CTNF 18/893,037 CTNF 101461 DETAILED ACTION Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA. Information Disclosure Statement The information disclosure statements (IDS) submitted on 09/23/2024 is being considered by the examiner. Examiner’s Remarks The Office would like to bring to applicant’s attention that inventor Abhishek Aich is listed as residing in “San Jose, CA, ICELAND” instead of “San Jose, CA, UNITED STATES” on the Application Data Sheet and Bib Data Sheet. The Office respectfully advises the applicant to review and update the data as needed. Claim Rejections - 35 USC § 103 07-06 AIA 15-10-15 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. 07-20-aia AIA 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. 07-21-aia AIA Claim s 1, 8, 11, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over THAWAKAR et al. (US 20240161334 A1), hereinafter referenced as THAWAKAR, in view of ZHANG et al. (US 20210056357 A1), hereinafter referenced as ZHANG . Regarding claim 1, THAWAKAR explicitly teaches a computer-implemented method for segmentation (Fig. 1 and 3A-3B. Paragraph [0045]-THAWAKAR discloses a solution is the disclosed transformer-based video instance segmentation architecture that can capture multi-scale spatio-temporal feature relationships in a video. Further in paragraph [0045]-THAWAKAR discloses video instance segmentation accurately delineates the target object in the presence of cluttered background.) , comprising: encoding an image (Fig. 3A-3B, #302 called video frames. Paragraph [0049]-THAWAKAR discloses a video clip x consisting T frames 302 of spatial size H.sup.0×W.sup.0 with a set of object instances is input to the backbone network 304. Latent feature maps for each frame are obtained from the backbone 304 at L multiple scales 2.sup.−l−2 with 1≤l≤L and passed through separate convolution filters for each scale. The output feature dimension for each of these convolution filters is set to C (wherein video frames are images and wherein obtaining feature maps is encoding an image).) using a backbone model (Fig. 3A, #304 called Backbone. Paragraph [0049]-THAWAKAR discloses a video clip x consisting T frames 302 of spatial size H.sup.0×W.sup.0 with a set of object instances is input to the backbone network 304. Latent feature maps for each frame are obtained from the backbone 304 at L multiple scales 2.sup.−l−2 with 1≤l≤L and passed through separate convolution filters for each scale.) to generate a plurality of feature maps (Fig. 3A-3B. Paragraph [0049]-THAWAKAR discloses latent feature maps for each frame are obtained from the backbone 304 at L multiple scales.) ; processing the feature maps with a dynamic transformer encoder (Fig. 3A-3B. Paragraph [0049]-THAWAKAR discloses latent feature maps for each frame are obtained from the backbone 304 at L multiple scales 2.sup.−l−2 with 1≤l≤L and passed through separate convolution filters for each scale. The output feature dimension for each of these convolution filters is set to C. The resulting multi-scale feature maps of each frame are then input to a transformer encoder comprising multi-scale deformable attention blocks 314 (wherein the transformer encoder is a dynamic transformer encoder).) that includes a plurality of layers (Fig. 4. Paragraph [0052]-THAWAKAR discloses these features {circumflex over (Z)} 406 are fused in each encoder layer with the base features output by the conventional deformable attention 314. While backbone features F 306 are used as input to the first encoder layer, the subsequent layers utilize the outputs from the preceding layer as input. As a result, multi-scale spatio-temporally enriched features C-dimensional E 316 are output by the encoder 310.) , exiting the dynamic transformer encoder at a layer identified by the exit point (Fig. 3A, #316 called enriched features E indicate an exit of the transformer encoder. Paragraph [0052]-ZHANG discloses while backbone features F 306 are used as input to the first encoder layer, the subsequent layers utilize the outputs from the preceding layer as input. As a result, multi-scale spatio-temporally enriched features C-dimensional E 316 are output by the encoder 310 (wherein outputting is exiting). Please see annotated Fig. 3A below.) ; and decoding an output of the dynamic transformer encoder (Fig. 3A-3B, block #330 illustrates the process of decoding in the decoder. Paragraph [0050]-THAWAKAR discloses these encoder output features maps from each frame along with n learnable instance query embeddings are then input to the transformer decoder comprising a series of self- and cross-attention blocks (wherein the decoder conducts decoding).) to output a segmentation of the image (Fig. 3A-3B. Paragraph [0050]-THAWAKAR discloses the learned box queries across T frames are then aggregated temporally to obtain n instance features I.sup.O ∈ custom-character.sup.C. These instance features output by the decoder are then used for video instance mask prediction. Further in paragraph [0053]-THAWAKAR disclose the encoder features E 316 along with the instance features I.sup.O 334 (aggregated temporally attended box queries) from the decoder 330 are utilized within the instance matching and segmentation block 340 to obtain the video instance mask prediction 342 (wherein the instance mask prediction is a segmentation of the image).) . PNG media_image1.png 578 789 media_image1.png Greyscale Annotated diagram of THAWAKAR’s Fig. 3A indicating the layers of the transformer encoder and the identified exit layer. THAWAKAR fails to explicitly teach selecting an exit point based on one of the plurality of feature maps. However, ZHANG explicitly teaches selecting an exit point (Fig. 15, #1510 called decision module and “yes” indicates an early exit result (i.e. an exit point). Paragraph [0105]-ZHANG discloses the decision module 1510 then takes the early prediction results generated by the early exit branch 1508 and makes a decision on whether to exit the inference process and output the early prediction results, or to continue the inference process and pass the generated feature maps to the next layer (wherein to exit the inference process is selecting an exit point).) based on one of the plurality of feature maps (Fig. 15. Paragraph [0105]-ZHANG discloses the smaller early exit branch network merely takes the intermediate features generated by the j.sup.th internal convolutional layers 1512-1514 of the base model and transforms them into early predictions. The decision module 1510 then takes the early prediction results generated by the early exit branch 1508 and makes a decision on whether to exit the inference process and output the early prediction results, or to continue the inference process and pass the generated feature maps to the next layer. In this way, a “base” model (which may have already been trained on a training dataset) can be modified to make it flexible and adaptive to the “difficulty” of the input data, without the memory requirements of a “bag of models” approach or the inefficiency of a “coarse” approach (wherein the features are feature maps).) ; Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR of a computer-implemented method for segmentation, comprising: encoding an image using a backbone model to generate a plurality of feature maps; processing the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decoding an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of ZHANG of selecting an exit point based on one of the plurality of feature maps. Wherein having THAWAKAR’s transformer encoder for segmenting images having selecting an exit point based on one of the plurality of feature maps. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and ZHANG are computer vision systems and are systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while ZHANG the systems and methods herein implement various features and aspects that make them useful for any situation in which it would be beneficial to reduce computational demand of a deep neural network application where the input data set has varying complexity. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and ZHANG et al. (US 20210056357 A1), Paragraph [0054]. Regarding claim 8, THAWAKAR in view of ZHANG explicitly teach the method of claim 1, THAWAKAR fails to explicitly teach wherein the segmentation of the image includes identification of objects within the image. However, ZHANG explicitly teaches wherein the segmentation of the image includes identification of objects within the image (Fig. 7A-7B, illustrate identifying information included within the image. Paragraph [0143]-ZHANG discloses a standard data processing flowpath 103 represents how the neural network 100 would classify the input 102 without the early exits 108-B, passing data through every base layer 104-D regardless of input, culminating in a final exit output 110. The Final Exit output would be an inference output giving likelihood that the input contained certain features of interest (e.g., an indication of likelihood the image contained the human activity of biking) (wherein features of interests are objects).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR of a computer-implemented method for segmentation, comprising: encoding an image using a backbone model to generate a plurality of feature maps; processing the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decoding an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of ZHANG of wherein the segmentation of the image includes identification of objects within the image. