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 .
Response to Arguments
The reply filed on 2/02/2026 has been entered. Applicant’s arguments with respect to claims 1-20 have been considered but are moot in view of new ground(s) of rejection caused by the amendments. Claims 1-20 are pending in this application and have been considered below.
Priority
Applicant claims the benefit of US Provisional Application No. 63/461,050, filed April 21, 2023. Claims 1-20 have been afforded the benefit of this filing date.
Information Disclosure Statement
The IDSs dated 01/04/2024 that have been previously considered remain placed in the application file.
1st 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.
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 1, 2, 3, 4, 5, 6, 7, 8, 10, 11, 12, 13, 14, 15, 16, 17, 18, and 19 are rejected under 35 U.S.C. 103 as obvious over US Patent Publication 2021 0004974 A1, (Guizilini et al.) in view of US Patent Publication 2015 0228110 A1, (Hecht) and European Patent Application Publication 1999 0903932 A2, (Metcalfe).
Claim 1
Regarding claim 1, Guizlini et al. teach a prediction system comprising: a memory storing instructions that, when executed by a processor, cause the processor to: ("a memory communicably coupled to one or more processors," par. 9) generate rays with camera intrinsics ("a memory communicably coupled to one or more processors," par. 9) to form a grid for an image frame ("the training module 230 may restrict the search involved in projecting the 3D points to a small h×w grid in the context image I.sub.c surrounding the (u, v) coordinates of the target pixel," par. 64); by an encoder during training of a learning model ("training module includes instructions to train the depth model," par. 9), and the pixel boundaries are defined by the grid ("the training module 230 may restrict the search involved in projecting the 3D points to a small h×w grid in the context image I.sub.c surrounding the (u, v) coordinates of the target pixel," par. 64); extract features from the rays; ("the encoder includes a variety of separate layers that operate on the monocular image, and ... convert the visual information of the monocular image into embedded state information in the form of encoded features," par. 48) and compare scaled depth estimates to a ground truth for a grid resolution using the features ("The particular supervised loss function implemented by the training module 230 for deriving the supervised loss may vary according to the implementation but generally compares the ground truth data 470 with corresponding points within the depth map 430 in order to compare the points with corresponding estimations and ground truths," par. 66) and adjust the learning model using the scaled depth estimates ("the training module 230 uses the loss value to train the depth model 260, the ray surface model 270, and the pose model 280 by backpropagating the loss value (e.g., supervised and self-supervised loss values) and updating parameters of the noted models," par. 67).
Guizilini et al. do not explicitly teach all of inject noise to individually perturb pixels randomly within pixel boundaries using a first range for the rays and the noise, the first range factors an image parameter for the image frame; and remove a subset of the rays randomly by the encoder.
However, Metcalfe teaches to inject noise to individually perturb pixels randomly within pixel boundaries ("perturbing the relationship between the threshold value and the multi-level grey signal according to an image classification of the multi-level grey scale pixel value, thereby effecting the output from said reduction means," col. 4, line 8) using a first range for the rays and the noise, ("This noise signal is fed to adder 408 wherein the noise signal is added to the threshold value so as to perturb the relationship between the threshold and the incoming video signal," col. 15, line 11) the first range factors an image parameter for the image frame ("a noise profile which may be selected based on the grey level of the video signal and the window effect pointer associated with the pixel being processed," col. 14, line 5).
Therefore, taking the teachings of Guizlini et al. and Metcalfe as a whole, 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 modify depth estimation system as taught by Guizilini et al. to use noise profiles as taught by Metcalfe. The suggestion/motivation for doing so would have been that, “An additional benefit of allowing both the location and the intensity of the threshold relationship perturbation to be programmable is a flexibility to properly apply the correct amount of perturbing noise to any type of preprocessed image prior to applying error diffusion. Preprocessing operations include tonal reproduction curve (TRC) input mapping, gain and offset adjustment, spot overlap compensation, etc. These preprocessing operations all tend to shift objectionable periodic patterns to input grey level location other than what has been conventionally expected. Moreover, the noise look-up table can be uniquely programmed and optimized to render images with good quality corresponding to any of the situations described above.” as noted by the Metcalfe disclosure in col. 13, line 61, which also motivates combination because the combination would predictably have a greater efficiency as there is a reasonable expectation that the trained model will achieve higher accuracy and better image quality by tailoring the noise used during training to specific, variable preprocessed input characteristics, thereby reducing artifacts and improving generalization; and/or because doing so merely combines prior art elements according to known methods to yield predictable results.
