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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 12th February, 2026 has been entered.
Response to Amendment
This action is in response to the amendment filed on 12th February, 2026. Claims 1, 4-6, 8, 11-13, 15, and 18-20 have been amended. Claims 1-20 remain rejected in the application.
Response to Arguments
Applicant's arguments with respect to Claims 1, 8, and 15 filed on 12th February, 2026, with respect to the rejection under 35 U.S.C. § 103, regarding that the prior art does not teach the limitation(s): "inputting the color input image and the opacity input image corresponding to the current frame, and a previous intermediate color output image and a previous intermediate opacity output image generated for a prior frame, into a trained neural network configured to generate an intermediate color output image and an intermediate opacity output image corresponding to the current frame, the intermediate color output image and the intermediate opacity output image having a same pixel resolution as the color input image and the opacity input image", "rendering an intermediate hair image based on a combination of the intermediate color output image and the intermediate opacity output image", and "rendering a final hair image based on a combination of the intermediate hair image and non-hair rendered components of the current frame" have been fully considered, but are moot because of new grounds for rejection. It has now been taught by the combination of Currius and Marks.
Regarding arguments to Claims 2-7, 9-14, and 16-20, they directly/indirectly depend on independent Claims 1, 8, and 15 respectively. Applicant does not argue anything other than independent Claims 1, 8, and 15. The limitations in those claims, in conjunction with combination, was previously established as explained.
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 1-2, 4-9, 11-16, and 18-20 are rejected under 35 U.S.C. 103 as being unpatentable over Currius et al. ("Real-Time Hair Filtering with Convolutional Neural Networks", previously cited), hereinafter referenced as Currius, in view of Marks et al. (US 20230112302 A1, previously cited), hereinafter referenced as Marks.
Regarding Claim 1, Currius discloses a method for hair rendering (Currius, [Section 3.1 Overview]: teaches a method for hair rendering using stochastic transparency and a trained neural network), the method comprising:
acquiring a color input image and an opacity input image corresponding to a current frame from a hair rendering system (Currius, [Section 3.3.1 Input and Output]: teaches the input features for the neural network being the color factor <read on color input image>, specular factor, the alpha (transparency) value <read on opacity input image>, the screen-space depth of the sample, and the view-space tangent as shown in FIG. 2; [Section 3.3.3 Training and Validation]: teaches producing multi-sampled images with different features <read on current frame> that the neural network receives),
PNG
media_image1.png
312
759
media_image1.png
Greyscale
the color input image and the opacity input image corresponding to the current frame having a first sample resolution (Currius, [Section 3.2 Input Rendering]: teaches obtaining more than one sample per pixel <read on first sample resolution> of an input <read on color and opacity input images corresponding to current frame>, the neural network uses a coverage mask for a multi-sample buffer);
inputting the color input image and the opacity input image corresponding to the current frame, and a previous intermediate color output image and a previous intermediate opacity output image generated for a prior frame, into a trained neural network (Currius, [Section 3.1 Overview]: teaches training a U-net <read on trained neural network> to reconstruct an input image using a few samples with stochastic transparency, where color <read on color input image> and additional features, such as tangents and depth <read on opacity input image>, are used as input for the neural network; [Section 3.3.4 Inference]: teaches rendering the input features into OpenGL multi-sampled color buffers <read on previous intermediate color and opacity output images for prior frame>, then move the result to cuDNN tensors <read on during rendering of current frame>, where the convolutional network is applied, and then the resulting tensors are copied back to an OpenGL texture to be composed into the final image) configured to
generate an intermediate color output image and an intermediate opacity output image corresponding to the current frame (Currius, FIG. 2 teaches using a CNN to filter stochastically sampled color factor, highlight, alpha, depth, and tangents to obtain filtered color factor <read on intermediate color output image>, highlight, and alpha <read out intermediate opacity output image>; [Section 3.1 Overview]: teaches the trained network denoising <read on performing anti-aliasing function> novel views and different hair styles with high quality results using color factor <read on color input image>, tangents, depth, highlight, and alpha <read on opacity input image> as input to the neural network),
PNG
media_image2.png
312
759
media_image2.png
Greyscale