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein the segmentation of the image includes identification of objects within the image. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and ZHANG are computer vision systems and are systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while ZHANG the systems and methods herein implement various features and aspects that make them useful for any situation in which it would be beneficial to reduce computational demand of a deep neural network application where the input data set has varying complexity. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and ZHANG et al. (US 20210056357 A1), Paragraph [0054]. Regarding claim 11, THAWAKAR explicitly teaches a system for segmentation (Fig. 1 and 3A-3B. Paragraph [0045]-THAWAKAR discloses a solution is the disclosed transformer-based video instance segmentation architecture that can capture multi-scale spatio-temporal feature relationships in a video. Further in paragraph [0045]-THAWAKAR discloses video instance segmentation accurately delineates the target object in the presence of cluttered background.) , comprising: a hardware processor (Fig. 11, #1100 called AI SoC. Paragraph [0071]-THAWAKAR discloses SOC 1100 may be a Qualcomm mobile processor, a Nvidia DRIVE processor, a Atom® processor from Intel Corporation of America, a Samsung mobile processor, or a Apple A mobile processor, or may be other processor types that would be recognized by one of ordinary skill in the art. Alternatively, the SOC 1100 may be implemented on an Field-Programmable Gate Array (FPGA), Application Specific Integrated Circuit (ASIC), Programmable Logic Device (PLD) or using discrete logic circuits, as one of ordinary skill in the art would recognize. Further, SOC 1100 may be implemented as multiple processors cooperatively working in parallel to perform the instructions of the inventive processes described above.) ; a memory (Fig. 11, #1102 called SDRAM Flash. Paragraph [0069]-THAWAKAR discloses the instructions may be stored in FLASH memory, Secure Digital Random Access Memory (SDRAM), Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read Only Memory (EEPROM), solid-state hard disk or any other information processing device with which the processing circuit 1126 communicates, such as a server or computer.) that stores a computer program (Fig. 11. Paragraph [0068]-THAWAKAR discloses the processing circuit 1126 includes an AI System on a Chip (SOC) 1100 which performs the processes described herein. The process data and instructions may be stored in memory 1102. These processes and instructions may also be stored on a portable storage medium or may be stored remotely (wherein instructions is a computer program).) which, when executed by the hardware processor, causes the hardware processor to (Fig. 11. Paragraph [0068]-THAWAKAR discloses the processing circuit 1126 includes an AI System on a Chip (SOC) 1100 which performs the processes described herein. The process data and instructions may be stored in memory 1102.) : encode an image (Fig. 3A-3B, #302 called video frames. Paragraph [0049]-THAWAKAR discloses a video clip x consisting T frames 302 of spatial size H.sup.0×W.sup.0 with a set of object instances is input to the backbone network 304. Latent feature maps for each frame are obtained from the backbone 304 at L multiple scales 2.sup.−l−2 with 1≤l≤L and passed through separate convolution filters for each scale. The output feature dimension for each of these convolution filters is set to C (wherein video frames are images and wherein obtaining feature maps is encoding an image).) using a backbone model (Fig. 3A, #304 called Backbone. Paragraph [0049]-THAWAKAR discloses a video clip x consisting T frames 302 of spatial size H.sup.0×W.sup.0 with a set of object instances is input to the backbone network 304. Latent feature maps for each frame are obtained from the backbone 304 at L multiple scales 2.sup.−l−2 with 1≤l≤L and passed through separate convolution filters for each scale.) to generate a plurality of feature maps (Fig. 3A-3B. Paragraph [0049]-THAWAKAR discloses latent feature maps for each frame are obtained from the backbone 304 at L multiple scales.) ; process the feature maps with a dynamic transformer encoder (Fig. 3A-3B. Paragraph [0049]-THAWAKAR discloses latent feature maps for each frame are obtained from the backbone 304 at L multiple scales 2.sup.−l−2 with 1≤l≤L and passed through separate convolution filters for each scale. The output feature dimension for each of these convolution filters is set to C. The resulting multi-scale feature maps of each frame are then input to a transformer encoder comprising multi-scale deformable attention blocks 314 (wherein the transformer encoder is a dynamic transformer encoder).) that includes a plurality of layers (Fig. 4. Paragraph [0052]-THAWAKAR discloses these features {circumflex over (Z)} 406 are fused in each encoder layer with the base features output by the conventional deformable attention 314. While backbone features F 306 are used as input to the first encoder layer, the subsequent layers utilize the outputs from the preceding layer as input. As a result, multi-scale spatio-temporally enriched features C-dimensional E 316 are output by the encoder 310.) , exiting the dynamic transformer encoder at a layer identified by the exit point (Fig. 3A, #316 called enriched features E indicate an exit of the transformer encoder. Paragraph [0052]-ZHANG discloses while backbone features F 306 are used as input to the first encoder layer, the subsequent layers utilize the outputs from the preceding layer as input. As a result, multi-scale spatio-temporally enriched features C-dimensional E 316 are output by the encoder 310 (wherein outputting is exiting). Please see annotated Fig. 3A below.) ; and decode an output of the dynamic transformer encoder (Fig. 3A-3B, block #330 illustrates the process of decoding in the decoder. Paragraph [0050]-THAWAKAR discloses these encoder output features maps from each frame along with n learnable instance query embeddings are then input to the transformer decoder comprising a series of self- and cross-attention blocks (wherein the decoder conducts decoding).) to output a segmentation of the image (Fig. 3A-3B. Paragraph [0050]-THAWAKAR discloses the learned box queries across T frames are then aggregated temporally to obtain n instance features I.sup.O ∈ custom-character.sup.C. These instance features output by the decoder are then used for video instance mask prediction. Further in paragraph [0053]-THAWAKAR disclose the encoder features E 316 along with the instance features I.sup.O 334 (aggregated temporally attended box queries) from the decoder 330 are utilized within the instance matching and segmentation block 340 to obtain the video instance mask prediction 342 (wherein the instance mask prediction is a segmentation of the image).) . PNG media_image1.png 578 789 media_image1.png Greyscale Annotated diagram of THAWAKAR’s Fig. 3A indicating the layers of the transformer encoder and the identified exit layer. THAWAKAR fails to explicitly teach select an exit point based on one of the plurality of feature maps. However, ZHANG explicitly teaches select an exit point (Fig. 15, #1510 called decision module and “yes” indicates an early exit result (i.e. an exit point). Paragraph [0105]-ZHANG discloses the decision module 1510 then takes the early prediction results generated by the early exit branch 1508 and makes a decision on whether to exit the inference process and output the early prediction results, or to continue the inference process and pass the generated feature maps to the next layer (wherein to exit the inference process is selecting an exit point).) based on one of the plurality of feature maps (Fig. 15. Paragraph [0105]-ZHANG discloses the smaller early exit branch network merely takes the intermediate features generated by the j.sup.th internal convolutional layers 1512-1514 of the base model and transforms them into early predictions. The decision module 1510 then takes the early prediction results generated by the early exit branch 1508 and makes a decision on whether to exit the inference process and output the early prediction results, or to continue the inference process and pass the generated feature maps to the next layer. In this way, a “base” model (which may have already been trained on a training dataset) can be modified to make it flexible and adaptive to the “difficulty” of the input data, without the memory requirements of a “bag of models” approach or the inefficiency of a “coarse” approach (wherein the features are feature maps).) ; Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR of a system for segmentation, comprising: a hardware processor; a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to: encode an image using a backbone model to generate a plurality of feature maps; process the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decode an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of ZHANG of select an exit point based on one of the plurality of feature maps. Wherein having THAWAKAR’s transformer encoder for segmenting images having select an exit point based on one of the plurality of feature maps. