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
Claims 1-15 and 21-25 are pending, with claims 21-25 withdrawn from consideration due to a restriction by original presentation (see below). Claims 16-20 are canceled.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 2/4/26 was filed after the mailing date of the Non-Final Office Action on 12/4/25. The submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Response to Arguments
Applicant’s arguments, see p. 9, filed 3/4/26, with respect to the drawings have been fully considered and are persuasive. The drawing objection of 12/4/25 has been withdrawn.
Applicant’s arguments, see p. 9, filed 3/4/26, with respect to claims 5, 10, 16, and 17 have been fully considered and are persuasive. The claim objections of 12/4/25 have been withdrawn.
Applicant’s arguments, see p. 9, filed 3/4/26, with respect to claims 1-20 have been fully considered and are persuasive. The 35 U.S.C. 112(b) rejections of 12/4/25 have been withdrawn.
Applicant's arguments filed 3/4/26 with respect to the 35 U.S.C. 103 rejections have been fully considered but they are not persuasive.
First, Applicant argues in p. 11-12 of the remarks, that Kim does not disclose the following limitations in claim 1: “iteratively performing: generating, using a feature extraction model characterized by first parameters and taking the training image as input, noised feature vectors by: determining a feature vector for individual pixels in the training image; and applying noise to each feature vector to produce the noised feature vectors.” However, the Examiner used the combination of Sohn and Kim to teach the above limitations, not Kim alone. Please see the 103 rejection below for claim mapping of these limitations.
Second, Applicant argues in p. 13 of the remarks, that in Kim, the noise applied is not to feature vectors. The Examiner respectfully disagrees. Since Applicant has not provided specifics as to what the feature vectors are in the claims, under broadest reasonable interpretation of the claim, the Examiner interprets the noise-infused event voxels, for example, as the feature vectors (see Kim: Pg. 3; Fig. 3; Pgs. 4-5). It is well known that voxels represent values (i.e., features), which is further evident in Kim (see Kim, Pg. 4: voxel value).
Third, Applicant argues in p. 13 of the remarks, that Kim does not disclose the noise being characterized by a reconstruction loss determined by a preceding iteration. Please note that the new prior art of record Han has been used to teach the iterative aspect of this limitation (see the 103 rejection below), and therefore the arguments are moot.
Fourth, Applicant argues in pg. 13 of the remarks, that Kim does not disclose determining a feature vector for individual pixels. The Examiner respectfully disagrees. Kim teaches obtaining a voxel grid to produce images using neural networks in which the networks are trained with noise-infused event voxels (see Kim, Pg. 3). As previously mentioned, the Examiner interprets the noise-infused event voxels, for example, as the feature vectors. Since voxels are volumetric pixels that have values (i.e., features), and Kim teaches a voxel grid, there are multiple individual voxels (i.e., pixels with values/features).
Election/Restrictions
Newly submitted claims 21-25 are directed to an invention that is independent or distinct from the invention originally claimed for the following reasons: Claim 21 relates to using a trained model to determine the location of a user device, whereas claims 1 and 10 relate to iteratively training a model. They are related as sub-combination usable together.
Since applicant has received an action on the merits for the originally presented invention, this invention has been constructively elected by original presentation for prosecution on the merits. Accordingly, claims 21-25 are withdrawn from consideration as being directed to a non-elected invention. See 37 CFR 1.142(b) and MPEP § 821.03.
To preserve a right to petition, the reply to this action must distinctly and specifically point out supposed errors in the restriction requirement. Otherwise, the election shall be treated as a final election without traverse. Traversal must be timely. Failure to timely traverse the requirement will result in the loss of right to petition under 37 CFR 1.144. If claims are subsequently added, applicant must indicate which of the subsequently added claims are readable upon the elected invention.
