Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment
Claims 1, 3-4, 10-11, 14, and 17 have been amended.
Claim 2 has been cancelled.
Claims 1-20 are still pending for consideration.
Response to Arguments
Applicant's arguments filed on 07/07/2026 have been fully considered but they are not persuasive.
Applicant on page 3 of the “Remark” asserts “The cited arts do not teach selecting a diffusion timestep based on a first quality level representing a degree of degradation”.
The argument is persuasive to the extent Applicant contends that Nie does not expressly characterize the input condition using the term “first quality level”. However, the rejection does not rely on Nie alone for that feature. Nie teaches selecting diffusion timestep t*, on para [0060] Nie disclose “a diffusion network can first add noise to adversarial input utilizing a forward SDE process 212 with a small diffusion timestep….. a process can utilize a selected diffusion timestep, which represents an amount of noise added”, and further on para [0064] disclose “a small diffusion timestep t* can be selected to obtain a diffused image… t* can be selected on a normalized scale that runs from t=0 to t=1, as illustrated in FIG. 2, where an image input at t=0 would be diffused to a point of pure Gaussian noise at time t=1”. Nie on para [0071] additionally disclose “a reverse-time SDE can be solved from timestep t=t* using diffused adversarial sample x(t*) as an initial value”. Zhang teaches characterizing an input image according to an image-qulity/degradation level. It would have been obvious to a person of ordinary skill in the art to select Nie’s diffusion timestep in accordance with the known quality/degradation level of the input image as taught by Zhang, because Nie already teaches selecting the timestep according to the extent of undesirable perturbation to be removed while maintaining image semantics.
Applicant further argues that Nie and Zhang fail to teach wherein the image generation model is trained to restore images depicting the entity using training data including a plurality of training images depicting the entity. The Examiner acknowledges that the prior combination did not expressly disclose the newly added limitation. However, newly found Aberman et al. (US 20240296596 A1) as listed below teaches the limitation. Accordingly, Aberman cures the deficiency identified by applicant.
Claim Rejections - 35 USC § 112
Claim 10 is rejected under 35 USC § 112 (b) as indefinite because the limitation “the training image” lacks sufficient antecedent basis.
Applicant amended the beginning of claim 10 from a singular “training image” to “a plurality of training images,” but later the claim still says, “wherein the image generation model is trained using the noisy image, the training image, and the guidance information…”.
As amended, “the training image” no longer appears to have antecedent basis. The antecedent is now “a plurality of training images.” it is unclear whether “the training image” means one selected image from the plurality, every image of the plurality, or something different.
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 text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claims 1, 3-4, and 9 are rejected under 35 U.S.C. 103 as being unpatentable over Nie et al. (US 20240104698 A1) in view of Zhang et al. (US 20220130017 A1), and further in view of Aberman et al. (US 20240296596 A1).
Regarding claim 1, Nie et al. teaches a method comprising: obtaining an input image depicting an entity (see para [0080]; “an image can be received 402 (or otherwise obtained)”, see also para [0056]; “such as input image 102 including a representation of at least one object”); adding noise to the input image by selecting a diffusion timestep based on the first quality level to obtain an intermediate noise image (see para [0060]; “a diffusion network can first add noise to adversarial input utilizing a forward SDE process 212 with a small diffusion timestep….. a process can utilize a selected diffusion timestep, which represents an amount of noise added”, see also para [0064]; “a small diffusion timestep t* can be selected to obtain a diffused image”); and generating, using an image generation model, a restored image depicting the entity by denoising from the intermediate noise image starting at the selected diffusion timestep (see para [0060]; “a reverse SDE process 214 can be utilized to recover, or reconstruct, a clean or purified image from a diffused adversarial image 204”, see also para [0071]; “a reverse-time SDE can be solved from timestep t=t* using diffused adversarial sample x(t*) as an initial value to get a final solution {circumflex over (x)}(0) of SDE”). However, Nie et al. does not teach and having a first quality level wherein the restored image has a second quality level higher than the first quality level, wherein the image generation model is trained to restore images depicting the entity using training data including a plurality of training images depicting the entity.