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and ZHANG are computer vision systems and are systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while ZHANG the systems and methods herein implement various features and aspects that make them useful for any situation in which it would be beneficial to reduce computational demand of a deep neural network application where the input data set has varying complexity. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and ZHANG et al. (US 20210056357 A1), Paragraph [0054]. Regarding claim 18, THAWAKAR in view of ZHANG explicitly teach the system of claim 11, THAWAKAR fails to explicitly teach wherein the segmentation of the image includes identification of objects within the image. However, ZHANG explicitly teaches wherein the segmentation of the image includes identification of objects within the image (Fig. 7A-7B, illustrate identifying information included within the image. Paragraph [0143]-ZHANG discloses a standard data processing flowpath 103 represents how the neural network 100 would classify the input 102 without the early exits 108-B, passing data through every base layer 104-D regardless of input, culminating in a final exit output 110. The Final Exit output would be an inference output giving likelihood that the input contained certain features of interest (e.g., an indication of likelihood the image contained the human activity of biking) (wherein features of interests are objects).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR of a system for segmentation, comprising: a hardware processor; a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to: encode an image using a backbone model to generate a plurality of feature maps; process the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decode an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of ZHANG of wherein the segmentation of the image includes identification of objects within the image. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein the segmentation of the image includes identification of objects within the image. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and ZHANG are computer vision systems and are systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while ZHANG the systems and methods herein implement various features and aspects that make them useful for any situation in which it would be beneficial to reduce computational demand of a deep neural network application where the input data set has varying complexity. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and ZHANG et al. (US 20210056357 A1), Paragraph [0054] . 07-21-aia AIA Claim s 2 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over THAWAKAR et al. (US 20240161334 A1), hereinafter referenced as THAWAKAR, in view of ZHANG et al. (US 20210056357 A1), hereinafter referenced as ZHANG, and further in view of SHELHAMER et al. (US 20220309285 A1), hereinafter referenced as SHELHAMER . Regarding claim 2, THAWAKAR in view of ZHANG explicitly teach the method of claim 1, THAWAKAR in view of ZHANG fail to explicitly teach wherein the exit point is selected based on a lowest-resolution feature map of the plurality of feature maps. However, SHELHAMER explicitly teaches wherein the exit point is selected (Fig. 3. Paragraph [0107]-SHELHAMER discloses as described above, the pixel-level prediction system 102 also applies the pixel mask generated from the first set of extracted features to further process unmasked features (e.g., features corresponding to a “1” in the pixel mask) utilizing a second early exit head (e.g., the second early exit head 410). For masked features (e.g., features corresponding to a “0” in the pixel mask), on the other hand, the pixel-level prediction system 102 preserves or reuses the predicted values generated via the previous early exit head (e.g., the first early exit head 404) (wherein using the predicted values from an early exit head is selecting an exit point as the exit point that corresponds to those values is selected).) based on a lowest-resolution feature map of the plurality of feature maps (Fig. 3. Paragraph [0065]-SHELHAMER discloses the pixel-level prediction system 102 utilizes the first early exit head 308 to generate a first pixel-wise classification prediction 312 from the first set of features 306. More specifically, the first early exit head 308 downsamples the first set of features 306 into a lower resolution utilizing an encoder 310a (wherein the down sampled features are the lowest resolution feature map when compared to first set of features #306 and wherein the plurality of feature maps includes the first set of features #306, the second set of features #318, and the down sampled features).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a computer-implemented method for segmentation, comprising: encoding an image using a backbone model to generate a plurality of feature maps; processing the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decoding an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of SHELHAMER of wherein the exit point is selected based on a lowest-resolution feature map of the plurality of feature maps. Wherein having THAWAKAR’s transformer encoder for segmenting images having wherein the exit point is selected based on a lowest-resolution feature map of the plurality of feature maps. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and SHELHAMER are image analysis systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while SHELHAMER the disclosed systems can mask the output of each exit head utilizing confidence thresholding that reduces the remaining computation in future exits. Utilizing this confidence adaptivity approach can further improve efficiency and flexibility of generating predictions at various exit heads without sacrificing accuracy in generating pixel-level predictions. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and SHELHAMER et al. (US 20220309285 A1), Paragraph [0006]. Regarding claim 12, THAWAKAR in view of ZHANG explicitly teach the system of claim 11, THAWAKAR in view of ZHANG fail to explicitly teach wherein the exit point is selected based on a lowest-resolution feature map of the plurality of feature maps. However, SHELHAMER explicitly teaches wherein the exit point is selected (Fig. 3. Paragraph [0107]-SHELHAMER discloses as described above, the pixel-level prediction system 102 also applies the pixel mask generated from the first set of extracted features to further process unmasked features (e.g., features corresponding to a “1” in the pixel mask) utilizing a second early exit head (e.g., the second early exit head 410). For masked features (e.g., features corresponding to a “0” in the pixel mask), on the other hand, the pixel-level prediction system 102 preserves or reuses the predicted values generated via the previous early exit head (e.g., the first early exit head 404) (wherein using the predicted values from an early exit head is selecting an exit point as the exit point that corresponds to those values is selected).) based on a lowest-resolution feature map of the plurality of feature maps (Fig. 3. Paragraph [0065]-SHELHAMER discloses the pixel-level prediction system 102 utilizes the first early exit head 308 to generate a first pixel-wise classification prediction 312 from the first set of features 306. More specifically, the first early exit head 308 downsamples the first set of features 306 into a lower resolution utilizing an encoder 310a (wherein the down sampled features are the lowest resolution feature map when compared to first set of features #306 and wherein the plurality of feature maps includes the first set of features #306, the second set of features #318, and the down sampled features)) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a system for segmentation, comprising: a hardware processor; a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to: encode an image using a backbone model to generate a plurality of feature maps; process the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decode an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of SHELHAMER of wherein the exit point is selected based on a lowest-resolution feature map of the plurality of feature maps. Wherein having THAWAKAR’s transformer encoder for segmenting images having wherein the exit point is selected based on a lowest-resolution feature map of the plurality of feature maps. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and SHELHAMER are image analysis systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while SHELHAMER the disclosed systems can mask the output of each exit head utilizing confidence thresholding that reduces the remaining computation in future exits. Utilizing this confidence adaptivity approach can further improve efficiency and flexibility of generating predictions at various exit heads without sacrificing accuracy in generating pixel-level predictions. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and SHELHAMER et al. (US 20220309285 A1), Paragraph [0006] . 