Additionally, Hecht teaches to remove a subset of the rays randomly by the encoder and extract features from the rays ("Depending on the underlying features of the render process, representative sample 116 may be computed at the intersection of view ray 108 with the boundary of a voxel (highlighted by the gray circles), at the voxel center, or based on a stochastic model," par. 25).
Therefore, taking the teachings of Guizlini et al., Metcalfe, and Hecht as a whole, 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 modify depth estimation system as taught by Guizilini et al. to use noise profiles as taught by Metcalfe and stochastic sampling as taught by Hecht. More specifically, stochastic sampling permits improved robustness to estimation errors, enhanced depth estimation accuracy, and reduced computational time/memory requirements. This known benefit is applicable to the depth estimation system and using noise profiles as they both share characteristics and capabilities, namely, they are directed to improving the accuracy and reliability of learning models. Therefore, it would have been recognized this combination would have yielded predictable results because (i) the level of ordinary skill in the art demonstrated by the references applied shows the ability to incorporate stochastic sampling in a depth estimation learning model and (ii) the benefits of such a combination would have been recognized by those of ordinary skill in the art.
The rejection of system claim 1 above applies mutatis mutandis to the corresponding limitations of non-transitory computer-readable medium claim 10 and method claim 12 while noting that the rejection above cites to both device and method disclosures. Claims 10 and 12 are mapped below for clarity of the record and to specify any new limitations not included in claim 1.
Claim 2
Regarding claim 2, Guizlini et al., Metcalfe, and Hecht teach the prediction system of claim 1 as noted above.
Guizilini et al. additionally teach to move the rays from centers of the pixel boundaries ("the neural camera model adjusts the predicted ray vectors according to the camera offset or a learned camera center point(s)," par. 85) to increase ray types in three dimensions, the ray types representing different geometries for objects within the image frame; ("the training module includes instructions to lift pixels to produce three-dimensional points using the ray surface ... the training module then projects the three-dimensional points onto a context image to synthesize the image," par. 56) wherein the image parameter is one of a color or a brightness associated with the image frame ("the depth smoothness component generally functions to regularize the depth in textureless image regions by penalizing high depth gradients in areas of low color gradients," par. 66).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 3
Regarding claim 3, Guizlini et al., Metcalfe, and Hecht teach the prediction system of claim 1 as noted above.
Guizilini et al. additionally teach to identify the features about objects within the image frame by the learning model using known priors ("the depth model 260 provides estimates of depths for different aspects depicted in the image," par. 50) representing appearance characteristics about the objects without depth information ("the training module 230 may initialize the training by using the ray surface template Q.sub.0 in combination with the learned residual ray surface {circumflex over (Q)}.sub.r to inject geometric priors into the learning framework," par. 55).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 4
Regarding claim 4, Guizlini et al., Metcalfe, and Hecht teach the prediction system of claim 3 as noted above.
Guizilini et al. additionally teach to interpolate between centers of the pixel boundaries ("the camera center coincides with the origin of the reference coordinate system," par. 59) and the pixels for interpreting location of the features including the noise ("the decoding layers generally function to up-sample, through sub-pixel convolutions and other mechanisms, the previously encoded features into the depth map," par. 49).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 5
Regarding claim 5, Guizlini et al., Metcalfe, and Hecht teach the prediction system of claim 3 as noted above.
Guizilini et al. additionally teach wherein the rays include expanded observations for a search space of the features ("the training module 230 may restrict the search involved in projecting the 3D points to a small h×w grid in the context image I.sub.c surrounding the (u, v) coordinates of the target pixel p.sub.t.," par. 64) and the scaled depth is independent of different resolutions ("the depth map 430 is a pixel-wise prediction of depths for the image," par. 50).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 6
Regarding claim 6, Guizlini et al., Metcalfe, and Hecht teach the prediction system of claim 1 as noted above.
