Prosecution Insights
Last updated: November 20, 2026
Application No. 18/802,764

DEVICE AND COMPUTER-IMPLEMENTED METHODS FOR MACHINE LEARNING, FOR PROVIDING A MACHINE LEARNING SYSTEM, OR FOR OPERATING A TECHNICAL SYSTEM

Non-Final OA §102§103
Filed
Aug 13, 2024
Priority
Sep 20, 2023 — EU 23 19 8631.6
Examiner
ZHAO, LEI
Art Unit
2668
Tech Center
2600 — Communications
Assignee
Robert Bosch GmbH
OA Round
1 (Non-Final)
74%
Grant Probability
Favorable
1-2
OA Rounds
9m
Est. Remaining
93%
With Interview

Examiner Intelligence

Grants 74% — above average
74%
Career Allowance Rate
57 granted / 77 resolved
+12.0% vs TC avg
Strong +19% interview lift
Without
With
+18.8%
Interview Lift
resolved cases with interview
Typical timeline
3y 0m
Avg Prosecution
22 currently pending
Career history
110
Total Applications
across all art units

Statute-Specific Performance

§101
6.8%
-33.2% vs TC avg
§103
64.0%
+24.0% vs TC avg
§102
26.5%
-13.5% vs TC avg
§112
2.7%
-37.3% vs TC avg
Black line = Tech Center average estimate • Based on career data from 77 resolved cases