In the same field of endeavor, Zhang et al. teaches and having a first quality level (see claim 2; “acquiring a first medical image having a first image quality, wherein the first medical image is an MR medical image of a subject”), wherein the restored image has a second quality level higher than the first quality level (see claim 2; “wherein the second medical image has a second image quality, and wherein the second image quality is higher than the first image quality”, see also claim 10; “wherein the second image quality has greater SNR, higher resolution, or less aliasing compared with the first image quality”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify the general use of techniques presented to remove unintended variations introduced into data of Nie rt al. in view of systems that can address various drawbacks of conventional systems of Zhang in order to improving image quality with shortened acquisition time (see claim 2). However, the combination of Nie et al. and Zhang et al. does not teach wherein the image generation model is trained to restore images depicting the entity using training data including a plurality of training images depicting the entity.
In the same field of endeavor, Aberman et al. teaches wherein the image generation model is trained to restore images depicting the entity using training data including a plurality of training images depicting the entity (see para [0043]; “The fine-tuning system 120 obtains the custom image dataset 104 and fine-tunes the pre-trained text-to-image model 102 on the obtained custom image dataset 104. The custom image dataset 104 includes a plurality of images that each depict a same particular subject instance of an object class”, see also para [0050]; “3-5 images, that each depict the particular subject instance”, and para [0075]; “The system generates a custom image dataset from a plurality of images that each depict a same particular subject instance (step 202)”, and para [0081]; “The system trains a text-to-image model on the custom image dataset”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify the general use of techniques presented to remove unintended variations introduced into data of Nie rt al. in view of systems that can address various drawbacks of conventional systems of Zhang and further in view of a method that trains a text-to-image model to generate an image conditioned on a text input of Aberman et al. in order to retrieve a specific subject instance from the model (see para [0043]).
Regarding claim 3, the rejection of claim 1 is incorporated herein.
Nie et al. in the combination further teach wherein generating the restored image comprises: iteratively removing noise from the intermediate noise image based on the selected timestep (see para [0059]; “a generative process in which a version of this image is generated that has at least some noise removed relative to a prior image in this noise removal sequence. In at least one embodiment, these iterations can continue to be performed until this process arrives at a “purified” image 206 that is substantially free of noise”).
Regarding claim 4, the rejection of claim 1 is incorporated herein.
Nie et al. in the combination further teach wherein: the selected timestep is based on the first quality level of the input image (see para [0060]; “such a process can utilize a selected diffusion timestep, which represents an amount of noise added during forward process 212. In at least one embodiment, this amount of noise is selected to be high enough to remove adversarial perturbations but not so high as to destroy label semantics, or other such features or aspects, of purified images”).
Regarding claim 9, the rejection of claim 1 is incorporated herein.
Zhang et al. in the combination further teaches wherein: the restored image has a higher image quality than the input image (see claim 2; “applying a deep network model to the first medical image to generate a second medical image, wherein the second medical image has a second image quality, and wherein the second image quality is higher than the first image quality”).
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Nie et al. and Zhang et al. in view of Aberman et al. as applied in claim 1 above, and further in view of Gandelsman et al. (US 20240161462 A1.)
Regarding claim 5, the rejection of claim 1 is incorporated herein. The combination Nie et al., Zhang et al. and Aberman et al. as a whole does not teach wherein: the image generation model has a constrained latent space based on training using at least one training image depicting the entity.
In the same field of endeavor, Gandelsman et al. teach wherein: the image generation model has a constrained latent space based on training using at least one training image depicting the entity (see para [0004]; “a pre-trained diffusion model is tuned based on a single target image so that the output of the model consistently maintains similarities to the target image”, see also para [0006]; “fine-tuning a pre-trained diffusion model based on a single image to obtain a tuned diffusion model”, and [0078]; “forward process for a latent diffusion model, the model maps an observed variable x.sub.0 (either in a pixel space or a latent space) intermediate variables x.sub.1, . . . , x.sub.T using a Markov chain”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify the general use of techniques presented to remove unintended variations introduced into data of Nie et al. in view of systems that can address various drawbacks of conventional systems of Zhang and further in view of a method that trains a text-to-image model to generate an image conditioned on a text input of Aberman et al. and image generation using machine learning of Gandelsman et al. in order to generate a modified image based on the image and the encoded text prompt (see para [0004]).