Guizilini et al. additionally teach to transfer scale priors ("inject geometric priors into the learning framework," par. 55) estimated during implementation to a vehicle having a sensor that acquires an image dataset, ("The monocular images are generally derived from one or more monocular videos that are comprised of a plurality of frames. As described herein, monocular images that comprise the monocular images are, for example, images from the camera 126 or another imaging device that is part of a video, and that encompasses a field-of-view (FOV) about the vehicle," par. 35) wherein the sensor has geometric properties that differ from the camera intrinsics ("The monocular images are generally derived from one or more monocular videos that are comprised of a plurality of frames. As described herein, monocular images that comprise the monocular images are, for example, images from the camera 126 or another imaging device that is part of a video, and that encompasses a field-of-view (FOV) about the vehicle," par. 35).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 7
Regarding claim 7, Guizlini et al., Metcalfe, and Hecht teach the prediction system of claim 6 as noted above.
Guizilini et al. additionally teach to transform the image dataset by rotation without perturbations of the image dataset ("a machine learning algorithm that generates a rigid-body transformation 440 according to the noted images," par. 51).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 8
Regarding claim 8, Guizlini et al., Metcalfe, and Hecht teach the prediction system of claim 1 as noted above.
Guizilini et al. additionally teach to interpolate the features for generating image embeddings using the rays, the image embeddings associated with visual characteristics about the image frame; ("the encoder may include a series of layers that function to fold (i.e., adapt dimensions of the feature map to retain the features) encoded features into separate channels iteratively reducing spatial dimensions of the image 410 while packing additional channels with information about embedded states of the features," par. 46) and execute back-propagation to adjust weights of the learning model ("backpropagating the loss value and updating the parameters," par. 67) from losses between the scaled depth estimates to the ground truth, ("compares the ground truth data with corresponding points within the depth map," par. 66) and the ground truth has real depth measurements about objects within the image frame ("depth system implements the training architecture to use sparse depth data as ground truth information," par. 24).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 10
Regarding claim 10, Guizlini et al. teach a non-transitory computer-readable medium comprising: instructions that, when executed by a processor, cause the processor to: ("a memory communicably coupled to one or more processors," par. 9) generate rays with camera intrinsics ("a memory communicably coupled to one or more processors," par. 9) to form a grid for an image frame ("the training module 230 may restrict the search involved in projecting the 3D points to a small h×w grid in the context image I.sub.c surrounding the (u, v) coordinates of the target pixel," par. 64); by an encoder during training of a learning model ("training module includes instructions to train the depth model," par. 9), and the pixel boundaries are defined by the grid ("the training module 230 may restrict the search involved in projecting the 3D points to a small h×w grid in the context image I.sub.c surrounding the (u, v) coordinates of the target pixel," par. 64); extract features from the rays; ("the encoder includes a variety of separate layers that operate on the monocular image, and ... convert the visual information of the monocular image into embedded state information in the form of encoded features," par. 48) and compare scaled depth estimates to a ground truth for a grid resolution using the features ("The particular supervised loss function implemented by the training module 230 for deriving the supervised loss may vary according to the implementation but generally compares the ground truth data 470 with corresponding points within the depth map 430 in order to compare the points with corresponding estimations and ground truths," par. 66) and adjust the learning model using the scaled depth estimates ("the training module 230 uses the loss value to train the depth model 260, the ray surface model 270, and the pose model 280 by backpropagating the loss value (e.g., supervised and self-supervised loss values) and updating parameters of the noted models," par. 67).
Guizilini et al. do not explicitly teach all of inject noise to individually perturb pixels randomly within pixel boundaries using a first range for the rays and the noise, the first range factors an image parameter for the image frame; and remove a subset of the rays randomly by the encoder.
However, Metcalfe teaches to inject noise to individually perturb pixels randomly within pixel boundaries ("perturbing the relationship between the threshold value and the multi-level grey signal according to an image classification of the multi-level grey scale pixel value, thereby effecting the output from said reduction means," col. 4, line 8) using a first range for the rays and the noise, ("This noise signal is fed to adder 408 wherein the noise signal is added to the threshold value so as to perturb the relationship between the threshold and the incoming video signal," col. 15, line 11) the first range factors an image parameter for the image frame ("a noise profile which may be selected based on the grey level of the video signal and the window effect pointer associated with the pixel being processed," col. 14, line 5).
Additionally, Hecht teaches to remove a subset of the rays randomly by the encoder and extract features from the rays ("Depending on the underlying features of the render process, representative sample 116 may be computed at the intersection of view ray 108 with the boundary of a voxel (highlighted by the gray circles), at the voxel center, or based on a stochastic model," par. 25).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 11
Regarding claim 11, Guizlini et al., Metcalfe, and Hecht teach the non-transitory computer-readable medium of claim 10 as noted above.