07-21-aia AIA Claim s 3-4, 6, 13-14, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over THAWAKAR et al. (US 20240161334 A1), hereinafter referenced as THAWAKAR, in view of ZHANG et al. (US 20210056357 A1), hereinafter referenced as ZHANG, and further in view of EHTESHAMI BEJNORDI et al. (US 20220157045 A1), hereinafter referenced as BEJNORDI . Regarding claim 3, THAWAKAR in view of ZHANG explicitly teach the method of claim 1, THAWAKAR in view of ZHANG fail to explicitly teach wherein selecting the exit point is performed using a gating network that includes a pooling layer and a linear layer to select a target layer of the plurality of layers. However, BEJNORDI explicitly teaches wherein selecting the exit point is performed using a gating network (Fig. 6, #600 called gate model. Paragraph [0103]-BEJNORDI discloses FIG. 6 depicts an example gate model 600 that has an efficient design, which beneficially reduces computational overhead.) that includes a pooling layer (Fig. 6, #604A and 604B called pooling components. Paragraph [0104]-BEJNORDI discloses in gate model 600, the input feature maps 602A and 602B are pooled by pooling components 604A and 604B thereby generating intermediate feature maps 606A and 606B (wherein pooling components are pooling layers).) and a linear layer (Fig. 6. Paragraph [0104]-BEJNORDI discloses the resulting intermediate feature maps 610A and 610B are then concatenated by a concatenation component 612 and linearly projected by a linear projection component 614, the output of which is fed to gate decision component 616, which may implement, for example, a Gumbel Softmax with the two possible outcomes described above (wherein linear projection component 614 is a linear layer).) to select a target layer of the plurality of layers (Fig 5, illustrates when early exit does occur (indicated as an arrow proceeding from “GATE 508A, GATE 508B, GATE 508C” to “INTERMEDIATE CLASSIFIED 510A, INTERMEDIATE CLASSIFIED 510B, INTERMEDIATE CLASSIFIED 510C” respectively) and when early exit does not occur (indicated as the arrows leaving “GATE 508A, GATE 508B, GATE 508C” to their respective receiving layers “GATE PRE-PROCESSING 512B, GATE PRE-PROCESSING 512C, GATE PRE-PROCESSING 512D”). Paragraph [0078-0079]-BEJNORDI discloses gate 508A decides whether its associated intermediate classifier 510A should process the frame, which represents an early exit from processing the entire clip 516, or whether model architecture 500 should continue processing additional frame data. If gate 508A decides to early exit, then the feature map is provided to intermediate classifier 510A, which generates model output, such as a classification of objects in the clip 516 based on the frame that has been processed (wherein the intermediate classifier is the target layer).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a computer-implemented method for segmentation, comprising: encoding an image using a backbone model to generate a plurality of feature maps; processing the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decoding an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of BEJNORDI of wherein selecting the exit point is performed using a gating network that includes a pooling layer and a linear layer to select a target layer of the plurality of layers. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein selecting the exit point is performed using a gating network that includes a pooling layer and a linear layer to select a target layer of the plurality of layers. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and BEJNORDI are image analysis systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while BEJNORDI what is needed are improved machine learning architectures that can provide the performance of larger models and the efficiency of smaller models in a single model architecture. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and EHTESHAMI BEJNORDI et al. (US 20220157045 A1), Paragraph [0006]. Regarding claim 4, THAWAKAR in view of ZHANG in view of BEJNORDI explicitly teach the method of claim 3, THAWAKAR in view of ZHANG fail to explicitly teach further comprising retraining the gating network to reflect a change in computational efficiency needs. However, BEJNORDI explicitly teaches further comprising retraining the gating network to reflect a change in computational efficiency needs (Fig. 2. Paragraph [0041]-BEJNORDI discloses gate models (e.g., 204) may be trained to maximize concurrent objectives (subject to tradeoff parameters) of model accuracy and processing sparsity, where sparsity is increased by exiting earlier from a model, and reduced by exiting later (wherein processing sparsity corresponds to computational efficiency).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a computer-implemented method for segmentation, comprising: encoding an image using a backbone model to generate a plurality of feature maps; processing the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decoding an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of BEJNORDI of further comprising retraining the gating network to reflect a change in computational efficiency needs. Wherein having THAWAKAR’s transformer encoder for segmenting images further comprising retraining the gating network to reflect a change in computational efficiency needs. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and BEJNORDI are image analysis systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while BEJNORDI what is needed are improved machine learning architectures that can provide the performance of larger models and the efficiency of smaller models in a single model architecture. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and EHTESHAMI BEJNORDI et al. (US 20220157045 A1), Paragraph [0006]. Regarding claim 6, THAWAKAR in view of ZHANG explicitly teach the method of claim 1, THAWAKAR in view of ZHANG fail to explicitly teach wherein selecting the exit point weighs segmentation quality against computational efficiency to maximize efficiency without sacrificing quality. However, BEJNORDI explicitly teaches wherein selecting the exit point weighs segmentation quality against computational efficiency to maximize efficiency without sacrificing quality (Fig. 2. Paragraph [0041]-BEJNORDI discloses gate models (e.g., 204) may be trained to maximize concurrent objectives (subject to tradeoff parameters) of model accuracy and processing sparsity, where sparsity is increased by exiting earlier from a model, and reduced by exiting later (wherein segmentation quality is how accurate the segmentation is and wherein computational efficiency is processing sparsity).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a computer-implemented method for segmentation, comprising: encoding an image using a backbone model to generate a plurality of feature maps; processing the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decoding an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of BEJNORDI of wherein selecting the exit point weighs segmentation quality against computational efficiency to maximize efficiency without sacrificing quality. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein selecting the exit point weighs segmentation quality against computational efficiency to maximize efficiency without sacrificing quality. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and BEJNORDI are image analysis systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while BEJNORDI what is needed are improved machine learning architectures that can provide the performance of larger models and the efficiency of smaller models in a single model architecture. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and EHTESHAMI BEJNORDI et al. (US 20220157045 A1), Paragraph [0006]. Regarding claim 13, THAWAKAR in view of ZHANG explicitly teach the system of claim 11, THAWAKAR in view of ZHANG fail to explicitly teach wherein the exit point is selected using a gating network that includes a pooling layer and a linear layer to select a target layer of the plurality of layers. However, BEJNORDI explicitly teaches wherein the exit point is selected using a gating network (Fig. 6, #600 called gate model. Paragraph [0103]-BEJNORDI discloses FIG. 6 depicts an example gate model 600 that has an efficient design, which beneficially reduces computational overhead.) that includes a pooling layer (Fig. 6, #604A and 604B called pooling components. Paragraph [0104]-BEJNORDI discloses in gate model 600, the input feature maps 602A and 602B are pooled by pooling components 604A and 604B thereby generating intermediate feature maps 606A and 606B (wherein pooling components are pooling layers).) and a linear layer (Fig. 6. Paragraph [0104]-BEJNORDI discloses the resulting intermediate feature maps 610A and 610B are then concatenated by a concatenation component 612 and linearly projected by a linear projection component 614, the output of which is fed to gate decision component 616, which may implement, for example, a Gumbel Softmax with the two possible outcomes described above (wherein linear projection component 614 is a linear layer).) to select a target layer of the plurality of layers (Fig 5, illustrates when early exit does occur (indicated as an arrow proceeding from “GATE 508A, GATE 508B, GATE 508C” to “INTERMEDIATE CLASSIFIED 510A, INTERMEDIATE CLASSIFIED 510B, INTERMEDIATE CLASSIFIED 510C” respectively) and when early exit does not occur (indicated as the arrows leaving “GATE 508A, GATE 508B, GATE 508C” to their respective receiving layers “GATE PRE-PROCESSING 512B, GATE PRE-PROCESSING 512C, GATE PRE-PROCESSING 512D”). Paragraph [0078-0079]-BEJNORDI discloses gate 508A decides whether its associated intermediate classifier 510A should process the frame, which represents an early exit from processing the entire clip 516, or whether model architecture 500 should continue processing additional frame data. If gate 508A decides to early exit, then the feature map is provided to intermediate classifier 510A, which generates model output, such as a classification of objects in the clip 516 based on the frame that has been processed (wherein the intermediate classifier is the target layer).