Claims 6-8 are rejected under 35 U.S.C. 103 as being unpatentable over Nie et al. and Zhang et al. in view of Aberman et al. as applied in claim 1 above, and further in view of Varanka et al. NPL “PFStorer: Personalized Face Restoration and Super-Resolution”.
Regarding claim 6, the rejection of claim 1 is incorporated herein. The combination Nie et al., Zhang et al. and Aberman et al. as a whole does not teach wherein: the restored image is generated without providing an image as guidance to an intermediate stage of the image generation model.
In the same field of endeavor, Varanka et al. teaches wherein: the restored image is generated without providing an image as guidance to an intermediate stage of the image generation model (see page 2375, Fig. 3; “During training the reference image is randomly sampled from a set of reference images for each training iteration. During inference, no reference images are required as the identity is learned in the personalization blocks as a neural representation”). Accordingly, it would have been obvious to one of ordinary skill inthe art before the effective filling date to the claimed invention to modify the general use of techniques presented to remove unintended variations introduced into data of Nie rt al. in view of systems that can address various drawbacks of conventional systems of Zhang and further in view of a method that trains a text-to-image model to generate an image conditioned on a text input of Aberman et al. and personalized face restoration and super-resolution of Varanka et al. in order to improve the training pipeline of face restoration models to enable an alignment-free approach (see page 2375, Fig. 3).
Regarding claim 7, the rejection of claim 1 is incorporated herein.
Varanka et al. in the combination further teaches further comprising: generating a synthetic image depicting the entity, wherein the image generation model is trained based on the synthetic image (see page 2375, 3.2; “During the personalization fine-tuning, the model takes as input a synthesized LQ image ILQ”, see also page 2377, left col. 3rd para; “Synthetic Noise Generation In order to generate LQ images for training,”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify the general use of techniques presented to remove unintended variations introduced into data of Nie rt al. in view of systems that can address various drawbacks of conventional systems of Zhang and further in view of a method that trains a text-to-image model to generate an image conditioned on a text input of Aberman et al. and personalized face restoration and super-resolution of Varanka et al. in order to generate high-quality image while mostly retaining the identity (see page 2375, 3.2).
Regarding claim 8, the rejection of claim 1 is incorporated herein.
Varanka et al. in the combination further teaches wherein: the restored image preserves an identity of the entity from the input image (see page 2378, 4.1; “CodeFormer is able to output a high-quality image while mostly retaining the identity. Our method is able to retain even small details such as the wrinkles and skin texture”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify the general use of techniques presented to remove unintended variations introduced into data of Nie rt al. in view of systems that can address various drawbacks of conventional systems of Zhang and further in view of a method that trains a text-to-image model to generate an image conditioned on a text input of Aberman et al. and personalized face restoration and super-resolution of Varanka et al. in order to generate both the qualitative results and quantitative metrics (see page 2378, 4.1).
Claims 10-11, and 14-16 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et al. (US 20240161327 A1) in view of Aberman et al. (US 20230325998 A1) herein after Aberman2
Regarding claim 10, Chen et al. teaches a method for training a machine learning model, comprising: obtaining a training set including a plurality of training images depicting an entity (see para [0125]; “a training image can be identified, where the training image includes at least one object. A training dataset can include a plurality of high-resolution training images of variable sizes. The training image(s) can include a ground truth, where the ground truth indicates an image with a predefined resolution and scale” Note: depicting an entity implies object ); generating a noisy image and guidance information based on the plurality of training images (see para [0006]; “obtaining a noise map and a global image code encoded from a training image and representing semantic content of the training image”, see also para [0103]; “noise can be added to the cropped patch, p.sub.c0, 1025 to create a noisy training patch, p.sub.t, 1030. The noise can be added to the cropped patch, p.sub.c0, 1025 by a diffusion model”, and para [0081]; “the encoder, E, 760 encodes the image 605 into a latent representation, global image code, z”, and para [0107]; “The local diffusion model can follow these guidance channels to generate correct, globally-consistent local patches”, Note: global image code is encoded from the training image to represent its semantic content, acting as the conditioning/guidance); and training an image generation model (see claim 9; “training the diffusion model to generate the plurality of image patches”), wherein the image generation model is trained using the noisy image, the training image, and the guidance information (see para [0006]; “obtaining a noise map and a global image code encoded from a training image… generating a plurality of predicted image patches based on the noise map and the global image code… and training the diffusion model to generate image patches by updating the parameters based on the loss function”, Note: the neural network looks at the noisy input and attempts to predict the noise or clean the patches at each step, using the guidance (global code)). Chen et al. does not explicitly disclose to generate a restored image depicting the entity based on an input image depicting the entity, to follow a constrained denoising path corresponding to a subspace of a generative space of the image generation model corresponding to the plurality of training images depicting the entity.
