DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Priority
Applicant claims the benefit of US Provisional Application No. 63562828, filed 3/8/2024. Claims 1-10 and 12-14 have been afforded the benefit of this filing date.
Applicant claims the benefit of US Provisional Application No. 63562828, filed 3/8/2024 but claims 11 and 15-20 contain subject matter not disclosed in the provisional application, thus the effective filing date of 3/10/2025 has been used.
If applicant intends to have these claims afforded the benefit of the earlier filing date, applicant may: (1) amend the claim limitation(s) to avoid it/them containing subject matter not in the provisional application; or (2) present a sufficient showing that the claim limitation(s) recite(s) material from specific sections of the provisional application.
Information Disclosure Statement
The IDS(s) dated 5/29/2026 has/have been considered and placed in the application file.
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.
Claim(s) 1, 6, 10, and 13 is/are rejected under 35 U.S.C. 103 as obvious over Chappell et al (WO 2021146177 A1, hereafter referred to as Chappell) in view of Lischinski et al (US 20140010448 A1, hereafter referred to as Lischinski).
Claim 1
Regarding Claim 1, Chappell teaches A computer-implemented method for eye gaze determination in images, comprising:
using a reference image luminous value to adjust a brightness of an image for input to a Convolutional Neural Network (CNN) model (Chappell in ¶¶18-27 discloses CNN processing of face/eye image data);
reducing one or more dimensions of the image (Chappell in ¶¶24-27, Fig. 4 discloses down sampling the face image and cropping higher resolution eye regions);
inputting the image to the CNN model (Chappell in ¶¶20-21, 27 discloses feeding eye region image data to a trained CNN/YOLO pupil finder);
determining a location of an eye pupil within the image (Chappell in ¶¶20, 27, Fig. 2 discloses CNN detection and (x,y) locations of the pupil centers); and
providing an indication of eye gaze direction for the image (Chappell in ¶¶19-21, Fig. 2 discloses using detected pupil positions to produce gaze point data).
Chappell does not explicitly teach all of using a reference image luminous value to adjust a brightness of an image for input to a Convolutional Neural Network (CNN) model.
However, Lischinski teaches using a reference image luminous value to adjust a brightness of an image for input to a Convolutional Neural Network (CNN) model (Lischinski in ¶¶106-111, Algorithm 1 discloses a luminance target learned from 400 well-exposed faces and bounded correction of the current face image toward that target).
Lischinski is analogous art because Lischinski, like Chappell, is in the field of digital face image processing and is reasonably pertinent to the problem of improving the brightness and visibility of underexposed face images used for pupil/gaze analysis.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the input face image preprocessing of Chappell with the reference face derived target and bounded digital exposure correction of Lischinski because Lischinski teaches that reference face derived bounded exposure correction provides improved facial brightness and visibility while avoiding excessive image change and brightening artifacts (¶¶106-111, Algorithm 1), and one of ordinary skill would have recognized that incorporating this feature into the method of Chappell would correct an underexposed input face image toward the well exposed reference target before deriving and feeding its reduced face/eye inputs to the CNN, with a reasonable expectation of success because Lischinski outputs a corrected digital face image through a pixel adjustment map and restored detail, while Chappell accepts face/eye pixel arrays in a color capable input implementation.
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claim 6
Regarding Claim 6, Chappell in view of Lischinski teaches The method of claim 1, further comprising:
dividing the reference image luminance value by the luminance value of the image to determine a reference ratio for adjusting the brightness of the image (Lischinski in Algorithm 1 discloses the learned target T=120 and current face skin percentile p in f=(T+p)/(2p)).
Claim 10
Regarding Claim 10, Chappell in view of Lischinski teaches The method of claim 1, further comprising:
indicating the eye gaze determination for the image by an “x”, “y” location of an eye center, pass/fail indication, visual, or audio cue (Chappell in ¶20, Fig. 2 discloses the produced (x,y) pupil center result used in the eye gaze determination).