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a system for segmentation, comprising: a hardware processor; a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to: encode an image using a backbone model to generate a plurality of feature maps; process the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decode an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of BEJNORDI of wherein the exit point is selected using a gating network that includes a pooling layer and a linear layer to select a target layer of the plurality of layers. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein the exit point is selected using a gating network that includes a pooling layer and a linear layer to select a target layer of the plurality of layers. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and BEJNORDI are image analysis systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while BEJNORDI what is needed are improved machine learning architectures that can provide the performance of larger models and the efficiency of smaller models in a single model architecture. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and EHTESHAMI BEJNORDI et al. (US 20220157045 A1), Paragraph [0006]. Regarding claim 14, THAWAKAR in view of ZHANG in view of BEJNORDI explicitly teach the system of claim 13, THAWAKAR in view of ZHANG fail to explicitly teach wherein the computer program further causes the hardware processor to retrain the gating network to reflect a change in computational efficiency needs. However, BEJNORDI explicitly teaches wherein the computer program further causes the hardware processor to retrain the gating network to reflect a change in computational efficiency needs (Fig. 2. Paragraph [0041]-BEJNORDI discloses gate models (e.g., 204) may be trained to maximize concurrent objectives (subject to tradeoff parameters) of model accuracy and processing sparsity, where sparsity is increased by exiting earlier from a model, and reduced by exiting later (wherein processing sparsity corresponds to computational efficiency).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a system for segmentation, comprising: a hardware processor; a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to: encode an image using a backbone model to generate a plurality of feature maps; process the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decode an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of BEJNORDI of wherein the computer program further causes the hardware processor to retrain the gating network to reflect a change in computational efficiency needs. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein the computer program further causes the hardware processor to retrain the gating network to reflect a change in computational efficiency needs. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and BEJNORDI are image analysis systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while BEJNORDI what is needed are improved machine learning architectures that can provide the performance of larger models and the efficiency of smaller models in a single model architecture. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and EHTESHAMI BEJNORDI et al. (US 20220157045 A1), Paragraph [0006]. Regarding claim 16, THAWAKAR in view of ZHANG explicitly teach the system of claim 11, THAWAKAR in view of ZHANG fail to explicitly teach wherein the exit point selection weighs segmentation quality against computational efficiency to maximize efficiency without sacrificing quality. However, BEJNORDI explicitly teaches wherein the exit point selection weighs segmentation quality against computational efficiency to maximize efficiency without sacrificing quality (Fig. 2. Paragraph [0041]-BEJNORDI discloses gate models (e.g., 204) may be trained to maximize concurrent objectives (subject to tradeoff parameters) of model accuracy and processing sparsity, where sparsity is increased by exiting earlier from a model, and reduced by exiting later (wherein segmentation quality is how accurate the segmentation is and wherein computation efficiency is processing sparsity).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a system for segmentation, comprising: a hardware processor; a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to: encode an image using a backbone model to generate a plurality of feature maps; process the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decode an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of BEJNORDI of wherein the exit point selection weighs segmentation quality against computational efficiency to maximize efficiency without sacrificing quality. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein the exit point selection weighs segmentation quality against computational efficiency to maximize efficiency without sacrificing quality. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and BEJNORDI are image analysis systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while BEJNORDI what is needed are improved machine learning architectures that can provide the performance of larger models and the efficiency of smaller models in a single model architecture. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and EHTESHAMI BEJNORDI et al. (US 20220157045 A1), Paragraph [0006] . 07-21-aia AIA Claim s 5 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over THAWAKAR et al. (US 20240161334 A1), hereinafter referenced as THAWAKAR, in view of ZHANG et al. (US 20210056357 A1), hereinafter referenced as ZHANG, and further in view of ZHOU et al. (US 20230376729 A1), hereinafter referenced as ZHOU . Regarding claim 5, THAWAKAR in view of ZHANG explicitly teach the method of claim 1, THAWAKAR further explicitly teaches the decoding is performed by a visual transformer decoder (Fig. 3B, #330 called transformer decoder. Paragraph [0050]-THAWAKAR discloses these encoder output features maps from each frame along with n learnable instance query embeddings I.sup.Q ∈ custom-character.sup.C are then input to the transformer decoder comprising a series of self- and cross-attention blocks. The n instance queries are further decomposed into n box queries B.sup.Q per-frame and are used to query the box features from the encoder feature maps of the corresponding frame. The learned box queries across T frames are then aggregated temporally to obtain n instance features I.sup.O ∈ custom-character.sup.C. These instance features output by the decoder are then used for video instance mask prediction (wherein the transformer decoder is a visual transformer decoder and wherein the transformer decoder performs decoding to obtain instance features I.sup.O).) . THAWAKAR in view of ZHANG fail to explicitly teach wherein the backbone model is a visual transformer encoder. However, ZHOU explicitly teaches wherein the backbone model is a visual transformer encoder (Fig. 3. Paragraph [0047]-ZHOU discloses a Vision Transformer (ViT) is used as an example as the encoder 301 but other architectures, such as CNN and MLP-based networks, can also be implemented as the encoder 301 in TAE. Accordingly, in the following, it is described how a ViT backbone network can be trained as an encoder within a TAE framework (wherein a ViT is a visual transformer encoder and wherein a backbone network is backbone model).) and Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a computer-implemented method for segmentation, comprising: encoding an image using a backbone model to generate a plurality of feature maps; processing the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decoding an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of ZHOU of wherein the backbone model is a visual transformer encoder. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein the backbone model is a visual transformer encoder. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and ZHOU relate to and employ visual transformers, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while ZHOU approaches for training a neural network are desirable which achieve good results without masking of the encoder input. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and ZHOU et al. (US 20230376729 A1), Paragraph [0007]. Regarding claim 15, THAWAKAR in view of ZHANG explicitly teach the system of claim 11, THAWAKAR further explicitly teaches the decoding is performed by a visual transformer decoder (Fig. 3B, #330 called transformer decoder. Paragraph [0050]-THAWAKAR discloses these encoder output features maps from each frame along with n learnable instance query embeddings I.sup.Q ∈ custom-character.sup.C are then input to the transformer decoder comprising a series of self- and cross-attention blocks. The n instance queries are further decomposed into n box queries B.sup.Q per-frame and are used to query the box features from the encoder feature maps of the corresponding frame. The learned box queries across T frames are then aggregated temporally to obtain n instance features I.sup.O ∈ custom-character.sup.C. These instance features output by the decoder are then used for video instance mask prediction (wherein the transformer decoder is a visual transformer decoder and wherein the transformer decoder performs decoding to obtain instance features I.sup.O).) . THAWAKAR in view of ZHANG fail to explicitly teach wherein the backbone model is a visual transformer encoder (Fig. 3. Paragraph [0047]-ZHOU discloses a Vision Transformer (ViT) is used as an example as the encoder 301 but other architectures, such as CNN and MLP-based networks, can also be implemented as the encoder 301 in TAE. Accordingly, in the following, it is described how a ViT backbone network can be trained as an encoder within a TAE framework (wherein a ViT is a visual transformer encoder and wherein a backbone network is backbone model).) and. However, ZHOU explicitly teaches wherein the backbone model is a visual transformer encoder and (Fig. 3. Paragraph [0047]-ZHOU discloses a Vision Transformer (ViT) is used as an example as the encoder 301 but other architectures, such as CNN and MLP-based networks, can also be implemented as the encoder 301 in TAE. Accordingly, in the following, it is described how a ViT backbone network can be trained as an encoder within a TAE framework (wherein a ViT is a visual transformer encoder and wherein a backbone network is backbone model).