In the same field of endeavor, Aberman2 teaches to generate a restored image depicting the entity based on an input image depicting the entity (see para [0066]; “that the generative model will be tasked with producing an output image that resembles the input image but has improved resolution, each of input images”, see also para [0003]; “the subject's identifying features will be reflected in the images the model produces), to follow a constrained denoising path corresponding to a subspace of a generative space of the image generation model (see para [0029]; “Personal priors 402 and 403 each represent a vector space within latent vector space W (302) including a subset of possible input codes that produce images that resemble a given subject”, see also Abstract; “using the personalized prior to confine the inputs of a generative model to a latent vector space associated with the given subject”, and para [0069]; “the generative model where it is confined to choosing codes within the personal prior 402”), corresponding to the plurality of training images depicting the entity (see para [0004]; “generating, using the one or more processors, a personalized prior for the subject based on a convex hull including each optimized code identified for each given image of the set of images of the subject”, see also para [0042]; “each image 602a-612a is assumed to be one of a set of n images of a subject”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify a method generating high-resolution images using a diffusion model of Cheng et al. in view of methods for identifying a personalized prior within a generative model's latent vector space based on a set of images of a given subject of Aberman2 in order perform inpainting to fill in masked areas, improve resolution, or deblur the image (see para [0066]).
Regarding claim 11, the rejection of claim 10 is incorporated herein.
Chen et al. in the combination further teaches wherein obtaining the training set comprises: generating a training image based on the input image (see para [0004]; “The diffusion model can be trained with input images at various resolution”, see also para [0041]; “the diffusion model 260 can be a stable diffusion model, which can be used as the base generative model”).
Regarding claim 14, the rejection of claim 10 is incorporated herein.
Chen et al. in the combination further teaches wherein training the image generation model comprises: generating a noise prediction based on the noisy image; computing a diffusion loss based on the noise prediction; and updating parameters of the image generation model based on the diffusion loss (see para [0039]; “the training component 240 can provide a training set of images having different resolutions, and calculate a loss for differences between a ground truth image and a predicted image. The loss can be used to update the diffusion model parameters and/or encoder parameters, so as to generate more accurate output image(s), where the output image(s) can have a different resolution than the input image(s)”, Note; the input and output images have different resolutions. This implies it is a specialized training mechanism (such as super-resolution) that handles training images of varying resolutions and includes updating "encoder parameters" in addition to standard diffusion parameters, which is common in Latent Diffusion Models).
Aberman2 in the combination further teaches and the plurality of training images(see para [0004]; “generating, using the one or more processors, a personalized prior for the subject based on a convex hull including each optimized code identified for each given image of the set of images of the subject”, see also para [0042]; “each image 602a-612a is assumed to be one of a set of n images of a subject”).
Regarding claim 15, the rejection of claim 10 is incorporated herein.
Chen et al. in the combination further teaches wherein: the noise prediction is generated based on the guidance information (see para [0080]; “an encoder 760 that can be configured and trained to generate an embedding of the initial image 605 that captures the image identity and summarizes the image content, as a global image code”, Note: the mechanism (encoder 760) used to create that "guidance information").
Regarding claim 16, the rejection of claim 10 is incorporated herein.