Claim 13
Regarding Claim 13, Chappell in view of Lischinski teaches The method of claim 10, further comprising:
indicating the eye gaze direction through a user interface (Chappell in ¶¶15, 17, and 19-21 discloses a display, gaze coordinates referenced to the display, and the gaze result available at the display).
Claim(s) 2 is/are rejected under 35 U.S.C. 103 as obvious over Chappell et al (WO 2021146177 A1, hereafter referred to as Chappell) and Lischinski et al (US 20140010448 A1, hereafter referred to as Lischinski), further in view of Mao et al (CN 113743254 A, hereafter referred to as Mao).
Claim 2
Regarding Claim 2, Chappell in view of Lischinski teaches The method of claim 1.
Chappell in view of Lischinski does not explicitly teach all of further comprising: extracting facial feature points that include eye features based on a digital library; annotating eye regions of the image; cropping the image based on the eye regions; and training the CNN model using the cropped image.
However, Mao teaches further comprising:
extracting facial feature points that include eye features based on a digital library (Mao in ¶¶[n0005]-n[0009], [n0020] discloses building a collection of digital face/eye training samples and obtaining eye contour landmark coordinates from those sample images);
annotating eye regions of the image (Mao in ¶¶[[n0019], [n0054]-[n0055] discloses marking pupil/eye contours in training face images with a Unity annotation tool);
cropping the image based on the eye regions (Mao in ¶¶[n0029]-[n0038], [n0058] discloses using eye region landmarks/contours to determine bounding boxes and crop training eye images); and
training the CNN model using the cropped image (Mao in ¶¶[n0042]-[n0043], [n0058]-[n0063] discloses training the neural model on the cropped annotated eye image samples).
Mao is analogous art because Mao, like Chappell in view of Lischinski, is in the field of digital eye/face image analysis and is reasonably pertinent to the problem of preparing localized, labeled training inputs for pupil and gaze estimation.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the supervised image CNN training of Chappell in view of Lischinski with eye landmark based cropping of a training image collection and eye/pupil annotation of Mao because Mao teaches that eye landmark based cropping of a training image collection and eye/pupil annotation provides localized eye samples whose spatial labels improve eye/pupil identification and gaze estimation (¶¶[n0019]-[n0020], [n0042]-[n0043], [n0053]-[n0063], [n0068]), and one of ordinary skill would have recognized that incorporating this feature into the method of Chappell in view of Lischinski would train the pupil/gaze image processing stages using the relevant labeled eye regions rather than unlocalized face pixels, with a reasonable expectation of success because Chappel already uses eye crops and trains image based models, while Mao supplies crop coordinates and sample associated annotations that can train the corresponding image input stages without changing the gaze point interface.
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claim(s) 3-5 is/are rejected under 35 U.S.C. 103 as obvious over Chappell et al (US 2021146177 A1, hereafter referred to as Chappell) and Lischinski et al (US 20140010448 A1, hereafter referred to as Lischinski), further in view of Krizhevsky et al (Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet classification with deep convolutional neural networks. Advances in neural information processing systems, 25, hereafter referred to as Krizhevsky).
Claim 3
Regarding Claim 3, Chappell in view of Lischinski teaches The method of claim 1.
Chappell in view of Lischinski does not explicitly teach all of wherein the CNN model comprises five convolution layers and three fully connected layers.
However, Krizhevsky teaches wherein the CNN model comprises five convolution layers and three fully connected layers (Krizhevsky in §§3, 3.5, Fig. 2 discloses 5 convolutional and 3 fully connected learned layers).
Krizhevsky is analogous art because Krizhevsky, like Chappell in view of Lischinski, is in the field of CNN based image analysis and is reasonably pertinent to the problem of choosing model structurer for pupil localization.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the pupil finder image CNN structure of Chappell in view of Lischinski with 5 convolutional and 3 fully connected learned layers of Krizhevsky because Krizhevsky teaches that 5 convolutional and 3 fully connected layers provides a deep image feature representation whose convolutional depth is important to recognition performance (§§3, 3.5), and one of ordinary skill would have recognized that incorporating this feature into the method of Chappell in view of Lischinski would supply spatial feature extraction followed by a trainable mapping to Chappel’s pupil center coordinates, with a reasonable expectation of success because convolution/fully-connected stages operate on cropped pixel arrays and can be retrained with Chappell’s known pupil labels.