Should applicant traverse on the ground that the inventions are not patentably distinct, applicant should submit evidence or identify such evidence now of record showing the inventions to be obvious variants or clearly admit on the record that this is the case. In either instance, if the examiner finds one of the inventions unpatentable over the prior art, the evidence or admission may be used in a rejection under 35 U.S.C. 103 or pre-AIA 35 U.S.C. 103(a) of the other invention.
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, 3-10, and 12-15 are rejected under 35 U.S.C. 103 as being unpatentable over Sohn et al. (US 2020/0143079 A1, hereinafter “Sohn”) in view of “Privacy-Preserving Visual Localization with Event Cameras” by Kim et al. (hereinafter “Kim”) and further in view of “Deep Image Prior for Super Resolution of Noisy Image” by Han et al. (hereinafter “Han”).
Regarding claim 1, Sohn teaches, A computer-implemented method, comprising (Sohn, Para. [0004]: “A computer-implemented method for protecting visual private data by preventing data reconstruction from latent representations of deep networks is presented”):
receiving a training image (Sohn, Para. [0004]: latent features are obtained from an input image and an encoder is trained; Sohn, Figs. 4 and 5: input image is input into an encoder);
and iteratively performing (Sohn, Para. [0037]; Sohn, Para. [0038]: decoder Dec and encoder Enc are updated; Sohn, Paras. [0046]-[0047]; Sohn, Fig. 4: updating the decoder; Sohn, Fig. 5: updating the encoder; Note: the Examiner interprets updating as iteratively performing):
generating, using a feature extraction model Sohn, Para. [0004]: latent features are obtained from an input image and an encoder is trained; Sohn, Fig. 3: an image 312 is input into an image encoder 314 and latent feature 316 is obtained; Sohn, Para. [0043]; Sohn, Fig. 7: step 701; Note: the examiner interprets the latent feature as a feature vector):
determining a feature vector Sohn, Para. [0004]: latent features are obtained from an input image and an encoder is trained; Sohn, Fig. 3: an image 312 is input into an image encoder 314 and latent feature 316 is obtained; Sohn, Para. [0043]; Sohn, Fig. 7: step 701; Note: the examiner interprets the latent feature as a feature vector);
generating, using a reconstructor model taking the noised feature vectors as input, a reconstructed imageSohn, Fig. 7: In step 705, a decoder is trained to reconstruct the input image from the latent features; Sohn, Paras. [0043]-[0045]; Sohn, As shown in Para. [0046] and Figs. 4 and 5, decoder 332 is trained to reconstruct an input X from latent encoding);
determining a reconstruction loss by at least comparing the training image with the reconstructed image (Sohn, Fig. 7: In step 707, an encoder is trained to maximize a reconstruction error; Sohn, Para. [0045], the encoder reconstructs the output image from the feature via the loss function (Reconstruction Loss 336); Sohn, As shown in Fig. 3, the input image 312 and reconstructed output image 334 are used to determine a reconstruction loss 336);
Sohn does not expressly disclose the following limitations: characterized by first parameters, noised feature vectors by: determining a feature vector for individual pixels; and applying noise to each feature vector to produce the noised feature vectors; the noised feature vectors as input, a reconstructed image, the reconstructor model characterized by second parameters; determining a noise loss using the noise applied to each feature vector, the noise characterized by the a reconstruction loss determined by a preceding iteration; and updating the first parameters based on the noise loss.