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a system for segmentation, comprising: a hardware processor; a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to: encode an image using a backbone model to generate a plurality of feature maps; process the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decode an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of ZHOU of wherein the backbone model is a visual transformer encoder. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein the backbone model is a visual transformer encoder. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and ZHOU relate to and employ visual transformers, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while ZHOU approaches for training a neural network are desirable which achieve good results without masking of the encoder input. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and ZHOU et al. (US 20230376729 A1), Paragraph [0007] . 07-21-aia AIA Claim s 7 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over THAWAKAR et al. (US 20240161334 A1), hereinafter referenced as THAWAKAR, in view of ZHANG et al. (US 20210056357 A1), hereinafter referenced as ZHANG, and further in view of KANAKATTE GURUMURTHY et al. (US 20240005512 A1), hereinafter referenced as GURUMURTHY . Regarding claim 7, THAWAKAR in view of ZHANG explicitly teach the method of claim 1, THAWAKAR further explicitly teaches wherein the plurality of feature maps include feature maps of different resolutions (Fig. 4. Paragraph [0054]-THAWAKAR discloses the MS-STS module 310 takes the backbone features 306 as input and produces multi-scale spatio-temporally enriched features 406, which are then fused with the standard features within the base framework (wherein multi-scale features are feature maps of different resolutions).) and THAWAKAR in view of ZHANG fail to explicitly teach wherein decoding the output includes generating respective segmentations for each of the plurality of feature maps. However, GURUMURTHY explicitly teaches wherein decoding the output includes generating respective segmentations for each of the plurality of feature maps (Fig. 2A-2C. Paragraph [0059]-GURUMURTHY discloses at step 208c, the concatenated feature map of each pre-processed training image, obtained at step 208b, is passed to the decoder network 302c, to predict a segmented image corresponding to each pre-processed training image (wherein predict a segmented image is generating respective segmentation and where the concatenated feature map of each pre-processed training image is each of the plurality of feature maps).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a computer-implemented method for segmentation, comprising: encoding an image using a backbone model to generate a plurality of feature maps; processing the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decoding an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of GURUMURTHY of wherein decoding the output includes generating respective segmentations for each of the plurality of feature maps. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein decoding the output includes generating respective segmentations for each of the plurality of feature maps. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and GURUMURTHY relate to encoding and decoding images through the use of neural networks, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while GURUMURTHY there are some techniques exist that deals with the 3-D images for the segmentation task. However, most of the techniques are inaccurate and inefficient in the image segmentation till the last slice of the image. Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and KANAKATTE GURUMURTHY et al. (US 20240005512 A1), Paragraph [0006-0007]. Regarding claim 17, THAWAKAR in view of ZHANG explicitly teach the system of claim 11, THAWAKAR further explicitly teaches wherein the plurality of feature maps include feature maps of different resolutions (Fig. 4. Paragraph [0054]-THAWAKAR discloses the MS-STS module 310 takes the backbone features 306 as input and produces multi-scale spatio-temporally enriched features 406, which are then fused with the standard features within the base framework (wherein multi-scale features are feature maps of different resolutions).) and THAWAKAR in view of ZHANG fail to explicitly teach wherein the computer program further causes the hardware processor to generate respective segmentations for each of the plurality of feature maps. However, GURUMURTHY explicitly teaches wherein the computer program further causes the hardware processor to generate respective segmentations for each of the plurality of feature maps (Fig. 2A-2C. Paragraph [0059]-GURUMURTHY discloses at step 208c, the concatenated feature map of each pre-processed training image, obtained at step 208b, is passed to the decoder network 302c, to predict a segmented image corresponding to each pre-processed training image (wherein predict a segmented image is generating respective segmentation and where the concatenated feature map of each pre-processed training image is each of the plurality of feature maps).) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a system for segmentation, comprising: a hardware processor; a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to: encode an image using a backbone model to generate a plurality of feature maps; process the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decode an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of GURUMURTHY of wherein the computer program further causes the hardware processor to generate respective segmentations for each of the plurality of feature maps. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein the computer program further causes the hardware processor to generate respective segmentations for each of the plurality of feature maps. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and GURUMURTHY relate to encoding and decoding images through the use of neural networks, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while GURUMURTHY there are some techniques exist that deals with the 3-D images for the segmentation task. However, most of the techniques are inaccurate and inefficient in the image segmentation till the last slice of the image. Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and KANAKATTE GURUMURTHY et al. (US 20240005512 A1), Paragraph [0006-0007] . 07-21-aia AIA Claim s 9-10 and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over THAWAKAR et al. (US 20240161334 A1), hereinafter referenced as THAWAKAR, in view of ZHANG et al. (US 20210056357 A1), hereinafter referenced as ZHANG, and further in view LEVI et al. (US 20220035378 A1), hereinafter referenced as LEVI . Regarding claim 9, THAWAKAR in view of ZHANG explicitly teach the method of claim 1, THAWAKAR in view of ZHANG fail to explicitly teach further comprising controlling an autonomous vehicle responsive to the segmentation to avoid an obstacle or hazard in a scene shown by the image. However, LEVI explicitly teaches further comprising controlling an autonomous vehicle (Fig. 1. Paragraph [0133]-LEVI discloses system 100 may provide a variety of features related to autonomous driving and/or driver assist technology. Further in paragraph [0133]-LEVI discloses when vehicle 200 navigates without human intervention, system 100 may automatically control the braking, acceleration, and/or steering of vehicle 200 (e.g., by sending control signals to one or more of throttling system 220, braking system 230, and steering system 240).) responsive to the segmentation (Fig. 5A-5F, illustrate performing image segmentation to assist an autonomous vehicle navigation. Paragraph [0150]-LEVI discloses navigational response module 408 may store software executable by processing unit 110 to determine a desired navigational response based on data derived from execution of monocular image analysis module 402 and/or stereo image analysis module 404 (wherein the image analysis modules perform image segmentation).) to avoid an obstacle or hazard (Fig. 5A-5F. Paragraph [0153]-LEVI discloses processing unit 110 may also execute monocular image analysis module 402 to detect various road hazards at step 520, such as, for example, parts of a truck tire, fallen road signs, loose cargo, small animals, and the like. Road hazards may vary in structure, shape, size, and color, which may make detection of such hazards more challenging. In some embodiments, processing unit 110 may execute monocular image analysis module 402 to perform multi-frame analysis on the plurality of images to detect road hazards (wherein hazards are obstacles).) in a scene shown by the image (Fig. 5A-5B. Paragraph [0152]-LEVI discloses at step 510, processing unit 110 may receive a plurality of images via data interface 128 between processing unit 110 and image acquisition unit 120. For instance, a camera included in image acquisition unit 120 (such as image capture device 122 having field of view 202) may capture a plurality of images of an area forward of vehicle 200.