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claim 4
Regarding Claim 4, Chappell in view of Lischinski teaches The method of claim 1.
Chappell in view of Lischinski does not explicitly teach all of further comprising: processing the image by a rectified linear activation function (ReLU) after dense and convolution layers and dropout after one or more dense layers as a part of the CNN model.
However, Krizhevsky teaches further comprising:
processing the image by a rectified linear activation function (ReLU) after dense and convolution layers and dropout after one or more dense layers as a part of the CNN model (Krizhevsky in §§3.1, 3.5, 4.2 discloses ReLU at convolutional/fully connected outputs and training time dropout at the first 2 fully connected hidden layers).
Krizhevsky is analogous art because Krizhevsky, like Chappell in view of Lischinski, is in the field of CNN based image analysis and is reasonably pertinent to the problem of slow nonlinear training and dense stage overfitting.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the processing of Chappell in view of Lischinski with ReLU after convolutional/dense transforms and training time dropout in the first 2 dense hidden layers of Krizhevsky because Krizhevsky teaches that ReLU after convolutional/dense transforms and dropout in the first 2 dense hidden layers provides faster training and reduced co-adaptation/overfitting in dense hidden units (§§3.1, 3.5, 4.2), and one of ordinary skill would have recognized that incorporating this feature into the method of Chappell in view of Lischinski would retain image feature learning while regularizing the trainable pupil coordinate mapping, with a reasonable expectation of success because ReLU and dropout act on the adopted convolution/FC hidden outputs and don’t require changing the pupil coordinate or downstream gaze interface.
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claim 5
Regarding Claim 5, Chappell and Lischinski, further in view of Krizhevsky teaches The method of claim 4, wherein the ReLU is used for all convolution layers and the ReLU and dropout are used after a first and second dense layers of the one or more dense layers (Krizhevsky in §§3.5, 4.2 discloses ReLU at every convolutional output and ReLU/dropout at the first two fully connected layers).
Claim(s) 7 is/are rejected under 35 U.S.C. 103 as obvious over Chappell et al (US 2021146177 A1, hereafter referred to as Chappell) and Lischinski et al (US 20140010448 A1, hereafter referred to as Lischinski), further in view of Huang et al (US 20100061653 A1, hereafter referred to as Huang).
Claim 7
Regarding Claim 7, Chappell in view of Lischinski teaches The method of claim 6.
Chappell in view of Lischinski does not explicitly teach all of further comprising: determining the luminance value of the image as the luminance element of a Hue Saturation Luminance (HSV).
However, Huang teaches further comprising:
determining the luminance value of the image as the luminance element of a Hue Saturation Luminance (HSV) (Huang in ¶¶19, 30-31 discloses per pixel brightness in an HSV representation).
Huang is analogous art because Huang, like Chappell in view of Lischinski, is in the field of image brightness processing and is reasonably pertinent to the problem of selecting a brightness representation for measuring and correcting illumination.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the brightness measurement and reference calibration stages of Chappell in view of Lischinski with the HSV value channel representation of Huang because Huang teaches that HSV value channel representation provides per pixel brightness measurements distinct from color saturation values (¶¶19, 30-31), and one of ordinary skill would have recognized that incorporating this feature into the method of Chappell in view of Lischinski would derive a reference face target and a corresponding current face skin percentile in the same HSV-V domain, with a reasonable expectation of success because Huang supplies the brightness channel and Lischinski teaches reference face calibration and percentile based correction.