Guizilini et al. additionally teach wherein the instructions to inject the noise further include to: move the rays from centers of the pixel boundaries ("the neural camera model adjusts the predicted ray vectors according to the camera offset or a learned camera center point(s)," par. 85) to increase ray types in three dimensions, the ray types representing different geometries for objects within the image frame; ("the training module includes instructions to lift pixels to produce three-dimensional points using the ray surface ... the training module then projects the three-dimensional points onto a context image to synthesize the image," par. 56) wherein the image parameter is one of a color or a brightness associated with the image frame ("the depth smoothness component generally functions to regularize the depth in textureless image regions by penalizing high depth gradients in areas of low color gradients," par. 66).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 12
Regarding claim 12, Guizlini et al. teach a method comprising: generating rays with camera intrinsics ("a memory communicably coupled to one or more processors," par. 9) to form a grid for an image frame ("the training module 230 may restrict the search involved in projecting the 3D points to a small h×w grid in the context image I.sub.c surrounding the (u, v) coordinates of the target pixel," par. 64); by an encoder during training of a learning model ("training module includes instructions to train the depth model," par. 9), and the pixel boundaries are defined by the grid ("the training module 230 may restrict the search involved in projecting the 3D points to a small h×w grid in the context image I.sub.c surrounding the (u, v) coordinates of the target pixel," par. 64); extracting features from the rays; ("the encoder includes a variety of separate layers that operate on the monocular image, and ... convert the visual information of the monocular image into embedded state information in the form of encoded features," par. 48) and comparing scaled depth estimates to a ground truth for a grid resolution using the features ("The particular supervised loss function implemented by the training module 230 for deriving the supervised loss may vary according to the implementation but generally compares the ground truth data 470 with corresponding points within the depth map 430 in order to compare the points with corresponding estimations and ground truths," par. 66) and adjust the learning model using the scaled depth estimates ("the training module 230 uses the loss value to train the depth model 260, the ray surface model 270, and the pose model 280 by backpropagating the loss value (e.g., supervised and self-supervised loss values) and updating parameters of the noted models," par. 67).
Guizilini et al. do not explicitly teach all of injecting noise to individually perturb pixels randomly within pixel boundaries using a first range for the rays and the noise, the first range factors an image parameter for the image frame; and removing a subset of the rays randomly by the encoder.
However, Metcalfe teaches injecting noise to individually perturb pixels randomly within pixel boundaries ("perturbing the relationship between the threshold value and the multi-level grey signal according to an image classification of the multi-level grey scale pixel value, thereby effecting the output from said reduction means," col. 4, line 8) using a first range for the rays and the noise, ("This noise signal is fed to adder 408 wherein the noise signal is added to the threshold value so as to perturb the relationship between the threshold and the incoming video signal," col. 15, line 11) the first range factors an image parameter for the image frame ("a noise profile which may be selected based on the grey level of the video signal and the window effect pointer associated with the pixel being processed," col. 14, line 5).
Additionally, Hecht teaches removing a subset of the rays randomly by the encoder and extract features from the rays ("Depending on the underlying features of the render process, representative sample 116 may be computed at the intersection of view ray 108 with the boundary of a voxel (highlighted by the gray circles), at the voxel center, or based on a stochastic model," par. 25).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 13
Regarding claim 13, Guizlini et al., Metcalfe, and Hecht teach the method of claim 12 as noted above.
Guizilini et al. additionally teach moving the rays from centers of the pixel boundaries ("the neural camera model adjusts the predicted ray vectors according to the camera offset or a learned camera center point(s)," par. 85) to increase ray types in three dimensions, the ray types representing different geometries for objects within the image frame; ("the training module includes instructions to lift pixels to produce three-dimensional points using the ray surface ... the training module then projects the three-dimensional points onto a context image to synthesize the image," par. 56) wherein the image parameter is one of a color or a brightness associated with the image frame ("the depth smoothness component generally functions to regularize the depth in textureless image regions by penalizing high depth gradients in areas of low color gradients," par. 66).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 14
Regarding claim 14, Guizlini et al., Metcalfe, and Hecht teach the method of claim 12 as noted above.