Chen et al. in the combination further teaches wherein training the image generation model comprises: performing a diffusion process at a first timestep using the guidance information; and performing a diffusion process at a second timestep without the guidance information (see para [0043]; “The model takes as input x (i.e., noisy or partially denoised image depending on the timestep), the timestep t, and conditioning information the model was trained to use. In some cases, the conditioning information can be a text prompt (e.g., TP, “ ”, and AP are text prompts). Classifier-free guidance is a mechanism to vary and control the influence of the conditioning on the sampled distribution at inference. In some cases, the conditioning can be replaced by the null token (i.e., the empty string, “ ”, in case of text conditioning) during training. A single scalar can control effect of the conditioning during inference” Note: using the guidance implies running the model with conditioning (e.g., text prompt), without the guidance implies running the model with a null token (empty string) and Classifier-free guidance connects these two separate runs (with/without guidance) to produce a combined, guided result).
Claims 12-13 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et al. in view of Aberman 2 as applied in claim 10 above, and further in view of Varanka et al.
Regarding claim 12, the rejection of claim 10 is incorporated herein. The combination of Chen et al. and Aberman2 as a whole does not teach wherein obtaining the training set comprises: obtaining a real image of the entity other than the input image
Varanka et al. teaches wherein obtaining the training set comprises: obtaining a real image of the entity other than the input image (see page 2373, right col. 2nd para; “Given a few HQ reference images (e.g. 3-5 selfies from a photo gallery) a restoration model is fine-tuned to a personalized restoration model. The reference images can have significantly different illumination, pose, expression and do not have to be aligned with the LQ image”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify a method generating high-resolution images using a diffusion model of Cheng et al. in view of methods for identifying a personalized prior within a generative model's latent vector space based on a set of images of a given subject of Aberman2 and personalized face restoration and super-resolution of Varanka et al. in order to improve the training pipeline of face restoration models to enable an alignment-free approach (see page 2375, Fig. 3).
Regarding claim 13, the rejection of claim 10 is incorporated herein.
Varanka et al. in the combination further teaches further comprising: initializing the image generation model based on a pre-trained image generation model (see page 2376, 3.3; “We train our model with the facial dataset FFHQ using the steps described below, which is initialized from the pre-trained Sta bleSR”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify a method generating high-resolution images using a diffusion model of Cheng et al. in view of methods for identifying a personalized prior within a generative model's latent vector space based on a set of images of a given subject of Aberman2 and personalized face restoration and super-resolution of Varanka et al. in order to generate high-quality image while mostly retaining the identity (see page 2375, 3.2).
Claims 17-18 rejected under 35 U.S.C. 103 as being unpatentable over et al. Nie et al. view of Zhang et al. and further in view of Aberman et al.
Regarding claim 17, Nie et al. teaches an apparatus comprising: at least one processor; at least one memory storing instructions executable by the at least one processor (see para [0228]; “a processor implementing a combination of instruction sets, or any other processor device, such as a digital signal processor, for example. …processor 1102 may be coupled to a processor bus 1110 that may transmit data signals between processor 1102 and other components in computer system 1100….. processor 1102 may include, without limitation, a Level 1 (“L1”) internal cache memory”); and an image generation model, comprising parameters stored in the at least one memory (see para [0062]; “storage 701 to store forward and/or output weight and/or input/output data, and/or other parameters to configure neurons or layers of a neural network trained”), and trained to generate a restored image based on an input image depicting an entity (see para [0059]; “generative process in which a version of this image is generated that has at least some noise removed relative to a prior image in this noise removal sequence. In at least one embodiment, these iterations can continue to be performed until this process arrives at a “purified” image 206 that is substantially free of noise, that is substantially free of adversarial perturbations, and has sufficient structure to enable an accurate classification of an object in this purified image”, see also para [0063]; “purified image 306 can be passed through a trained classifier 308 that can generate a correct label 310 for an object in purified image 306”), wherein the input image is combined with a noise input by selecting a diffusion timestep based on the first quality level to obtain a noisy image (see para [0063]; “noise can first be added to an adversarial image 302 by following a forward process with a small diffusion timestep”, see also para [0060]; “a process can utilize a selected diffusion timestep, which represents an amount of noise added during forward process”), wherein the restored image is generated based on the noisy image by denoising from the noisy image starting at the selected diffusion timestep (see para [0081]; “removing noise can be accomplished by generating a new version of this image at each reverse iteration that has less noise than a prior version of this image. In at least one embodiment, this process can be continued until a final iteration, in which a purified image can be generated”, see also para [0071]; “a reverse-time SDE can be solved from timestep t=t* using diffused adversarial sample x(t*) as an initial value to get a final solution”). However, Gandelsman et al. does not teach and having a first quality level representing a degree of degradation, and wherein the image generation model is trained to restore images depicting the entity using training data including a plurality of training images using a training image depicting the entity.