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claim(s) 9 is/are rejected under 35 U.S.C. 103 as obvious over Chappell et al (US 2021146177 A1, hereafter referred to as Chappell) and Lischinski et al (US 20140010448 A1, hereafter referred to as Lischinski), further in view of Kingma et al (Kingma, D. P., & Ba, J. (2014). Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, hereafter referred to as Kingma).
Claim 9
Regarding Claim 9, Chappell in view of Lischinski teaches The method of claim 1.
Chappell in view of Lischinski does not explicitly teach all of further comprising: training the CNN model using an Adaptive Moment Estimation algorithm (Adam) as an optimizer.
However, Kingma teaches further comprising:
training the CNN model using an Adaptive Moment Estimation algorithm (Adam) as an optimizer (Kingma in §§1-2, Algorithm 1 discloses Adam’s adaptive first and second gradient moment updates for a stochastic differentiable objective).
Kingma is analogous art because Kingma, like Chappell in view of Lischinski, is in the field of training image neural networks and is reasonably pertinent to the problem of efficient parameter updates for a differentiable CNN loss.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the supervised CNN parameter update workflow of Chappell in view of Lischinski with adaptive first and second gradient moment Adam updates of Kingma because Kingma teaches that adaptive first and second gradient moment Adam updates provides computationally efficient, low memory optimization of large, noisy stochastic objectives (§§1-2, Algorithm 1), and one of ordinary skill would have recognized that incorporating this feature into the method of Chappell in view of Lischinski would update the pupil/gaze model parameters with per parameter adaptive moment estimates, with a reasonable expectation of success because Chappel’s backpropagation training supplies differentiable loss gradients, and Adam consumes those gradients without changing the models image inputs or pupil/gaze outputs..
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claim(s) 11-12 is/are rejected under 35 U.S.C. 103 as obvious over Chappell et al (US 2021146177 A1, hereafter referred to as Chappell) and Lischinski et al (US 20140010448 A1, hereafter referred to as Lischinski), further in view of Walker et al (US 20150146026 A1, hereafter referred to as Walker).
Claim 11
Regarding Claim 11, Chappell in view of Lischinski teaches The method of claim 1.
Chappell in view of Lischinski does not explicitly teach all of further comprising: recording the image when compliant with a selected eye gaze direction.
However, Walker teaches further comprising:
recording the image when compliant with a selected eye gaze direction (Walker in ¶¶29, 51-53, 269-274, 285, 317-318 discloses buffering an acquired image for evaluation, treating subjects looking away from the camera as a lower quality condition, and saving or discarding that evaluated image).
Walker is analogous art because Walker, like Chappell in view of Lischinski, is in the field of computer image processing and image retention and is reasonably pertinent to the problem of recording an image where the subject looks toward the camera.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the pose gaze processing of Chappell in view of Lischinski with a buffered image evaluation and storage workflow of Walker because Walker teaches that a buffered image evaluation and storage workflow provides retention of desirable analyzed images and deleting/overwriting undesirable ones, with a subject looking away from the camera identified as a lower quality condition (¶¶29, 51-53, 269-274, 285, 317-318), and one of ordinary skill would have recognized that incorporating this feature into the method of Chappell in view of Lischinski would retain the image processed by the inherited brightness/CNN/pupil/gaze pipeline when its computed gaze meets the selected camera facing condition, with a reasonable expectation of success because Walker processes an acquired pixel array held in memory before deciding whether to save the array, Chappell supplies the gaze result for the subject, and Lischinski’s brightness correction and Chappell’s dimension reduction can be performed on that retained frame before the storage decision.
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claim 12
Regarding Claim 12, Chappell and Lischinski, further in view of Walker teaches The method of claim 11, further comprising:
providing audio or visual instructions for a live (real time) recording to bring the subject’s eye gaze into compliance with the selected eye gaze direction (Walker in ¶¶136, 139, 269, 285 discloses an audio speaker, audio prompts guiding image capture, a live acquisition/selective storage sequence, and look away as an undesirable image condition. Chappell in ¶¶15-22 discloses frame rate facial image acquisition and computed gaze results).