the intermediate color output image and the intermediate opacity output image having a same pixel resolution as the color input image and the opacity input image (Currius, [Section 3.3.3 Training and Validation]: teaches "the target training data is composed of images rendered at very high resolution and down-sampled to the same size <read on same pixel resolution> as the input <read on color input and opacity images>" and "the hair translucency <read on opacity output image> is approximated by averaging the images <read on intermediate color and opacity output images> rendered with stochastic transparency in the super-sampled resolution, until converged");
rendering an [[intermediate]] hair image based on a combination of the intermediate color output image and the intermediate opacity output image (Currius, FIG. 2 teaches using a CNN to filter stochastically sampled color factor, highlight, alpha, depth, and tangents to obtain filtered color factor <read on intermediate color output image of hair image>, highlight, and alpha <read out intermediate opacity output image of hair image>, which are then composited <read on combination> to produce the final image); and
rendering a final hair image based on a combination of the intermediate hair image [[and non-hair rendered components]] of the current frame (Currius, FIG. 2 teaches compositing sampled images <read on intermediate hair image> of the hair image to produce a final image <read on render final image> of the hair).
PNG
media_image3.png
312
759
media_image3.png
Greyscale
However, Currius does not expressly disclose
rendering an intermediate hair image based on a combination of the intermediate color output image and the intermediate opacity output image; and
rendering a final hair image based on a combination of the intermediate hair image and non-hair rendered components of the current frame.
Marks discloses
rendering an intermediate hair image based on a combination of the intermediate color output image and the intermediate opacity output image (Marks, [0056]: teaches a "masked hair image 330 <read on intermediate hair image> is reconstructed from the image 102 using a hair-manipulation algorithm," which uses extracted features 322 that correspond to hair from image 102); and
rendering a final hair image based on a combination of the intermediate hair image and non-hair rendered components of the current frame (Marks, [0057]: teaches "the combination of the masked rendered face image 318 <read on non-hair rendered components> and the masked hair image 330 <read on intermediate hair image> generate a combined synthetic face image 332 from the face in the image 102," which is then refined to generate a photorealistic refined synthetic image face 336 <read on final hair image> as shown in FIG. 3; Note: it should be noted that non-hair rendered components can be interpreted as face portions and/or hair accessories).
PNG
media_image4.png
409
491
media_image4.png
Greyscale
Marks is analogous art with respect to Currius because they are from the same field of endeavor, namely utilizing machine learning for improved hair rendering output. Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to utilize a segmentation network to further segment the input image into a plurality of segmented layers as taught by Marks into the teaching of Currius. The suggestion for doing so would allow specific layers, such as the hair layer, to be modified and improved, which can then be used to train a neural network to obtain improved and desired results. Therefore, it would have been obvious to combine Marks with Currius.
Regarding Claim 8, it recites the limitations that are similar in scope to Claim 1, but in an apparatus. As shown in the rejection, the combination of Currius and Marks discloses the limitations of Claim 1. Additionally, Currius discloses an apparatus for hair rendering, the apparatus (Currius, [Section 4 Results]: teaches running experiments for a hair rendering system with neural networks on an Nvidia RTX 2080 GPU <read on apparatus>) comprising:…
Thus, Claim 8 is met by Currius according to the mapping presented in the rejection of Claim 1, given the method corresponds to an apparatus.
Regarding Claim 15, it recites the limitations that are similar in scope to Claim 1, but in a non-transitory computer-readable storage medium. As shown in the rejection, the combination of Currius and Marks discloses the limitations of Claim 1. Additionally, Currius discloses a non-transitory computer-readable storage medium storing computer-readable instructions thereon, which, when executed by processing circuitry, cause the processing circuitry to perform a method for hair rendering (Currius, [Section 4 Results]: teaches running experiments for a hair rendering system with neural networks on an Nvidia RTX 2080 GPU <read on processing circuitry>, where convolution parameters and data tensors are stored <read on non-transitory computer-readable storage medium> as NHWC, and the network uses the tensor cores of the CPU for acceleration <read on computer-readable instructions>), the method comprising:…
Thus, Claim 15 is met by Currius according to the mapping presented in the rejection of Claim 1, given the method corresponds to a non-transitory computer-readable storage medium.