However, Kim teaches, characterized by first parameters, noised feature vectors by (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks. These neural networks have neural network parameters; Kim, Fig. 3: the network is trained using noise-infused event voxels; Kim, Pgs. 4-5: during training, a network takes fixed noise Enoise as additional input; Note: the Examiner interprets noise-infused voxels as noised feature vectors):
determining a feature vector for individual pixels (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks; Kim, Fig. 3: the network is trained using noise-infused event voxels; Kim, Pgs. 4-5: during training, a network takes fixed noise Enoise as additional input; Note: the Examiner interprets noise-infused voxels as noised feature vectors. Voxels are 3D pixels);
and applying noise to each feature vector to produce the noised feature vectors (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks; Kim, Fig. 3: the network is trained using noise-infused event voxels; Kim, Pgs. 4-5: during training, a network takes fixed noise Enoise as additional input; Note: since noise is added as input/the voxels are infused with noise, noise is applied to the voxel/feature vector);
the noised feature vectors as input, a reconstructed image, the reconstructor model characterized by second parameters (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks. These neural networks have neural network parameters; Kim, Fig. 3: the network is trained using noise-infused event voxels and loss functions to prevent possible attacks from the service provider to reconstruct images using shared information; Kim, Pgs. 4-5: The network learns to reconstruct images from noise-infused voxel grids in which the noise includes Enoise. During training, a network takes fixed noise Enoise as additional input and network weights learned from training are obfuscated);
determining a noise loss using the noise applied to each feature vector, the noise characterized by a reconstruction loss (Kim, Fig. 3: the network is trained using noise-infused event voxels and loss functions to prevent possible attacks from the service provider to reconstruct images using shared information. Both reconstruction loss and adversarial loss are determined using the noise-infused event voxels (i.e., noisy events); Note: the Examiner interprets adversarial loss as a noise loss; Kim, Pg. 5: adversarial loss enforces the new network weights to deviate from the original network);
and updating the first parameters based on the noise loss (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks. These neural networks have neural network parameters; Kim, Pg. 5: adversarial loss enforces the new network weights to deviate from the original network; Note: the Examiner interprets the new network weights as updated parameters).
It would have been obvious, before the effective filing date of the claimed invention, to one of ordinary skill in the art to combine generating noised feature vectors, determining a feature vector for individual pixels, applying noise to each feature vector to produce the noised feature vectors, generating a reconstructed image using noised feature vectors, determining a noise loss, and updating parameters based on the noise loss as taught by Kim with the method of Sohn in order to robustly localize in challenging scenarios and effectively hide sensitive visual content while preserving localization performance (Kim, Pg. 2). Therefore, one of ordinary skill in the art would be capable to have combined these elements claimed by known methods, and that in combination, each element merely performs the same function as it does separately.
The combination of Sohn and Kim does not expressly disclose the following limitation: a reconstruction loss determined by a preceding iteration.
However, Han teaches, a reconstruction loss determined by a preceding iteration (Han, Pg. 4: reconstruction loss; Han, Pg. 5: “The proposed framework is optimized through several iterations. In each optimization step, the proposed network outputs the reconstructed image. We hypothesize that the result of an earlier stage can be used as a constraint to reconstruct a noiseless HR image for the following stage. Accordingly, our self-supervision loss utilizes the output of the previous iteration step during training. SSL compares the output image of the current step with that of the previous step. By performing this, the reconstructed image maintains the learned signal without following the noise in the target image.”; Han, Pg. 6: Algorithm 1; Note: Since the result of an earlier stage/previous step can be used as a constraint to reconstruct a noiseless HR image for the following stage, it shows that noise is characterized by the reconstruction).
It would have been obvious, before the effective filing date of the claimed invention, to one of ordinary skill in the art to combine a reconstruction loss being determined by a preceding iteration and noise being characterized by the reconstruction as taught by Han with the combined method of Sohn and Kim in order to prevent the reconstructed image from becoming noisy (Han, Pg. 2) . Therefore, one of ordinary skill in the art would be capable to have combined these elements claimed by known methods, and that in combination, each element merely performs the same function as it does separately. It is for at least the aforementioned that the Examiner has reached a conclusion of obviousness with respect to claim 1.
Regarding claim 3, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 1.
The combination of Sohn, Kim, and Han further teaches, The method of claim 1 (see claim 1), further comprising updating the second parameters based on the reconstruction loss (Kim, Pgs. 4-5: The network learns to reconstruct images from noise-infused voxel grids in which the noise includes Enoise. During training, a network takes fixed noise Enoise as additional input and network weights learned from training are obfuscated. One of the two losses imposed during training is reconstruction loss; Note: the Examiner interprets learning of network weights as updating parameters).