Claim(s) 14 is/are rejected under 35 U.S.C. 103 as obvious over Chappell et al (US 2021146177 A1, hereafter referred to as Chappell) and Lischinski et al (US 20140010448 A1, hereafter referred to as Lischinski), further in view of Ioffe et al (Ioffe, S., & Szegedy, C. (2015, June). Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International conference on machine learning (pp. 448-456). pmlr, hereafter referred to as Ioffe).
Claim 14
Regarding Claim 14, Chappell in view of Lischinski teaches The method of claim 1.
Chappell in view of Lischinski does not explicitly teach all of further comprising: normalizing the image using batch normalization after a first convolution layer and after the one or more dense layers as a part of the CNN model.
However, Ioffe teaches further comprising:
normalizing the image using batch normalization after a first convolution layer and after the one or more dense layers as a part of the CNN model (Ioffe in §§3.1-3.2 discloses batch normalization of convolutional and fully connected affine outputs before nonlinearities, with population statistics inference).
Ioffe is analogous art because Ioffe, like Chappell in view of Lischinski, is in the field of CNN image processing and training and is reasonably pertinent to the problem of unstable distributions at trainable convolutional and dense stages.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the convolution/fully connected image CNN stages of Chappell in view of Lischinski with batch normalization of convolutional and dense outputs of Ioffe because Ioffe teaches that batch normalization of convolutional and dense outputs provides more stable activation distributions and accelerated training while retaining network capacity (§§3.1-3.2), and one of ordinary skill would have recognized that incorporating this feature into the method of Chappell in view of Lischinski would normalize the feature data after the first convolution and after the dense hidden stages of Chappell’s image CNN architecture, with a reasonable expectation of success because Chappell discloses convolutional/fully connected image processing and supervised training, Ioffe’s differentiable transform preserves each stages shape and supplies fixed population statistics for inference, and the pupil-finder remains available to determine pupil locations.
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claim(s) 15-20 is/are rejected under 35 U.S.C. 103 as obvious over Mao et al (CN 113743254 A, hereafter referred to as Mao) in view of Ioffe et al (Ioffe, S., & Szegedy, C. (2015, June). Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International conference on machine learning (pp. 448-456). pmlr, hereafter referred to as Ioffe).
Claim 15
Regarding Claim 15, Mao teaches A method for training a CNN model to detect eye gaze direction, comprising:
pre-processing a training image, the training image comprising a known eye pupil location or a known eye gaze direction (Mao in ¶¶[n0029]-[n0038], [n0053]-[n0058] discloses preparing cropped eye samples with pupil/eye contour position annotations and known gaze vector annotations);
inputting the training image to the CNN model (Mao in ¶¶[n0059]-[n0061] discloses inputting the annotated sample eye images to the neural model);
…
producing an output from the CNN model, wherein the output comprises an eye pupil location output or an eye gaze direction output of the training image (Mao in ¶¶[n0042]-[n0043], [n0059]-[n0063] discloses two model outputs being a spatial eye/pupil location heatmap and an eye gaze vector);
comparing the eye pupil location output to the known eye pupil location or the eye gaze direction output to the known eye gaze direction, wherein a difference in the eye pupil location output to the known eye pupil location and the eye gaze direction output to the known eye gaze direction comprises an error amount (Mao in ¶¶[n0062]-[n0063], [n0068] discloses comparing predicted eye/pupil spatial labels against annotated contours using cross entropy and predicted gaze against the annotated gaze vector using mean square error, then summing both losses);
adjusting internal parameters of the CNN model based at least in part on the error amount (Mao in ¶[n0063] discloses adjusting the network parameters when the joint loss has not converged); and
creating a training CNN model having adjusted internal parameters (Mao in ¶¶[n0063]-[n0065] discloses obtaining the trained gaze estimation network after convergence with adjusted parameters).
Mao does not explicitly teach all of processing the training image through the CNN model, the CNN model comprising: one or more convolution layers; one or more fully connected layers; one or more activation functions; and one or more batch normalizations.