Office Action

§102 §103
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 . Election/Restrictions Applicant’s election without traverse of Group I, claims 1-9 and 14-15 in the reply filed on 8/17/2026 is acknowledged. Claims 10-13 are canceled; claims 16-24 are newly added. Claim Rejections - 35 USC § 102 The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action: A person shall be entitled to a patent unless – (a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention. (a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention. Claims 1, 14, 15, 20 and 22 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Kim (Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models, arXiv:2211.17091v4 [cs.CV] 4 Jun 2023). Regarding claim 1, Kim teaches a computer-implemented method (The proposed method, Discriminator Guidance, aims to improve sample generation of pre-trained diffusion models. Abstract. We release the code at https: //github.com/alsdudrla10/DG. Abstract) comprising: generating a synthetic digital image ( PNG media_image1.png 428 1250 media_image1.png Greyscale ): (i) for generating training data and/or test data for training a machine learning system (Instead, we train the discriminator after score training, making discriminator training stable and fast to converge. Abstract), or (ii) for training an image classifier, the generating comprising the following steps: providing a latent variable ( PNG media_image2.png 214 608 media_image2.png Greyscale . Page 2 right column 1st paragraph) of a diffusion model (Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models. Title) representing the synthetic digital image ( PNG media_image3.png 428 1250 media_image3.png Greyscale ), wherein the providing of the latent variable includes: (i) sampling the latent variable from random noise ( PNG media_image4.png 490 604 media_image4.png Greyscale . Page 5 right column 1st paragraph), or (ii) providing a digital reference image, and adding noise with the diffusion model to the digital reference image to determine the latent variable, or (iii) mapping the digital reference image with an encoder to an input of the diffusion model, and adding noise with the diffusion model to the input to determine the latent variable (The claim language is interpreted disjunctively.); and (i) mapping the latent variable with the diffusion model depending on parameters of the diffusion model ( PNG media_image5.png 562 606 media_image5.png Greyscale . Page 5 right column 2nd paragraph) to features of the diffusion model (Classifier Guidance provides supervision information on a sample path, evaluating whether the sample is correctly classified by the class label y, or not. Page 5 right column 1st paragraph. As sampling from the joint distribution of (xt, y) requires accurate score estimation, Discriminator Guidance and Classifier Guidance can be combined for a synergistic effect. Page 5 right column 2nd paragraph), mapping the features with the diffusion model ( PNG media_image6.png 562 606 media_image6.png Greyscale . Page 5 right column 2nd paragraph) depending on parameters of the diffusion model ( PNG media_image5.png 562 606 media_image5.png Greyscale . Page 5 right column 2nd paragraph) to an output of the diffusion model ( PNG media_image3.png 428 1250 media_image3.png Greyscale ), wherein the synthetic digital image comprises the output ( PNG media_image3.png 428 1250 media_image3.png Greyscale ), or (ii) mapping the output with a decoder to the synthetic digital image, mapping the features or the synthetic digital image with a discriminator to a class of a predetermined set of classes for a real digital image or to a class for a synthetic digital image, depending on parameters of the discriminator (The claim language is interpreted disjunctively), and learning at least one parameter of the diffusion model and/or the discriminator ( PNG media_image5.png 562 606 media_image5.png Greyscale . Page 5 right column 2nd paragraph) depending on a loss for the discriminator ( PNG media_image7.png 356 612 media_image7.png Greyscale . Page 4 left column 1st paragraph) that is defined depending on the class (As sampling from the joint distribution of (xt, y) requires accurate score estimation, Discriminator Guidance and Classifier Guidance can be combined for a synergistic effect. Page 5 right column 2nd paragraph) and a predetermined reference class (Classifier Guidance provides supervision information on a sample path, evaluating whether the sample is correctly classified by the class label y, or not. Page 5 right column 1st paragraph. As sampling from the joint distribution of (xt, y) requires accurate score estimation, Discriminator Guidance and Classifier Guidance can be combined for a synergistic effect. Page 5 right column 2nd paragraph). Regarding claim 20, Kim teaches the method according to claim 1, wherein the generating of the synthetic digital image includes the mapping of the latent variable with the diffusion model ( PNG media_image2.png 214 608 media_image2.png Greyscale . Page 2 right column 1st paragraph) depending on the parameters of the diffusion model ( PNG media_image5.png 562 606 media_image5.png Greyscale . Page 5 right column 2nd paragraph) to features of the diffusion model ( PNG media_image6.png 562 606 media_image6.png Greyscale . Page 5 right column 2nd paragraph), mapping the features with the diffusion model depending on parameters of the diffusion model ( PNG media_image5.png 562 606 media_image5.png Greyscale . Page 5 right column 2nd paragraph) to the output of the diffusion model ( PNG media_image3.png 428 1250 media_image3.png Greyscale ), and wherein the synthetic digital image comprises the output ( PNG media_image3.png 428 1250 media_image3.png Greyscale ). Regarding claim 22, Kim teaches the method according to claim 1, wherein the providing of the latent variable includes the sampling of the latent variable from the random noise ( PNG media_image4.png 490 604 media_image4.png Greyscale . Page 5 right column 1st paragraph). Apparatus claim 14 is drawn to the apparatus corresponding to the method of using same as claimed in claim 1. Therefore apparatus claim 14 corresponds to method claim 1, and is rejected for the same reasons of anticipation as used above. Claim 15 is drawn to a non-transitory computer-readable storage medium having executable instructions stored for carrying out the method of claim 1. Therefore, claim 15 corresponds to method claim 1, and is rejected for the same reasons of anticipation as used above. 