The proposed combination as well as the motivation for combining the Sohn, Kim, and Han references presented in the rejection of claim 1 apply to claim 3 and are incorporated herein by reference. Therefore, the method recited in claim 3 is met by Sohn, Kim, and Han.
Regarding claim 4, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 3.
The combination of Sohn, Kim, and Han further teaches, The method of claim 3 (see claim 3 above), wherein updating the second parameters comprises updating the second parameters to minimize the reconstruction loss (Kim, Pgs. 4-5: The network learns to reconstruct images from noise-infused voxel grids in which the noise includes Enoise. During training, a network takes fixed noise Enoise as additional input and network weights learned from training are obfuscated. One of the two losses imposed during training is reconstruction loss; Note: the Examiner interprets learning of network weights as updating parameters; Sohn, Para. [0023]: “the input X is reconstructed from the feature Z by minimizing the reconstruction loss”).
The proposed combination as well as the motivation for combining the Sohn, Kim, and Han references presented in the rejection of claim 3 apply to claim 4 and are incorporated herein by reference. Therefore, the method recited in claim 4 is met by Sohn, Kim, and Han.
Regarding claim 5, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 1.
The combination of Sohn, Kim, and Han further teaches, The method of claim 1 (see claim 1 above), wherein the feature extraction model comprises a noiser network characterized by at least one of the first parameters (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks. These neural networks have neural network parameters; Kim, Fig. 3: the network is trained using noise-infused event voxels; Kim, Pgs. 4-5: during training, a network takes fixed noise Enoise as additional input; Note: the Examiner interprets the network trained with noise-infused event voxels as the noiser network).
The proposed combination as well as the motivation for combining the Sohn, Kim, and Han references presented in the rejection of claim 1 apply to claim 5 and are incorporated herein by reference. Therefore, the method recited in claim 5 is met by Sohn, Kim, and Han.
Regarding claim 6, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 1.
The combination of Sohn, Kim, and Han further teaches, The method of claim 1 (see claim 1 above), wherein the training image comprises an identifiable element (Sohn, Para. [0004]: latent features are obtained from an input image and an encoder is trained; Sohn, Figs. 4 and 5: input image is input into an encoder; Sohn, Para. [0042]: the privacy of private faces images is protected; Note: the Examiner interprets faces as an identifiable element).
Regarding claim 7, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 6.
The combination of Sohn, Kim, and Han further teaches, The method of claim 6 (see claim 6 above), wherein the reconstructed image comprises an obfuscated element corresponding to the identifiable element of the training image (Kim, Pg. 7: a face detection algorithm is applied on the image reconstructions in which the number of detected faces largely decreases after filtering. The filtering obfuscates facial features; Sohn, Para. [0042]: the privacy of private faces images is protected).
The proposed combination as well as the motivation for combining the Sohn, Kim, and Han references presented in the rejection of claim 6 apply to claim 7 and are incorporated herein by reference. Therefore, the method recited in claim 7 is met by Sohn, Kim, and Han.
Regarding claim 8, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 1.
The combination of Sohn, Kim, and Han further teaches, The method of claim 1 (see claim 1 above), wherein determining a reconstruction loss comprises computing a similarity measure between the training image and the reconstructed image (Kim, As shown in Fig. 3, reconstruction loss is determined based on a similarity between the clean events (i.e., training images) and noisy events (i.e., reconstructed images); Kim, Pg. 5: equation (3) provides the reconstruction loss in which d is the LPIPS distance; Note: LPIPS computes the similarity between two image patches).
The proposed combination as well as the motivation for combining the Sohn, Kim, and Han references presented in the rejection of claim 1 apply to claim 8 and are incorporated herein by reference. Therefore, the method recited in claim 8 is met by Sohn, Kim, and Han.
Regarding claim 9, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 1.