Guizilini et al. additionally teach identifying the features about objects within the image frame by the learning model using known priors ("the depth model 260 provides estimates of depths for different aspects depicted in the image," par. 50) representing appearance characteristics about the objects without depth information ("the training module 230 may initialize the training by using the ray surface template Q.sub.0 in combination with the learned residual ray surface {circumflex over (Q)}.sub.r to inject geometric priors into the learning framework," par. 55).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 15
Regarding claim 15, Guizlini et al., Metcalfe, and Hecht teach the method of claim 14 as noted above.
Guizilini et al. additionally teach interpolating between centers of the pixel boundaries ("the camera center coincides with the origin of the reference coordinate system," par. 59) and the pixels for interpreting location of the features including the noise ("the decoding layers generally function to up-sample, through sub-pixel convolutions and other mechanisms, the previously encoded features into the depth map," par. 49) .
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 16
Regarding claim 16, Guizlini et al., Metcalfe, and Hecht teach the method of claim 14 as noted above.
Guizilini et al. additionally teach wherein the rays include expanded observations for a search space of the features ("the training module 230 may restrict the search involved in projecting the 3D points to a small h×w grid in the context image I.sub.c surrounding the (u, v) coordinates of the target pixel p.sub.t.," par. 64) and the scaled depth is independent of different resolutions ("the depth map 430 is a pixel-wise prediction of depths for the image," par. 50).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 17
Regarding claim 17, Guizlini et al., Metcalfe, and Hecht teach the method of claim 12 as noted above.
Guizilini et al. additionally teach transferring scale priors estimated during implementation ("inject geometric priors into the learning framework," par. 55) to a vehicle having a sensor that acquires an image dataset, ("The monocular images are generally derived from one or more monocular videos that are comprised of a plurality of frames. As described herein, monocular images that comprise the monocular images are, for example, images from the camera 126 or another imaging device that is part of a video, and that encompasses a field-of-view (FOV) about the vehicle," par. 35) wherein the sensor has geometric properties that differ from the camera intrinsics ("leveraging a non-parametric camera model to permit learning of arbitrary camera types for monocular depth estimation," par. 23).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 18
Regarding claim 18, Guizlini et al., Metcalfe, and Hecht teach the method of claim 17 as noted above.
Guizilini et al. additionally teach transforming the image dataset by rotation without perturbations of the image dataset ("a machine learning algorithm that generates a rigid-body transformation 440 according to the noted images," par. 51).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
Claim 19
Regarding claim 19, Guizlini et al., Metcalfe, and Hecht teach the method of claim 12 as noted above.
Guizilini et al. additionally teach interpolating the features for generating image embeddings using the rays, the image embeddings associated with visual characteristics about the image frame; ("the encoder may include a series of layers that function to fold (i.e., adapt dimensions of the feature map to retain the features) encoded features into separate channels iteratively reducing spatial dimensions of the image 410 while packing additional channels with information about embedded states of the features," par. 46) and executing back-propagation to adjust weights of the learning model ("backpropagating the loss value and updating the parameters," par. 67) from losses between the scaled depth estimates to the ground truth, ("compares the ground truth data with corresponding points within the depth map," par. 66) and the ground truth has real depth measurements about objects within the image frame ("depth system implements the training architecture to use sparse depth data as ground truth information," par. 24).
Guizlini et al., Metcalfe, and Hecht are combined as per claim 1.
2nd Claim Rejections - 35 USC § 103
Claims 9 and 20 are rejected under 35 U.S.C. 103 as obvious over US Patent Publication 2021 0004974 A1, (Guizilini et al.), US Patent Publication 2015 0228110 A1, (Hecht), and European Patent Application Publication 1999 0903932 A2, (Metcalfe) in further view of US Patent Publication 2024 0331872 A1, (Arora et al.) and US Patent Publication 2022 0391633 A1, (Harikumar et al.).
Claim 9
Regarding claim 9, Guizlini et al., Metcalfe, and Hecht teach the prediction system of claim 2 as noted above.
Guizilini et al. additionally teach the search space including latent representations about the image frame ("the encoder may include a series of layers that function to fold (i.e., adapt dimensions of the feature map to retain the features) encoded features into separate channels iteratively reducing spatial dimensions of the image 410 while packing additional channels with information about embedded states of the features," par. 46), by the encoder during a training iteration, ("iteratively processing different images and updating through the disclosed training process," par. 78) embeddings associated with the rays, ("representations about embedded states of features included in the image," par. 52) using the image parameter ("the depth smoothness component generally functions to regularize the depth in textureless image regions by penalizing high depth gradients in areas of low color gradients," par. 66).