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 2, 6-7 and 9 are rejected under 35 U.S.C. 103 as being unpatentable over Kim (Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models, arXiv:2211.17091v4 [cs.CV] 4 Jun 2023) hereinafter Kim, in view of Sushko (OASIS: Only Adversarial Supervision for Semantic Image Synthesis, International Journal of Computer Vision (2022) 130:2903–2923), hereinafter Sushko. Regarding claim 2, Kim teaches the method according to claim 1, wherein the synthetic digital image includes pixels ( PNG media_image3.png 428 1250 media_image3.png Greyscale ). Kim does not teach the following limitations as further recited, but Sushko further teaches wherein the generating further comprises determining the class pixel-wise (With this design, the network cannot ignore the provided label maps, as it has to predict a correct class label for each pixel of an image. Page 2904 right column last paragraph), and learning the at least one parameter ( PNG media_image8.png 174 614 media_image8.png Greyscale . Page 2909 left column 1st paragraph) depending on a loss ( PNG media_image9.png 434 614 media_image9.png Greyscale . Page 2909 left column 1st paragraph) that is defined depending on the pixel-wise classes ( PNG media_image10.png 174 614 media_image10.png Greyscale . Page 2909 left column 1st paragraph) and respective predetermined pixel-wise reference classes (The segmentation task of the discriminator is formulated to predict the per-pixel class label of the real images, using the given semantic label maps as ground truth. In addition to the N semantic classes from the label maps, all pixels of fake images are categorized as one extra class. Page 2908 right column last paragraph. PNG media_image11.png 410 1264 media_image11.png Greyscale ). It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Kim to incorporate the teachings of Sushko to learn the parameter depending on a loss that is defined depending on the pixel-wise classes and respective predetermined pixel-wise reference classes in order to generate synthetic images from user-specified layouts. Regarding claim 6, Sushko in the combination teaches the method according claim 1, wherein the generating further comprises providing a condition including a label map ( PNG media_image11.png 410 1264 media_image11.png Greyscale ), modulating the features (In this work, we enable multi-modal synthesis directly through sampling of a 3D noise tensor which is injected at every layer of the network. Page 2907 right column 2nd paragraph) depending on the condition ( PNG media_image12.png 566 1254 media_image12.png Greyscale ), and mapping the modulated features (In this work, we enable multi-modal synthesis directly through sampling of a 3D noise tensor which is injected at every layer of the network. Page 2907 right column 2nd paragraph): (i) with the diffusion model to the output, and/or (ii) with the discriminator to the class (Thus, we propose to cast the discriminator task as a multi-class semantic segmentation problem to directly utilize label maps for supervision, and accordingly alter its architecture to an encoder-decoder segmentation network (see Fig. 3). Page 2908 right column 3rd paragraph PNG media_image13.png 412 1264 media_image13.png Greyscale ). Regarding claim 7, Sushko in the combination teaches the method according to claim 6, wherein the generating further comprises learning the at least one parameter ( PNG media_image8.png 174 614 media_image8.png Greyscale . Page 2909 left column 1st paragraph) depending on the loss that is defined for the discriminator ( PNG media_image9.png 434 614 media_image9.png Greyscale . Page 2909 left column 1st paragraph), wherein the loss that is defined for the discriminator is defined depending on the condition ( PNG media_image14.png 564 620 media_image14.png Greyscale . Page 2909 left column last paragraph). Regarding claim 9, Sushko in the combination teaches the method according to claim 2, wherein the generating further comprises providing a condition including a label map ( PNG media_image11.png 410 1264 media_image11.png Greyscale ), modulating the features (In this work, we enable multi-modal synthesis directly through sampling of a 3D noise tensor which is injected at every layer of the network. Page 2907 right column 2nd paragraph) depending on the condition ( PNG media_image12.png 566 1254 media_image12.png Greyscale ), and mapping the modulated features (In this work, we enable multi-modal synthesis directly through sampling of a 3D noise tensor which is injected at every layer of the network. Page 2907 right column 2nd paragraph): (i) with the diffusion model to the output, and/or (ii) with the discriminator to the class (Thus, we propose to cast the discriminator task as a multi-class semantic segmentation problem to directly utilize label maps for supervision, and accordingly alter its architecture to an encoder-decoder segmentation network (see Fig. 3). Page 2908 right column 3rd paragraph PNG media_image13.png 412 1264 media_image13.png Greyscale ), and wherein the condition comprises the predetermined reference class or the pixel-wise reference classes (With this design, the network cannot ignore the provided label maps, as it has to predict a correct class label for each pixel of an image. Page 2904 right column last paragraph). Claims 3-5, 21, 23 and 24 are rejected under 35 U.S.C. 103 as being