The combination of Sohn, Kim, and Han further teaches, The method of claim 1 (see claim 1 above), wherein the feature vector comprises a plurality of feature values encoding information associated with the pixel corresponding to the feature vector (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks; Kim, Fig. 3: the network is trained using noise-infused event voxels; Kim, Pgs. 4-5: Voxels have values and during training, a network takes fixed noise Enoise as additional input; Note: the Examiner interprets noise-infused voxels as feature vectors and the voxel values as feature values. Voxels are 3D pixels), and wherein applying the noise to each feature vector comprises perturbing one or more of the plurality of feature values (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks; Kim, Fig. 3: the network is trained using noise-infused event voxels; Kim, Pgs. 4-5: Voxels values are replaced and during training, a network takes fixed noise Enoise as additional input; Note: the Examiner interprets noise-infused voxels as feature vectors, the voxel values as feature values, and replacing/infusing of noise in voxels as perturbing feature values).
The proposed combination as well as the motivation for combining the Sohn, Kim, and Han references presented in the rejection of claim 1 apply to claim 9 and are incorporated herein by reference. Therefore, the method recited in claim 9 is met by Sohn, Kim, and Han.
Regarding claim 10, Sohn teaches, A system, comprising (Sohn, Para. [0006]: “A system for protecting visual private data by preventing reconstruction from latent representations of deep networks is presented”):
one or more processors (Sohn, Para. [0006]: “The system includes a memory and one or more processors in communication with the memory”; Sohn, Para. [0050]);
and one or more memories storing computer-executable instructions that, when executed by the one or more processors, cause the system to at least (Sohn, Para. [0006]: “The system includes a memory and one or more processors in communication with the memory”; Sohn, Para. [0050]; Sohn, Para. [0071]: computer program instructions are executed by a processor; Sohn, Para. [0072]: computer program instructions may be stored in a computer readable medium):
receive a training image (Sohn, Para. [0004]: latent features are obtained from an input image and an encoder is trained; Sohn, Figs. 4 and 5: input image is input into an encoder);
and iteratively (Sohn, Para. [0037]; Sohn, Para. [0038]: decoder Dec and encoder Enc are updated; Sohn, Paras. [0046]-[0047]; Sohn, Fig. 4: updating the decoder; Sohn, Fig. 5: updating the encoder; Note: the Examiner interprets updating as iteratively performing):
generate, using a feature extraction model Sohn, Para. [0004]: latent features are obtained from an input image and an encoder is trained; Sohn, Fig. 3: an image 312 is input into an image encoder 314 and latent feature 316 is obtained; Sohn, Para. [0043]; Sohn, Fig. 7: step 701; Note: the examiner interprets the latent feature as a feature vector):
determining a feature vector Sohn, Para. [0004]: latent features are obtained from an input image and an encoder is trained; Sohn, Fig. 3: an image 312 is input into an image encoder 314 and latent feature 316 is obtained; Sohn, Para. [0043]; Sohn, Fig. 7: step 701; Note: the examiner interprets the latent feature as a feature vector);
generate, using a reconstructor model taking the Sohn, Fig. 7: In step 705, a decoder is trained to reconstruct the input image from the latent features; Sohn, Paras. [0043]-[0045]; Sohn, As shown in Para. [0046] and Figs. 4 and 5, decoder 332 is trained to reconstruct an input X from latent encoding);
determine a reconstruction loss by at least comparing the training image with the reconstructed image (Sohn, Fig. 7: In step 707, an encoder is trained to maximize a reconstruction error; Sohn, Para. [0045], the encoder reconstructs the output image from the feature via the loss function (Reconstruction Loss 336); Sohn, As shown in Fig. 3, the input image 312 and reconstructed output image 334 are used to determine a reconstruction loss 336);
Sohn does not expressly disclose the following limitations: characterized by first parameters, noised feature vectors by: determining a feature vector for individual pixels; and applying noise to each feature vector to produce the noised feature vectors; the noised feature vectors as input, a reconstructed image, the reconstructor model characterized by second parameters; determine a noise loss using the noise applied to each feature vector, the noise characterized by a reconstruction loss determined by a preceding iteration; and update the first parameters based on the noise loss.