Regarding Claims 2, 9, and 16, the combination of Currius and Marks discloses the method, the apparatus, and the non-transitory computer-readable storage medium of Claims 1, 8, and 15 respectively. Additionally, Currius further discloses wherein
the trained neural network is trained using training images having a second sample resolution that is greater than the first sample resolution (Currius, [Section 3.3.3 Training and Validation]: teaches "the target training
data <read on second sample resolution> is composed of images <read on training images> rendered at very high resolution <read on second sample resolution being greater than first sample resolution> and down-sampled to the same size as the input").
Regarding Claims 4, 11, and 18, the combination of Currius and Marks discloses the method, the apparatus, and the non-transitory computer-readable storage medium of Claims 1, 8, and 15 respectively. Additionally, Currius further discloses wherein
the previous intermediate color output image and the previous intermediate opacity output image are generated by the trained neural network for the prior frame (Currius, [Section 3.2 Input Rendering]: teaches using a multi-sample buffer to average input samples <read on providing previous intermediate color output image and previous intermediate opacity output image output> for the neural network; Note: it should be noted that although not expressly stated, one skilled in the art would understand that performant real-time frame generation is a multitasked process; in addition, a frame buffer stores historical frames, where prior frames are used as context for advanced image processing, such as image sharpening and anti-aliasing), and
the trained neural network generates the intermediate color output image and the intermediate opacity output image for the current frame based on(i) the color input image and the opacity input image acquired for the current frame (Currius, [Section 3.3.1 Input and Output]: teaches the input features for the neural network to render the current frame being the color factor <read on color input image>, specular factor, the alpha (transparency) value <read on opacity input image>, the screen-space depth of the sample, and the view-space tangent as shown in FIG. 2), and
(ii) the previous intermediate color output image and the previous intermediate opacity output image output by the trained neural network for the prior frame (Currius, [Section 3.3.4 Inference]: teaches using a multi-sample color buffer for real-time hair rendering that utilizes a neural network; [Section 3.2 Input Rendering]: teaches using a multi-sample buffer <read on previous intermediate color and opacity output images> to average input samples for the neural network to reduce temporal noise, which requires data from previous/prior frames).
Regarding Claims 5, 12, and 19, the combination of Currius and Marks discloses the method, the apparatus, and the non-transitory computer-readable storage medium of Claims 1, 8, and 15 respectively. Additionally, Currius further discloses
performing deferred hair rendering by, during rendering of the current frame, providing a color input image and an opacity input image corresponding to the prior frame to the trained neural network (Currius, [Section 3.3.1 Input and Output]: teaches the input features for the neural network to render <read on deferred hair rendering> the current frame being the color factor <read on color input image>, specular factor, the alpha (transparency) value <read on opacity input image>, the screen-space depth of the sample, and the view-space tangent as shown in FIG. 2; [Section 3.3.4 Inference]: teaches rendering the input features into OpenGL multi-sampled color buffers <read on prior frame preceding current frame>, then move the result to cuDNN tensors <read on during rendering of current frame>, where the convolutional network is applied, and then the resulting tensors are copied back to an OpenGL texture to be composed into the final image); and
in response to the trained neural network outputting the intermediate color output image and the previous intermediate opacity output image corresponding to the prior frame (Currius, FIG. 2 teaches using a CNN to filter stochastically sampled color factor, highlight, alpha, depth, and tangents to obtain filtered color factor <read on intermediate color output image>, highlight, and alpha <read out intermediate opacity output image>),
rendering the current frame [[by combining the non-hair rendered components of the current frame with the final rendered hair image]] based on the previous intermediate color output image and the previous intermediate opacity output image corresponding to the prior frame (Currius, FIG. 2 teaches using a CNN to filter stochastically sampled color factor, highlight, alpha, depth, and tangents to obtain filtered color factor <read on intermediate color output image>, highlight, and alpha <read out intermediate opacity output image>, which are then composited to produce the final image <read on generate final rendered hair image>; [Section 3.2 Input Rendering]: teaches using a multi-sample buffer <read on previous intermediate color and opacity output images> to average input samples for the neural network to reduce temporal noise, which requires data from previous/prior frames).