unpatentable over Kim (Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models, arXiv:2211.17091v4 [cs.CV] 4 Jun 2023) hereinafter Kim, in view of Amit (SegDiff: Image Segmentation with Diffusion Probabilistic Models, arXiv:2112.00390v3 [cs.CV] 7 Sep 2022), hereinafter Amit. Regarding claim 3, Kim teaches the method according to claim 1, wherein the generating further comprises determining a sequence of outputs comprising the output ( PNG media_image15.png 428 1250 media_image15.png Greyscale ). Kim does not teach the following limitations as further recited, but Amit further teaches wherein determining the sequence of outputs includes determining a first output of the sequence depending on a result of encoding the input and decoding the encoded input ( PNG media_image16.png 322 522 media_image16.png Greyscale . Page 4 left column 2nd paragraph), and successively determining a next output in the sequence depending on a result of encoding the output preceding a next output in the sequence and decoding the encoded output preceding a next output in the sequence ( PNG media_image17.png 464 1056 media_image17.png Greyscale ), wherein encoding or decoding includes determining the features ( PNG media_image18.png 464 1056 media_image18.png Greyscale ). It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Kim to incorporate the teachings of Amit to successively determine a next output in the sequence depending on a result of encoding the output preceding a next output in the sequence and decoding the encoded output preceding a next output in the sequence, wherein encoding or decoding includes determining the features, in order to apply diffusion model to generate images when the ground truth result is unique. Regarding claim 4, Kim in the combination teaches the method according to claim 3, wherein the generating further comprises determining, with a respective discriminator, a class for a plurality of outputs of the sequence (We train a class-conditional discriminator for the class-conditional generation. Page 7 left column 1st paragraph) depending on features collected from respective encoding (We use the encoder of U-Net structure as our discriminator network. Page 7 left column 1st paragraph) or decoding, and learning the parameters ( PNG media_image5.png 562 606 media_image5.png Greyscale . Page 5 right column 2nd paragraph) depending on losses ( PNG media_image7.png 356 612 media_image7.png Greyscale . Page 4 left column 1st paragraph) that are defined for each of the respective discriminators depending on the class predicted by the respective discriminator (We train a class-conditional discriminator for the class-conditional generation. Page 7 left column 1st paragraph) and the predetermined reference class (Classifier Guidance provides supervision information on a sample path, evaluating whether the sample is correctly classified by the class label y, or not. Page 5 right column 1st paragraph. As sampling from the joint distribution of (xt, y) requires accurate score estimation, Discriminator Guidance and Classifier Guidance can be combined for a synergistic effect. Page 5 right column 2nd paragraph). Regarding claim 5, Kim in the combination teaches the method according to claim 4, wherein the generating includes predicting a noise in the next output of the sequence depending on the output preceding the next output in the sequence of outputs ( PNG media_image19.png 144 608 media_image19.png Greyscale . Page 4 left column 2nd paragraph), and learning the parameters depending a loss for noise ( PNG media_image20.png 236 612 media_image20.png Greyscale . Page 3 left column 3rd paragraph) and depending on the losses that are defined for the respective discriminators ( PNG media_image7.png 356 612 media_image7.png Greyscale . Page 4 left column 1st paragraph), wherein the loss for noise is defined depending on the predicted noise and a predetermined random noise sampled from a Gaussian distribution ( PNG media_image21.png 430 613 media_image21.png Greyscale . Page 3 left column 3rd paragraph). Regarding claim 21, Kim in the combination teaches the method according to claim 1, wherein the generating of the synthetic digital image includes mapping the features or the synthetic digital image with the discriminator to the class of the predetermined set of classes for the real digital image (Classifier Guidance provides supervision information on a sample path, evaluating whether the sample is correctly classified by the class label y, or not. Page 5 right column 1st paragraph. As sampling from the joint distribution of (xt, y) requires accurate score estimation, Discriminator Guidance and Classifier Guidance can be combined for a synergistic effect. Page 5 right column 2nd paragraph) or to the class for the synthetic digital image, depending on the parameters of the discriminator ( PNG media_image5.png 562 606 media_image5.png Greyscale ), and learning at least one parameter of the diffusion model and/or the discriminator ( PNG media_image5.png 562 606 media_image5.png Greyscale . Page 5 right column 2nd paragraph) depending on the Loss for the discriminator ( PNG media_image7.png 356 612 media_image7.png Greyscale . Page 4 left column 1st paragraph) that is defined depending on the class (As sampling from the joint distribution of (xt, y) requires accurate score estimation, Discriminator Guidance and Classifier Guidance can be combined for a synergistic effect. Page 5 right column 2nd paragraph) and the predetermined reference class (Classifier Guidance provides supervision information on a sample path, evaluating whether the sample is correctly classified by the class label y, or not. Page 5 right column 1st paragraph. As sampling from the joint distribution of (xt, y) requires accurate score estimation, Discriminator Guidance and Classifier Guidance can be combined for a synergistic effect. Page 5 right column 2nd paragraph). Amit in the combination further teaches the mapping of the output with