However, Kim teaches, characterized by first parameters, noised feature vectors by (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks. These neural networks have neural network parameters; Kim, Fig. 3: the network is trained using noise-infused event voxels; Kim, Pgs. 4-5: during training, a network takes fixed noise Enoise as additional input; Note: the Examiner interprets noise-infused voxels as noised feature vectors):
determining a feature vector for individual pixels (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks; Kim, Fig. 3: the network is trained using noise-infused event voxels; Kim, Pgs. 4-5: during training, a network takes fixed noise Enoise as additional input; Note: the Examiner interprets noise-infused voxels as noised feature vectors. Voxels are 3D pixels);
and applying noise to each feature vector to produce the noised feature vectors (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks; Kim, Fig. 3: the network is trained using noise-infused event voxels; Kim, Pgs. 4-5: during training, a network takes fixed noise Enoise as additional input; Note: since noise is added as input/the voxels are infused with noise, noise is applied to the voxel/feature vector);
the noised feature vectors as input, a reconstructed image, the reconstructor model characterized by second parameters (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks. These neural networks have neural network parameters; Kim, Fig. 3: the network is trained using noise-infused event voxels and loss functions to prevent possible attacks from the service provider to reconstruct images using shared information; Kim, Pgs. 4-5: The network learns to reconstruct images from noise-infused voxel grids in which the noise includes Enoise. During training, a network takes fixed noise Enoise as additional input and network weights learned from training are obfuscated);
determine a noise loss using the noise applied to each feature vector, the noise characterized by a reconstruction loss (Kim, Fig. 3: the network is trained using noise-infused event voxels and loss functions to prevent possible attacks from the service provider to reconstruct images using shared information. Both reconstruction loss and adversarial loss are determined using the noise-infused event voxels (i.e., noisy events); Note: the Examiner interprets adversarial loss as a noise loss; Kim, Pg. 5: adversarial loss enforces the new network weights to deviate from the original network);
and update the first parameters based on the noise loss (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks. These neural networks have neural network parameters; Kim, Pg. 5: adversarial loss enforces the new network weights to deviate from the original network; Note: the Examiner interprets the new network weights as updated parameters).
It would have been obvious, before the effective filing date of the claimed invention, to one of ordinary skill in the art to combine generating noised feature vectors, determining a feature vector for individual pixels, applying noise to each feature vector to produce the noised feature vectors, generating a reconstructed image using noised feature vectors, determining a noise loss, and updating parameters based on the noise loss as taught by Kim with the method of Sohn in order to robustly localize in challenging scenarios and effectively hide sensitive visual content while preserving localization performance (Kim, Pg. 2). Therefore, one of ordinary skill in the art would be capable to have combined these elements claimed by known methods, and that in combination, each element merely performs the same function as it does separately.
The combination of Sohn and Kim does not expressly disclose the following limitation: a reconstruction loss determined by a preceding iteration.
However, Han teaches, a reconstruction loss determined by a preceding iteration (Han, Pg. 4: reconstruction loss; Han, Pg. 5: “The proposed framework is optimized through several iterations. In each optimization step, the proposed network outputs the reconstructed image. We hypothesize that the result of an earlier stage can be used as a constraint to reconstruct a noiseless HR image for the following stage. Accordingly, our self-supervision loss utilizes the output of the previous iteration step during training. SSL compares the output image of the current step with that of the previous step. By performing this, the reconstructed image maintains the learned signal without following the noise in the target image.”; Han, Pg. 6: Algorithm 1; Note: Since the result of an earlier stage/previous step can be used as a constraint to reconstruct a noiseless HR image for the following stage, it shows that noise is characterized by the reconstruction).