In the same field of endeavor, Zhang et al. teaches and having a first quality level representing a degree of degradation (see claim 2; “acquiring a first medical image having a first image quality, wherein the first medical image is an MR medical image of a subject”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify the general use of techniques presented to remove unintended variations introduced into data of Nie rt al. in view of systems that can address various drawbacks of conventional systems of Zhang in order to improving image quality with shortened acquisition time (see claim 2). However, the combination of Nie et al. and Zhang et al. does not teach wherein the image generation model is trained to restore images depicting the entity using training data including a plurality of training images depicting the entity.
In the same field of endeavor, Aberman et al. teaches and wherein the image generation model is trained to restore images depicting the entity using training data including a plurality of training images using a training image depicting the entity (see para [0043]; “The fine-tuning system 120 obtains the custom image dataset 104 and fine-tunes the pre-trained text-to-image model 102 on the obtained custom image dataset 104. The custom image dataset 104 includes a plurality of images that each depict a same particular subject instance of an object class”, see also para [0050]; “3-5 images, that each depict the particular subject instance”, and para [0075]; “The system generates a custom image dataset from a plurality of images that each depict a same particular subject instance (step 202)”, and para [0081]; “The system trains a text-to-image model on the custom image dataset”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify the general use of techniques presented to remove unintended variations introduced into data of Nie rt al. in view of systems that can address various drawbacks of conventional systems of Zhang and further in view of a method that trains a text-to-image model to generate an image conditioned on a text input of Aberman et al. in order to retrieve a specific subject instance from the model (see para [0043]).
Regarding claim 18, the rejection of claim 17 is incorporated herein.
Nie et al. in the combination further teaches wherein: the image generation model comprises a diffusion model (see para [0059]; “a pre-trained diffusion model can be utilized”).
Claims 19-20 rejected under 35 U.S.C. 103 as being unpatentable over et al. Nie et al. and Zhang et al. view of Aberman2 and further in view of Gandelsman et al.
Regarding claim 19, the rejection of claim 17 is incorporated herein. The combination of Nie et al., Zhang et al. and Aberman et al. does not teach wherein: the image generation model comprises a U-Net architecture.
In the same field of endeavor, Gandelsman et al. teaches wherein: the image generation model comprises a U-Net architecture (see para [0012]; “a diffusion model using a U-Net”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify the general use of techniques presented to remove unintended variations introduced into data of Nie rt al. in view of systems that can address various drawbacks of conventional systems of Zhang and further in view of a method that trains a text-to-image model to generate an image conditioned on a text input of Aberman et al. and machine learning model trained to produce output images consistent with a given input image of Gandelsman et al. in order to generate different versions of the image (see para [0012]).
Regarding claim 20, the rejection of claim 17 is incorporated herein.
Gandelsman et al. in the combination further teaches further comprising: an output space of the image generation model is constrained to images depicting the entity based on the training (see para [0024]; “The present disclosure uses diffusion models to generate images that retain the identity of a target image (i.e., recognizable characteristics that are not captured in a semantic label)”). Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filling date to the claimed invention to modify the general use of techniques presented to remove unintended variations introduced into data of Nie rt al. in view of systems that can address various drawbacks of conventional systems of Zhang and further in view of a method that trains a text-to-image model to generate an image conditioned on a text input of Aberman et al. and machine learning model trained to produce output images consistent with a given input image of Gandelsman et al. in order to generate different versions of the image (see para [0024]).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to WINTA GEBRESLASSIE whose telephone number is (571)272-3475. The examiner can normally be reached Monday-Friday9:00-5:00.
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, Andrew Bee can be reached at 571-270-5180. 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.
/WINTA GEBRESLASSIE/ Examiner, Art Unit 2677