Guizilini et al. do not explicitly teach all of to expand a search space for the features by the encoder resizing the image frame randomly, select embeddings associated with the rays within a second range.
However, Arora et al. teach to expand a search space for the features by the encoder resizing the image frame randomly ("all the scans in a batch are augmented randomly using the following methods such as resizing the image by a random factor," par. 40).
Therefore, taking the teachings of Guizlini et al., Metcalfe, Hecht, and Arora et al. as a whole, 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 modify depth estimation system as taught by Guizilini et al., stochastic sampling as taught by Metcalfe, and noise profiles as taught by Hecht to use random image resizing as taught by Arora et al. The suggestion/motivation for doing so would have been that, “The above augmentations are intended to make the models robust to this variance. Data augmentation is a well-accepted technique in the training of deep learning algorithms” as noted by the Arora et al. disclosure in paragraph [0040], which also motivates combination because the combination would predictably have a greater efficiency as there is a reasonable expectation that the resulting model would demonstrate increased generalization, improved convergence speeds, and enhanced performance across diverse input scenarios, without requiring significant retraining efforts; and/or because doing so merely combines prior art elements according to known methods to yield predictable results.
Additionally, Harikumar et al. teaches to select embeddings within a second range ("the instance extraction system 106 identifies digital image embeddings of the plurality of digital image embeddings that are within a threshold similarity range of the embedding," par. 98).
Therefore, taking the teachings of Guizlini et al., Metcalfe, Hecht, Arora et al., and Harikumar et al. as a whole, 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 modify depth estimation system as taught by Guizilini et al., stochastic sampling as taught by Metcalfe, and noise profiles as taught by Hecht to use random image resizing as taught by Arora et al. and selecting image embeddings using a range as taught by Harikumar et al. The suggestion/motivation for doing so would have been that, “the instance extraction system 106 compares the embedding 714 with the plurality of digital image embeddings to identify the similar instance images ” as noted by the Harikumar et al. disclosure in paragraph [0098], which also motivates combination because the combination would predictably have additional utility as there is a reasonable expectation that depth estimation accuracy, robustness to input variations, and identification of similar instance images would be improved by efficiently processing image embeddings and reducing noise in the depth map generation; and/or because doing so merely combines prior art elements according to known methods to yield predictable results.
Claim 20
Regarding claim 20, Guizlini et al., Metcalfe, and Hecht teach the method of claim 13 as noted above.
Guizilini et al. additionally teach the search space including latent representations about the image frame ("the encoder may include a series of layers that function to fold (i.e., adapt dimensions of the feature map to retain the features) encoded features into separate channels iteratively reducing spatial dimensions of the image 410 while packing additional channels with information about embedded states of the features," par. 46), by the encoder during a training iteration, ("iteratively processing different images and updating through the disclosed training process," par. 78) embeddings associated with the rays, ("representations about embedded states of features included in the image," par. 52) using the image parameter ("the depth smoothness component generally functions to regularize the depth in textureless image regions by penalizing high depth gradients in areas of low color gradients," par. 66).
Guizilini et al. do not explicitly teach all of expanding a search space for the features by the encoder resizing the image frame randomly, selecting embeddings associated with the rays within a second range.
However, Arora et al. teach expanding a search space for the features by the encoder resizing the image frame randomly ("all the scans in a batch are augmented randomly using the following methods such as resizing the image by a random factor," par. 40).
Additionally, Harikumar et al. teaches selecting embeddings associated with the rays within a second range ("the instance extraction system 106 identifies digital image embeddings of the plurality of digital image embeddings that are within a threshold similarity range of the embedding," par. 98).
Guizlini et al., Metcalfe, Hecht, Arora et al., and Harikumar et al. are combined as per claim 9.
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 Karsten F. Lantz whose telephone number is (571)272-4564. The examiner can normally be reached Monday-Friday 8:00-4:00.
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, Ms. Jennifer Mehmood can be reached on 571-272-2976. 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.
/Karsten F. Lantz/Examiner, Art Unit 2664
Date: 4/29/2026
/JENNIFER MEHMOOD/Supervisory Patent Examiner, Art Unit 2664
Prosecution Insights