It would have been obvious, before the effective filing date of the claimed invention, to one of ordinary skill in the art to combine a reconstruction loss being determined by a preceding iteration and noise being characterized by the reconstruction as taught by Han with the combined method of Sohn and Kim in order to prevent the reconstructed image from becoming noisy (Han, Pg. 2) . Therefore, one of ordinary skill in the art would be capable to have combined these elements claimed by known methods, and that in combination, each element merely performs the same function as it does separately. It is for at least the aforementioned that the Examiner has reached a conclusion of obviousness with respect to claim 10.
Regarding claim 12, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 10.
The combination of Sohn, Kim, and Han further teaches, The system of claim 10 (see claim 10 above), further comprising updating the second parameters based on the reconstruction loss (Kim, Pgs. 4-5: The network learns to reconstruct images from noise-infused voxel grids in which the noise includes Enoise. During training, a network takes fixed noise Enoise as additional input and network weights learned from training are obfuscated. One of the two losses imposed during training is reconstruction loss; Note: the Examiner interprets learning of network weights as updating parameters).
The proposed combination as well as the motivation for combining the Sohn, Kim, and Han references presented in the rejection of claim 10 apply to claim 12 and are incorporated herein by reference. Therefore, the system recited in claim 12 is met by Sohn, Kim, and Han.
Regarding claim 13, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 12.
The combination of Sohn, Kim, and Han further teaches, The system of claim 12 (see claim 12 above), wherein updating the second parameters comprises updating the second parameters to minimize the reconstruction loss (Kim, Pgs. 4-5: The network learns to reconstruct images from noise-infused voxel grids in which the noise includes Enoise. During training, a network takes fixed noise Enoise as additional input and network weights learned from training are obfuscated. One of the two losses imposed during training is reconstruction loss; Note: the Examiner interprets learning of network weights as updating parameters; Sohn, Para. [0023]: “the input X is reconstructed from the feature Z by minimizing the reconstruction loss”).
The proposed combination as well as the motivation for combining the Sohn, Kim, and Han references presented in the rejection of claim 12 apply to claim 13 and are incorporated herein by reference. Therefore, the system recited in claim 13 is met by Sohn, Kim, and Han.
Regarding claim 14, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 10.
The combination of Sohn, Kim, and Han further teaches, The system of claim 10 (see claim 10 above), wherein determining a reconstruction loss comprises computing a similarity measure between the training image and the reconstructed image (Kim, As shown in Fig. 3, reconstruction loss is determined based on a similarity between the clean events (i.e., training images) and noisy events (i.e., reconstructed images); Kim, Pg. 5: equation (3) provides the reconstruction loss in which d is the LPIPS distance; Note: LPIPS computes the similarity between two image patches).
The proposed combination as well as the motivation for combining the Sohn, Kim, and Han references presented in the rejection of claim 10 apply to claim 14 and are incorporated herein by reference. Therefore, the system recited in claim 14 is met by Sohn, Kim, and Han.
Regarding claim 15, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 10.
The combination of Sohn, Kim, and Han further teaches, The system of claim 10 (see claim 10 above), wherein the feature vector comprises a plurality of feature values encoding information associated with the pixel corresponding to the feature vector (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks; Kim, Fig. 3: the network is trained using noise-infused event voxels; Kim, Pgs. 4-5: Voxels have values and during training, a network takes fixed noise Enoise as additional input; Note: the Examiner interprets noise-infused voxels as feature vectors and the voxel values as feature values. Voxels are 3D pixels), and wherein applying the noise to each feature vector comprises perturbing one or more of the plurality of feature values (Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks; Kim, Fig. 3: the network is trained using noise-infused event voxels; Kim, Pgs. 4-5: Voxels values are replaced and during training, a network takes fixed noise Enoise as additional input; Note: the Examiner interprets noise-infused voxels as feature vectors, the voxel values as feature values, and replacing/infusing of noise in voxels as perturbing feature values).