the decoder to the synthetic digital image ( PNG media_image22.png 464 1056 media_image22.png Greyscale ). Regarding claim 23, Amit in the combination teaches the method according to claim 1, wherein the providing of the latent variable includes the providing of the digital reference image (This method allows the generation of images based on a given reference image without any additional learning. Page 2 left column 5th paragraph), and the adding of noise ( PNG media_image23.png 104 510 media_image23.png Greyscale . Page 2 right column 4th paragraph) with the diffusion model to the digital reference image to determine the latent variable ( PNG media_image24.png 464 1056 media_image24.png Greyscale ). Regarding claim 24, Amit in the combination teaches the method according to claim 1, wherein the providing of the latent variable includes the mapping of the digital reference image with the encoder to the input of the diffusion model ( PNG media_image25.png 464 1056 media_image25.png Greyscale ), and the adding of noise with the diffusion model to the input to determine the latent variable ( PNG media_image23.png 104 510 media_image23.png Greyscale . Page 2 right column 4th paragraph). Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Kim (Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models, arXiv:2211.17091v4 [cs.CV] 4 Jun 2023) hereinafter Kim, in view of Amit (SegDiff: Image Segmentation with Diffusion Probabilistic Models, arXiv:2112.00390v3 [cs.CV] 7 Sep 2022), hereinafter Amit, further in view of Sushko (OASIS: Only Adversarial Supervision for Semantic Image Synthesis, International Journal of Computer Vision (2022) 130:2903–2923), hereinafter Sushko. Regarding claim 8, Kim and Amit teach all of the elements of the claimed invention as stated in claim 4 except for the following limitations as further recited. However, Sushko teaches wherein the generating further comprises providing a condition including a label map ( PNG media_image11.png 410 1264 media_image11.png Greyscale ), modulating the features (In this work, we enable multi-modal synthesis directly through sampling of a 3D noise tensor which is injected at every layer of the network. Page 2907 right column 2nd paragraph) depending on the condition ( PNG media_image12.png 566 1254 media_image12.png Greyscale ), and mapping the modulated features (In this work, we enable multi-modal synthesis directly through sampling of a 3D noise tensor which is injected at every layer of the network. Page 2907 right column 2nd paragraph): (i) with the diffusion model to the output, and/or (ii) with the discriminator to the class (Thus, we propose to cast the discriminator task as a multi-class semantic segmentation problem to directly utilize label maps for supervision, and accordingly alter its architecture to an encoder-decoder segmentation network (see Fig. 3). Page 2908 right column 3rd paragraph PNG media_image13.png 412 1264 media_image13.png Greyscale ), and learning the parameters ( PNG media_image8.png 174 614 media_image8.png Greyscale . Page 2909 left column 1st paragraph) depending on the losses ( PNG media_image26.png 314 618 media_image26.png Greyscale . Page 2910 left column 3rd paragraph) that are defined for the respective discriminators (This new discriminator provides semantically-aware pixel-level feedback to the generator, partitioning the image into segments belonging to one of the N real semantic classes or the fake class. Page 2904 right column 1st paragraph), wherein the losses are defined ( PNG media_image26.png 314 618 media_image26.png Greyscale . Page 2910 left column 3rd paragraph) depending on the condition ( PNG media_image12.png 566 1254 media_image12.png Greyscale ). It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Kim and Amit to incorporate the teachings of Sushko to map the modulated features depending on the label map with the discriminator to the class and learn the parameters depending on the losses that are defined for the respective discriminators in order to generate synthetic images from user-specified layouts. Claims 16-17 are rejected under 35 U.S.C. 103 as being unpatentable over Kim (Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models, arXiv:2211.17091v4 [cs.CV] 4 Jun 2023) hereinafter Kim, in view of Wolleb (Diffusion Models for Implicit Image Segmentation Ensembles, Proceedings of Machine Learning Research 172:1–13, 2022), hereinafter Wolleb. Regarding claim 16, Kim teaches all of the elements of the claimed invention as stated in claim 1 except for the following limitations as further recited. However, Wolleb teaches wherein the generating of the synthetic digital image is for generating the training data and/or the test data for training the machine learning system (In this paper, we present a novel semantic segmentation method based on diffusion models. By modifying the training and sampling scheme, we show that diffusion models can perform lesion segmentation of medical images. Abstract), wherein the machine learning system comprises a neural network for semantic segmentation (In this paper, we present a novel semantic segmentation method based on diffusion models. By modifying the training and sampling scheme, we show that diffusion models can perform lesion segmentation of medical images. Abstract), object recognition, or classification of objects, and wherein the method further comprises training, testing, verifying, or validating the machine learning system for semantic segmentation (In this paper, we present a novel semantic segmentation method based on diffusion models. By modifying the training and sampling scheme, we show that diffusion models can perform lesion segmentation of medical images. Abstract), object recognition, or classification of objects depending on the synthetic digital image ( PNG media_image27.png 224 1110 media_image27.png Greyscale ). It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Kim to incorporate the teachings of Wolleb to generate