However, Ioffe teaches processing the training image through the CNN model (Ioffe in §§3.2, 4.2 discloses image feature stages in a batch normalized image CNN), the CNN model comprising:
one or more convolution layers (Ioffe in §§3.2, 4.2 discloses convolutional image feature stages in a batch normalized image CNN);
one or more fully connected layers (Ioffe in §§3.2, 4.1-4.2 discloses fully connected affine stages and the top fully connected output of its image CNN);
one or more activation functions (Ioffe in §§3.2, 4.2 discloses ReLU nonlinearities in the convolutional image model stages); and
one or more batch normalizations (Ioffe in §§3-3.2, 4.2, Algorithms 1-2 discloses differentiable mini-batch normalization in convolutional image network stages);
Ioffe is analogous art because Ioffe, like Mao, is in the field of training image neural networks and is reasonably pertinent to the problem of supplying trainable image feature/output stages with stable activation distributions during joint loss training.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the feature extraction and gaze output architecture of Mao with the convolutional/fully connected image network stages, nonlinear activations and differentiable batch normalization of Ioffe because Ioffe teaches that a batch normalized image network architecture provides stable activation distributions and accelerated training while preserving trainable representation capacity (§§3-3.2, 4.2, Algorithms 1-2), and one of ordinary skill would have recognized that incorporating this feature into the method of Mao would supply a BN/ReLU convolutional feature trunk and a fully connected affine gaze vector head while retaining Mao’s spatial heatmap branch and summed pupil location/gaze losses, with a reasonable expectation of success because Mao supplies both output tensors and differentiable losses, Ioffe’s transforms preserve feature shape and propagate gradients, and the terminal fully connected output can be retrained to Mao’s gaze vector dimension.
Thus, the claimed subject matter would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention.
Claim 16
Regarding Claim 16, Mao in view of Ioffe teaches The method of claim 15, further comprising:
validating the trained CNN model using a test set of images with known parameters (Ioffe in §§4.1-4.2.2 discloses held-out labeled image test/validation evaluation; Mao in ¶¶[n0053]-[n0058] discloses the pupil region/gaze annotations used as known parameters for the proposed held-out eye image set).
Claim 17
Regarding Claim 17, Mao in view of Ioffe teaches The method of claim 15, further comprising:
comparing the eye gaze determination to ground truth data associated with the training image to determine accuracy of the CNN model (Mao in ¶¶[n0056]-[n0057], [n0062]-[n0063], [n0068] discloses using annotated ground truth gaze vector values and mean squared gaze error to assess prediction accuracy).
Claim 18
Regarding Claim 18, Mao in view of Ioffe teaches The method of claim 15, wherein adjusting the internal parameters of the CNN model comprises minimizing errors in the eye gaze determination (Mao in ¶¶[n0062]-[n0064], [n0068] discloses adjusting parameters to reduce the combined loss including the gaze vector mean squared error term until convergence).
Claim 19
Regarding Claim 19, Mao in view of Ioffe teaches The method of claim 15, wherein the pre-processing of the training image comprises:
extracting facial feature points (Mao in ¶¶[n0029]-[n0038], [n0058] discloses extracting eye contour landmark coordinates from the training face image before cropping).
Claim 20
Regarding Claim 20, Mao in view of Ioffe teaches The method of claim 19, wherein the pre-processing of the training image further comprises:
splitting the training image into one or more sub-images (Mao in [n0019], [n0032]-[n0039], [n0055], [n0058] discloses deriving left/right eye region images and cropped annotated eye samples from face training images).
Allowable Subject Matter
Claim 8 is objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JUSTIN P CASCAIS whose telephone number is (703) 756-5576. 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, Mr. O'Neal Mistry can be reached on (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 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.
/J.P.C./Examiner, Art Unit 2674
/ONEAL R MISTRY/Supervisory Patent Examiner, Art Unit 2674
Date: 10/9/2026