arrangement of claim 17, wherein the controller is configured to determine an amount of compensation applied based on a distance in pixels between a location of a reference point of the candidate object and a nearest one of the confidence dips (Zhang Fig. 1: Classification confidence varies as a continuous function of pixel shift – the magnitude of the confidence drop depends on the pixel distance from a periodic peak. Objects exactly at a dip suffer the largest drop; objects slightly off a dip suffer less.).
Zhang's Figure 1 shows confidence as a smooth function of pixel position. The
farther the object is from a dip, the less confidence is suppressed and the less compensation is needed. Scaling the compensation amount based on pixel distance to the nearest dip is the natural engineering approach when the confidence-vs-position relationship is known. The compensation mechanism is Koivisto’s adjustable per-class threshold (Koivisto ¶78). Koivisto computes statistics over “the locations ( e.g., detected object region coordinates)” of detected objects (Koivisto ¶111), so scaling the corrective by the pixel distance between the object’s reference point and the nearest dip uses a quantity Koivisto already derives.
Claim 22.
The object detection arrangement of claim 19, wherein the controller is configured to compensate the classification of the candidate object by the multiscale CNN by forming a shifted image and classifying the candidate object by processing the shifted image via the multi-scale CNN, and wherein the controller is configured to form the shifted image by shifting the image data by a defined number of pixels along one or both axes of the image to increase a distance in pixels between a reference point of the candidate object and a nearest one of the periodic dips (Zhang §3.1: "circular shifting and convolution” When shifting, pixels are rolled off one edge to the other side. §4.1 (lmageNet): "An alternative is to take a shifted crop from a larger image." Fig. 1: Explicitly demonstrates that shifting the image changes classification confidence – shifting by even 1 pixel can dramatically change output. Koivisto ¶79: Multi-scale inferencing, "infer the same images multiple times at different scales." Already processes the same image multiple times.).
Zhang explicitly demonstrates that shifting the image changes classification confidence (Fig. 1). Zhang's entire experimental methodology involves classifying shifted versions of images. One of ordinary skill, understanding from Zhang that an object at a confidence dip would have higher confidence if the image were shifted by a few pixels, would have found it obvious to shift the image by a defined number of pixels to move the object's reference point away from the nearest dip. Koivisto already supports processing images at different scales; adding a shifted version is a trivial extension of the existing multi-inference pipeline.
Claim 27.
The object detection arrangement of claim 17, wherein the object detection arrangement is a smartphone or a tablet computer (Koivisto ¶194: "a client device 1420 may be embodied as a personal computer (PC), a laptop computer, a mobile device, a smartphone, a tablet computer, a smart watch, a wearable computer"; Koivisto ¶35: “other types of devices may be used to implement that various approaches described herein, such as robots, camera systems, weather forecasting devices, medical imaging devices, etc.”). The limitation is met by Koivisto’s express disclosure. Official Notice is no longer relied upon for this claim.
Claim 28.
The object detection arrangement of claim 17, wherein the object detection arrangement is an optical see-through device (Official Notice: Optical see-through AR devices (e.g., Hololens, Magic Leap) employing CNN-based object detection were well-known before the EFD. It would have been obvious to implement Koivisto's object detection on an optical see-through device for augmented reality applications.). Official Notice was taken of this fact in the non-final action mailed March 11, 2026. Applicant did not traverse the noticed fact in the response filed June 4, 2026, and it is therefore taken to be admitted prior art. See MPEP 2144.03(C).
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded 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 filed 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 nonprovisional extension fee (37 CFR 1.17(a)) 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 mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Ross Varndell whose telephone number is (571)270-1922. The examiner can normally be reached M-F, 9-5 EST.
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, O’Neal 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 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 https://ppair-my.uspto.gov/pair/PrivatePair. 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.
/Ross Varndell/Primary Examiner, Art Unit 2674
Prosecution Insights