the training data for training a neural network for semantic segmentation depending on the synthetic digital image without having to train a new model for each segmentation mask. Regarding claim 17, Wolleb in the combination teaches the method according to claim 16, further comprising providing a condition comprising a label map (Let b be the given brain MR image of dimension (c, h, w), where c denotes the number of channels, and (h, w) denote the image height and image width. The ground truth segmentation of the tumor for the input image b is denoted as xb (which reads on “a label map”), and is of dimension (1, h, w). Page 4 last paragraph) for generating the synthetic digital image ( PNG media_image28.png 504 1078 media_image28.png Greyscale ), and using the label map as ground truth (The ground truth segmentation of the tumor for the input image b is denoted as xb (which reads on “a label map”), and is of dimension (1, h, w). Page 4 last paragraph) for training, testing, verifying, or validating the machine learning system (Using the stochastic sampling process, our method allows implicit ensembling of different segmentation masks for the same input brain MR image, without having to train a new model. We could show that ensembling those segmentation masks increases the performance of the model with respect to different segmentation scores. Page 8 last paragraph). Claims 18-19 are rejected under 35 U.S.C. 103 as being unpatentable over Kim (Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models, arXiv:2211.17091v4 [cs.CV] 4 Jun 2023) hereinafter Kim, in view of Brehar (Pedestrian Street-Cross Action Recognition in Monocular Far Infrared Sequences, IEEE Access, vol. 9, pp. 74302-74324, 2021), hereinafter Brehar. Regarding claim 18, Kim teaches the method according to claim 1, wherein the generating of the synthetic digital image is for generating the training data and/or the test data for training the machine learning system (We show that the discriminator-guided samples are closer to the real-world data than the non-guided samples, theoretically and empirically. Page 2 left column last paragraph), and wherein the method further comprises: training, testing, verifying, or validating the machine learning system for semantic segmentation, object recognition, or classification of objects (We show that the discriminator-guided samples are closer to the real-world data than the non-guided samples, theoretically and empirically. Page 2 left column last paragraph) depending on the synthetic digital image (We show that the discriminator-guided samples are closer to the real-world data than the non-guided samples, theoretically and empirically. Page 2 left column last paragraph). Kim does not teach the following limitations as further recited, but Brehar further teaches capturing, with a sensor, an operational digital image (The used image sensor consists of a FLIR PathFindIR camera, incorporating an uncooled 320 240 Vox microbolometer, with 8-14 m spectral response. Page 74306 right column 4th paragraph), wherein the operational digital image is a video image, a radar image, a LiDAR image, an ultrasonic image, a motion image, or a thermal image (The proposed dataset contains sequences of infrared images grabbed in winter and spring time. Page 74306 right column last paragraph); and determining (Action recognition module - uses a Long Short Term Memory Network that given a sequence of states and the measured features for a pedestrian in each state (frame) provides the probability the pedestrian is engaged in a cross or not cross action. Page 74307 right column 2nd paragraph), depending on the operational digital image (In our experiments we have used the PyTorch implementation and the pretraied encoder provided by ERFNet. This encoder was pre-trained on ImageNet while the decoder was trained from scratch on the infrared images. We have modified the number of classes to two (road and non road pixels), and trained the network with a batch size of 6 for 150 epochs. Page 74312 left column last paragraph) and using the machine learning system (The extraction of these features emerges from the combination of a deep learning based pedestrian detector with an original tracking algorithm, a semantic segmentation of the road surface and a time series long-short term memory network based action recognition. Abstract), a control signal for operating a technical system comprising a computer-controlled machine (The early recognition and understanding of the actions performed by pedestrians in traffic scenes leads to an anticipation of pedestrian intentions in advance and helps in the process of collision warning and avoidance in the context of autonomous vehicles. Abstract). It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Kim to incorporate the teachings of Brehar to generate a control signal for operating an autonomous vehicle using the thermal image and the machine learning system that generates synthetic digital image in order to help in the process of collision warning and avoidance in the context of autonomous vehicles. Regarding claim 19, Brehar in the combination teaches the method according to claim 18, wherein the computer-controlled machine is a robotic system, a vehicle ( PNG media_image29.png 484 630 media_image29.png Greyscale ), a domestic appliance, a power tool, a manufacturing machine, a personal assistant, or an access control system (Note: the claim language is interpreted as disjunctive.). Conclusion Any inquiry concerning this communication or earlier communications from the examiner should be directed to LEI ZHAO whose telephone number is (703)756-1922. The examiner can normally be reached Monday - Friday 8:00 am - 5:00 pm. Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, VU LE can be reached at (571)272-7332. 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. /LEI ZHAO/Examiner, Art Unit 2668 /VU LE/Supervisory Patent Examiner, Art Unit 2668
Read full office action