However, Currius does not expressly disclose
rendering the current frame by combining the non-hair rendered components of the current frame with the final rendered hair image based on the previous intermediate color output image and the previous intermediate opacity output image corresponding to the prior frame.
Marks discloses
rendering the current frame by combining the non-hair rendered components of the current frame with the final rendered hair image based on the previous intermediate color output image and the previous intermediate opacity output image corresponding to the prior frame (Marks, [0057]: teaches "the combination of the masked rendered face image 318 <read on non-hair rendered components> and the masked hair image 330 generate a combined synthetic face image 332 from the face in the image 102," which is then refined to generate a photorealistic refined synthetic image face 336 <read on final hair image> as shown in FIG. 3).
Marks is analogous art with respect to Currius because they are from the same field of endeavor, namely utilizing machine learning for improved hair rendering output. Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to utilize a segmentation network to further segment the input image into a plurality of segmented layers as taught by Marks into the teaching of Currius. The suggestion for doing so would allow specific layers, such as the hair layer, to be modified and improved, which can then be used to train a neural network to obtain improved and desired results. Therefore, it would have been obvious to combine Marks with Currius.
Regarding Claims 6, 13, and 20, the combination of Currius and Marks discloses the method, the apparatus, and the non-transitory computer-readable storage medium of Claims 5, 12, and 19 respectively. Additionally, Currius further discloses wherein the performing the deferred hair rendering further comprises,
during rendering of the current frame, generating the color input image and the opacity input image corresponding to the current frame (Currius, [Section 3.3.4 Inference]: teaches using a multi-sample color buffer <read on during rendering of current frame> for real-time hair rendering that utilizes a neural network; [Section 3.3.1 Input and Output]: teaches the input features for the neural network being the color factor <read on color input image>, specular factor, the alpha (transparency) value <read on opacity input image>, the screen-space depth of the sample, and the view-space tangent as shown in FIG. 2), and
a color input image and an opacity input image corresponding to the rendered current frame are provided to the trained neural network during rendering of a next frame after the current frame (Currius, [Section 3.3.3 Training and Validation]: teaches "the input training data is obtained by rendering the hair with stochastic transparency" using input features <read on generated color input image and opacity input image corresponding to current frame>; [Section 3.3.4 Inference]: teaches rendering the input features into OpenGL multi-sampled color buffers, where "the result is moved to cuDNN tensors and the convolutional network is applied <read on rendering a next frame after current frame>"; Note: it should be noted that this neural network is to be used for real-time applications, such as video games, where real-time rendering is required).
Regarding Claims 7 and 14, the combination of Currius and Marks discloses the method and the apparatus of Claims 1 and 8 respectively. Additionally, Currius further discloses wherein
the color input image, the opacity input image, the intermediate color output image, and the intermediate opacity output image corresponding to the current frame include only hair data (Currius, [Section 3.3.1 Input and Output]: teaches the input features for the neural network being the color factor <read on color input image>, specular factor, the alpha (transparency) value <read on opacity input image>, the screen-space depth of the sample, and the view-space tangent; FIG. 2 teaches obtaining a filtered color factor <read on intermediate color output image>, highlight, and alpha <read on intermediate opacity output image> which depicts only hair), wherein
[[the hair data is pre-segmented by the hair rendering system prior to being input into the trained neural network.]]
However, Currius does not expressly disclose
the hair data is pre-segmented by the hair rendering system prior to being input into the trained neural network.
Marks discloses
the hair data is pre-segmented by the hair rendering system prior to being input into the trained neural network (Marks, FIG. 3 teaches the hair generator 324 <read on hair rendering system> generating hair segment 328 <read on pre-segmented hair data> which is then combined with the face image before being refined by the refiner neural network <read on trained neural network>).