The proposed combination as well as the motivation for combining the of Sohn, Kim, and Han references presented in the rejection of claim 10 apply to claim 15 and are incorporated herein by reference. Therefore, the system recited in claim 15 is met by of Sohn, Kim, and Han.
Claims 2 and 11 are rejected under 35 U.S.C. 103 as being unpatentable over Sohn et al. (US 2020/0143079 A1, hereinafter “Sohn”) in view of “Privacy-Preserving Visual Localization with Event Cameras” by Kim et al. (hereinafter “Kim”) and further in view of “Deep Image Prior for Super Resolution of Noisy Image” by Han et al. (hereinafter “Han”) and Liu et al. (US 2022/0148191 A1, hereinafter “Liu”).
Regarding claim 2, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 1.
The method of claim 1 (see claim 1 above), wherein updating the first parameters comprises updating the first parameters Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks. These neural networks have neural network parameters; Kim, Pg. 5: adversarial loss enforces the new network weights to deviate from the original network; Note: the Examiner interprets the new network weights as updated parameters).
The proposed combination as well as the motivation for combining the Sohn, Kim, and Han references presented in the rejection of claim 1 apply to claim 2 and are incorporated herein by reference.
The combination of Sohn, Kim, and Han does not expressly disclose the following limitation: to minimize the noise loss.
However, Liu teaches, to minimize the noise loss (Liu, Para. [0037]: the source domain image includes a noisy image; Liu, Para. [0042]: parameter of the network is updated by minimizing the adversarial loss).
It would have been obvious, before the effective filing date of the claimed invention, to one of ordinary skill in the art to combine minimizing noise loss as taught by Liu with the combined method of Sohn, Kim, and Han in order to make predicted results increasingly similar (Liu, Para. 0042). Therefore, one of ordinary skill in the art would be capable to have combined these elements claimed by known methods, and that in combination, each element merely performs the same function as it does separately. It is for at least the aforementioned that the Examiner has reached a conclusion of obviousness with respect to claim 2.
Regarding claim 11, the combination of Sohn, Kim, and Han teaches the limitations as explained above in claim 10.
The combination of Sohn, Kim, and Han further teaches, The system of claim 10 (see claim 10 above), wherein updating the first parameters comprises updating the first parameters Kim, Pg. 3: voxel grid is obtained as input to produce images using neural networks. These neural networks have neural network parameters; Kim, Pg. 5: adversarial loss enforces the new network weights to deviate from the original network; Note: the Examiner interprets the new network weights as updated parameters).
The proposed combination as well as the motivation for combining the Sohn, Kim, and Han references presented in the rejection of claim 10 apply to claim 11 and are incorporated herein by reference.
The combination of Sohn, Kim, and Han does not expressly disclose the following limitation: to minimize the noise loss.
However, Liu teaches, to minimize the noise loss (Liu, Para. [0037]: the source domain image includes a noisy image; Liu, Para. [0042]: parameter of the network is updated by minimizing the adversarial loss).
It would have been obvious, before the effective filing date of the claimed invention, to one of ordinary skill in the art to combine minimizing noise loss as taught by Liu with the combined method of Sohn, Kim, and Han in order to make predicted results increasingly similar (Liu, Para. 0042). Therefore, one of ordinary skill in the art would be capable to have combined these elements claimed by known methods, and that in combination, each element merely performs the same function as it does separately. It is for at least the aforementioned that the Examiner has reached a conclusion of obviousness with respect to claim 11.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Kansy et al. (US 2023/0377214 A1)
Hu et al. (US 10,430,708 B1)
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.
Contact Information
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Daniella M. DiGuglielmo whose telephone number is (571)272-0183. The examiner can normally be reached Monday - Friday 8:00 AM - 4:00 PM.
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, Emily Terrell can be reached at (571)270-3717. 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.
/Daniella M. DiGuglielmo/Examiner, Art Unit 2666
/EMILY C TERRELL/Supervisory Patent Examiner, Art Unit 2666
Prosecution Insights