Prosecution Timeline

Aug 13, 2024
Application Filed
Oct 22, 2026
Non-Final Rejection mailed — §102, §103 (current)

Precedent Cases

Applications granted by this same examiner with similar technology

Patent 12797159
APPARATUS AND METHOD FOR GENERATING VIRTUAL MODEL
4y 4m to grant Granted Nov 17, 2026
Patent 12802201
METHOD FOR INSPECTING A CORRECT EXECUTION OF A PROCESSING STEP OF COMPONENTS, IN PARTICULAR A WIRING HARNESS, DATA STRUCTURE, AND SYSTEM
3y 5m to grant Granted Nov 17, 2026
Patent 12793235
MOTION-BASED TARGET TRACKING
5y 2m to grant Granted Nov 10, 2026
Patent 12784245
Image Processing and Segmentation of Sets of Z-Stacked Images of Three-Dimensional Biological Samples
3y 4m to grant Granted Nov 03, 2026
Patent 12784280
ENCODING AND DECODING METHOD, APPARATUS, AND DEVICE, STORAGE MEDIUM, COMPUTER PROGRAM, AND COMPUTER PROGRAM PRODUCT
2y 11m to grant Granted Nov 03, 2026
Study what changed to get past this examiner. Based on 5 most recent grants.

Strategy Recommendation AI-generated — please review before filing

Get a prosecution strategy drawn from examiner precedents, rejection analysis, and claim mapping.
Typically takes 5-10 seconds — AI-generated, attorney review required before filing

Prosecution Projections

1-2
Expected OA Rounds
74%
Grant Probability
93%
With Interview (+18.8%)
3y 0m (~9m remaining)
Median Time to Grant
Low
PTA Risk
Based on 77 resolved cases by this examiner. Grant probability derived from career allowance rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month