Marks is analogous art with respect to Currius because they are from the same field of endeavor, namely utilizing machine learning for improved hair rendering output. Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to utilize a segmentation network to further segment the input image into a plurality of segmented layers as taught by Marks into the teaching of Currius. The suggestion for doing so would allow specific layers, such as the hair layer, to be modified and improved, which can then be used to train a neural network to obtain improved and desired results. Therefore, it would have been obvious to combine Marks with Currius.
Claims 3, 10, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Currius et al. ("Real-Time Hair Filtering with Convolutional Neural Networks", previously cited), hereinafter referenced as Currius, in view of Marks et al. (US 20230112302 A1, previously cited), hereinafter referenced as Marks as applied to Claims 1, 8, and 15 above respectively, and further in view of Luebke et al. (US 20180096516 A1, previously cited), hereinafter referenced as Luebke.
Regarding Claims 3, 10, and 17, the combination of Currius and Marks discloses the method, the apparatus, and the non-transitory computer-readable storage medium of Claims 2, 9, and 16 respectively. The combination of Currius and Marks does not expressly disclose the limitations of Claims 3, 10, and 17; however, Luebke discloses wherein
the first sample resolution is 1 sample per pixel (Luebke, [0109]: teaches using a ray cast to sample ray cast hit points, where temporal anti-aliasing techniques "may rely on per-sample <read on first sample resolution being 1 sample per pixel> or per-pixel velocity vectors"),
the second sample resolution is n samples per pixel, where n is an integer greater than 1 (Luebke, [0103]: teaches "multiple samples may be explicitly tracked and incorporated per pixel <read on second sample resolution>," where "the pixel may first be divided into a fixed number of subpixels <read on n samples per pixel>, and then a target range of samples <read on n being an integer greater than 1> may be computed per pixel by, for example, examining the local contrast in the previous frame"; [0135]: teaches when utilizing stable ray tracing, "it may be adaptively decided whether to re-use a given shade or compute it afresh," which "is an improvement on temporal AA, which may always perform one shade per pixel and decide whether to blend it with the previously computed shade from prior frames"), and
each sample represents a ray cast toward a pixel corresponding to the respective sample (Luebke, [0109]: teaches using a ray cast to sample ray cast hit points, where temporal anti-aliasing techniques "may rely on per-sample or per-pixel velocity vectors, but such vectors may be imprecise (e.g., movement of points on a spinning tire cannot be predicted by a simple vector)").
Luebke is analogous art with respect to Currius, in view of Marks because they are from the same field of endeavor, namely high quality and stable anti-aliasing techniques. Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to implement stable ray tracing as taught by Luebke into the teaching of Currius, in view of Marks. The suggestion for doing so would resolve the issue of temporal stability, thereby resulting in sharp image quality of hair rendered in real-time. Therefore, it would have been obvious to combine Luebke with Currius, in view of Marks.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Aluru et al. (US 20200279440 A1) discloses a virtual representation creation and display for generating virtual hairstyles;
Fanello et al. (US 20230419600 A1) discloses techniques for volumetric performance capture with neural rendering;
Chai et al. ("Neural Hair Rendering") discloses a neural-based hair rendering pipeline that synthesizes photo-realistic images from virtual 3D hair models;
Rosu et al. ("Neural Strands: Learning Hair Geometry and Appearance from Multi-View Images") discloses a learning framework for modeling accurate hair geometry and appearance from multi-view image inputs; and
Wang et al. ("HVH: Learning a Hybrid Neural Volumetric Representation for Dynamic Hair Performance Capture") discloses a volumetric hair representation that is composed of thousands of primitives, where each primitive is rendered efficiently and realistically.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KARL TRUONG whose telephone number is (703)756-5915. The examiner can normally be reached 10:30 AM - 7:30 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, Kent Chang can be reached at (571) 272-7667. 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.
/K.D.T./Examiner, Art Unit 2614
/KENT W CHANG/Supervisory Patent Examiner, Art Unit 2614
Prosecution Insights