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a computer-implemented method for segmentation, comprising: encoding an image using a backbone model to generate a plurality of feature maps; processing the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decoding an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of LEVI of further comprising controlling an autonomous vehicle responsive to the segmentation to avoid an obstacle or hazard in a scene shown by the image. Wherein having THAWAKAR’s transformer encoder for segmenting images further comprising controlling an autonomous vehicle responsive to the segmentation to avoid an obstacle or hazard in a scene shown by the image. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and LEVI are computer vision systems and are systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while LEVI the sheer quantity of data (e.g., captured image data, map data, GPS data, sensor data, etc.) that an autonomous vehicle may need to analyze, access, and/or store poses challenges that can in fact limit or even adversely affect autonomous navigation. Furthermore, if an autonomous vehicle relies on traditional mapping technology to navigate, the sheer volume of data needed to store and update the map poses daunting challenges. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and LEVI et al. (US 20220035378 A1), Paragraph [0003]. Regarding claim 10, THAWAKAR in view of ZHANG and further in view of LEVI explicitly teach the method of claim 9, THAWAKAR in view of ZHANG fail to explicitly teach wherein controlling the autonomous vehicle includes performing a steering, accelerating, or decelerating action. However, LEVI explicitly teaches wherein controlling the autonomous vehicle includes performing a steering, accelerating, or decelerating action (Fig. 5A, #530 called “CAUSE ONE OR MORE NAVIGATIONAL RESPONSES BASED ON THE MONOCULAR IMAGE ANALYSIS”. Paragraph [0154]-LEVI discloses multiple navigational responses may occur simultaneously, in sequence, or any combination thereof. For instance, processing unit 110 may cause vehicle 200 to shift one lane over and then accelerate by, for example, sequentially transmitting control signals to steering system 240 and throttling system 220 of vehicle 200. Alternatively, processing unit 110 may cause vehicle 200 to brake while at the same time shifting lanes by, for example, simultaneously transmitting control signals to braking system 230 and steering system 240 of vehicle 200.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a computer-implemented method for segmentation, comprising: encoding an image using a backbone model to generate a plurality of feature maps; processing the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decoding an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of LEVI of wherein controlling the autonomous vehicle includes performing a steering, accelerating, or decelerating action. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein controlling the autonomous vehicle includes performing a steering, accelerating, or decelerating action. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and LEVI are computer vision systems and are systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while LEVI the sheer quantity of data (e.g., captured image data, map data, GPS data, sensor data, etc.) that an autonomous vehicle may need to analyze, access, and/or store poses challenges that can in fact limit or even adversely affect autonomous navigation. Furthermore, if an autonomous vehicle relies on traditional mapping technology to navigate, the sheer volume of data needed to store and update the map poses daunting challenges. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and LEVI et al. (US 20220035378 A1), Paragraph [0003]. Regarding claim 19, THAWAKAR in view of ZHANG explicitly teach the system of claim 11, THAWAKAR in view of ZHANG fail to explicitly teach wherein the computer program further causes the hardware processor to perform a steering, accelerating, or decelerating action on an autonomous vehicle responsive to the segmentation to avoid an obstacle or hazard in a scene shown by the image. However, LEVI explicitly teaches wherein the computer program further causes the hardware processor to perform a steering, accelerating, or decelerating action on an autonomous vehicle (Fig. 5A, #530 called “CAUSE ONE OR MORE NAVIGATIONAL RESPONSES BASED ON THE MONOCULAR IMAGE ANALYSIS”. Paragraph [0154]-LEVI discloses multiple navigational responses may occur simultaneously, in sequence, or any combination thereof. For instance, processing unit 110 may cause vehicle 200 to shift one lane over and then accelerate by, for example, sequentially transmitting control signals to steering system 240 and throttling system 220 of vehicle 200. Alternatively, processing unit 110 may cause vehicle 200 to brake while at the same time shifting lanes by, for example, simultaneously transmitting control signals to braking system 230 and steering system 240 of vehicle 200.) responsive to the segmentation (Fig. 5A-5F, illustrate performing image segmentation to assist an autonomous vehicle navigation. Paragraph [0150]-LEVI discloses navigational response module 408 may store software executable by processing unit 110 to determine a desired navigational response based on data derived from execution of monocular image analysis module 402 and/or stereo image analysis module 404 (wherein the image analysis modules perform image segmentation).) to avoid an obstacle or hazard (Fig. 5A-5F. Paragraph [0153]-LEVI discloses processing unit 110 may also execute monocular image analysis module 402 to detect various road hazards at step 520, such as, for example, parts of a truck tire, fallen road signs, loose cargo, small animals, and the like. Road hazards may vary in structure, shape, size, and color, which may make detection of such hazards more challenging. In some embodiments, processing unit 110 may execute monocular image analysis module 402 to perform multi-frame analysis on the plurality of images to detect road hazards (wherein hazards are obstacles).) in a scene shown by the image (Fig. 5A-5B. Paragraph [0152]-LEVI discloses at step 510, processing unit 110 may receive a plurality of images via data interface 128 between processing unit 110 and image acquisition unit 120. For instance, a camera included in image acquisition unit 120 (such as image capture device 122 having field of view 202) may capture a plurality of images of an area forward of vehicle 200.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of a system for segmentation, comprising: a hardware processor; a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to: encode an image using a backbone model to generate a plurality of feature maps; process the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; and decode an output of the dynamic transformer encoder to output a segmentation of the image with the teachings of LEVI of wherein the computer program further causes the hardware processor to perform a steering, accelerating, or decelerating action on an autonomous vehicle responsive to the segmentation to avoid an obstacle or hazard in a scene shown by the image. Wherein having THAWAKAR’s transformer encoder for segmenting images wherein the computer program further causes the hardware processor to perform a steering, accelerating, or decelerating action on an autonomous vehicle responsive to the segmentation to avoid an obstacle or hazard in a scene shown by the image. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and LEVI are computer vision systems and are systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while LEVI the sheer quantity of data (e.g., captured image data, map data, GPS data, sensor data, etc.) that an autonomous vehicle may need to analyze, access, and/or store poses challenges that can in fact limit or even adversely affect autonomous navigation. Furthermore, if an autonomous vehicle relies on traditional mapping technology to navigate, the sheer volume of data needed to store and update the map poses daunting challenges. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and LEVI et al. (US 20220035378 A1), Paragraph [0003]. Regarding claim 20, THAWAKAR discloses an autonomous vehicle (Fig. 2. Paragraph [0042]-THAWAKAR discloses FIG. 2 is non-limiting exemplary computer vision system in the context of a vehicle. The vehicle can have camera-based driver assistance and various levels of automated driving, including partial automation, conditional automation, high automation, and full automation.) , comprising: a camera (Fig. 2, #102 called video cameras. Paragraph [0043]-THAWAKAR discloses a vehicle may include one or more video cameras 102, on one side or multiple sides of the vehicle body. Cameras may also be mounted at other locations, including the vehicle roof, adjacent to a windshield, inside the vehicle facing through a window, to name a few.) that captures an image of a scene (Fig. 12, illustrates capturing an image of a scene. Paragraph [0077]-THAWAKAR discloses FIG. 12 illustrates a computer vision task in the case of a road sign being viewed by a camera on a moving vehicle. Road signs are often signs that display a symbol and may be displayed in positions along the side of a road or in overhead displays that cross over a road. Although the road sign itself is stationary, the vehicle, with the video camera 1221 mounted thereon, is moving at a speed and direction 1220 such that the sequence of images of the road sign may be obtained at different angles, at a range of distances and perspectives, over a short time period.) ; a hardware processor (Fig. 11, #1100 called AI SoC. Paragraph [0071]-THAWAKAR discloses SOC 1100 may be a Qualcomm mobile processor, a Nvidia DRIVE processor, a Atom® processor from Intel Corporation of America, a Samsung mobile processor, or a Apple A mobile processor, or may be other processor types that would be recognized by one of ordinary skill in the art. Alternatively, the SOC 1100 may be implemented on an Field-Programmable Gate Array (FPGA), Application Specific Integrated Circuit (ASIC), Programmable Logic Device (PLD) or using discrete logic circuits, as one of ordinary skill in the art would recognize. Further, SOC 1100 may be implemented as multiple processors cooperatively working in parallel to perform the instructions of the inventive processes described above.) ; a memory (Fig. 11, #1102 called SDRAM Flash. Paragraph [0069]-THAWAKAR discloses the instructions may be stored in FLASH memory, Secure Digital Random Access Memory (SDRAM), Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read Only Memory (EEPROM), solid-state hard disk or any other information processing device with which the processing circuit 1126 communicates, such as a server or computer.) that stores a computer program (Fig. 11. Paragraph [0068]-THAWAKAR discloses the processing circuit 1126 includes an AI System on a Chip (SOC) 1100 which performs the processes described herein. The process data and instructions may be stored in memory 1102. These processes and instructions may also be stored on a portable storage medium or may be stored remotely (wherein instructions is a computer program).) which, when executed by the hardware processor, causes the hardware processor to (Fig. 11. Paragraph [0068]-THAWAKAR discloses the processing circuit 1126 includes an AI System on a Chip (SOC) 1100 which performs the processes described herein. The process data and instructions may be stored in memory 1102.) : encode an image (Fig. 3A-3B, #302 called video frames. Paragraph [0049]-THAWAKAR discloses a video clip x consisting T frames 302 of spatial size H.sup.0×W.sup.0 with a set of object instances is input to the backbone network 304. Latent feature maps for each frame are obtained from the backbone 304 at L multiple scales 2.sup.−l−2 with 1≤l≤L and passed through separate convolution filters for each scale. The output feature dimension for each of these convolution filters is set to C (wherein video frames are images and wherein obtaining feature maps is encoding an image).) using a backbone model (Fig. 3A, #304 called Backbone. Paragraph [0049]-THAWAKAR discloses a video clip x consisting T frames 302 of spatial size H.sup.0×W.sup.0 with a set of object instances is input to the backbone network 304. Latent feature maps for each frame are obtained from the backbone 304 at L multiple scales 2.sup.−l−2 with 1≤l≤L and passed through separate convolution filters for each scale.) to generate a plurality of feature maps (Fig. 3A-3B. Paragraph [0049]-THAWAKAR discloses latent feature maps for each frame are obtained from the backbone 304 at L multiple scales.) ; process the feature maps with a dynamic transformer encoder (Fig. 3A-3B. Paragraph [0049]-THAWAKAR discloses latent feature maps for each frame are obtained from the backbone 304 at L multiple scales 2.sup.−l−2 with 1≤l≤L and passed through separate convolution filters for each scale. The output feature dimension for each of these convolution filters is set to C. The resulting multi-scale feature maps of each frame are then input to a transformer encoder comprising multi-scale deformable attention blocks 314 (wherein the transformer encoder is a dynamic transformer encoder).) that includes a plurality of layers (Fig. 4. Paragraph [0052]-THAWAKAR discloses these features {circumflex over (Z)} 406 are fused in each encoder layer with the base features output by the conventional deformable attention 314. While backbone features F 306 are used as input to the first encoder layer, the subsequent layers utilize the outputs from the preceding layer as input. As a result, multi-scale spatio-temporally enriched features C-dimensional E 316 are output by the encoder 310.) , exiting the dynamic transformer encoder at a layer identified by the exit point (Fig. 3A, #316 called enriched features E indicate an exit of the transformer encoder. Paragraph [0052]-ZHANG discloses while backbone features F 306 are used as input to the first encoder layer, the subsequent layers utilize the outputs from the preceding layer as input. As a result, multi-scale spatio-temporally enriched features C-dimensional E 316 are output by the encoder 310 (wherein outputting is exiting). Please see annotated Fig. 3A below.) ; and decode an output of the dynamic transformer encoder (Fig. 3A-3B, block #330 illustrates the process of decoding in the decoder. Paragraph [0050]-THAWAKAR discloses these encoder output features maps from each frame along with n learnable instance query embeddings are then input to the transformer decoder comprising a series of self- and cross-attention blocks (wherein the decoder conducts decoding).) to output a segmentation of the image (Fig. 3A-3B. Paragraph [0050]-THAWAKAR discloses the learned box queries across T frames are then aggregated temporally to obtain n instance features I.sup.O ∈ custom-character.sup.C. These instance features output by the decoder are then used for video instance mask prediction. Further in paragraph [0053]-THAWAKAR disclose the encoder features E 316 along with the instance features I.sup.O 334 (aggregated temporally attended box queries) from the decoder 330 are utilized within the instance matching and segmentation block 340 to obtain the video instance mask prediction 342 (wherein the instance mask prediction is a segmentation of the image).) . PNG media_image1.png 578 789 media_image1.png Greyscale Annotated diagram of THAWAKAR’s Fig. 3A indicating the layers of the transformer encoder and the identified exit layer. THAWAKAR fails to explicitly teach select an exit point based on one of the plurality of feature maps. However, ZHANG explicitly teaches select an exit point (Fig. 15, #1510 called decision module and “yes” indicates an early exit result (i.e. an exit point). Paragraph [0105]-ZHANG discloses the decision module 1510 then takes the early prediction results generated by the early exit branch 1508 and makes a decision on whether to exit the inference process and output the early prediction results, or to continue the inference process and pass the generated feature maps to the next layer (wherein to exit the inference process is selecting an exit point).) based on one of the plurality of feature maps (Fig. 15. Paragraph [0105]-ZHANG discloses the smaller early exit branch network merely takes the intermediate features generated by the j.sup.th internal convolutional layers 1512-1514 of the base model and transforms them into early predictions. The decision module 1510 then takes the early prediction results generated by the early exit branch 1508 and makes a decision on whether to exit the inference process and output the early prediction results, or to continue the inference process and pass the generated feature maps to the next layer. In this way, a “base” model (which may have already been trained on a training dataset) can be modified to make it flexible and adaptive to the “difficulty” of the input data, without the memory requirements of a “bag of models” approach or the inefficiency of a “coarse” approach (wherein the features are feature maps).) ; Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR of an autonomous vehicle, comprising: a camera that captures an image of a scene; a hardware processor; and a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to: encode an image using a backbone model to generate a plurality of feature maps; process the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; decode an output of the dynamic transformer encoder to output a segmentation of the image; and with the teachings of ZHANG of select an exit point based on one of the plurality of feature maps. Wherein having THAWAKAR’s transformer encoder for segmenting images having select an exit point based on one of the plurality of feature maps. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and ZHANG are computer vision systems and are systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while ZHANG the systems and methods herein implement various features and aspects that make them useful for any situation in which it would be beneficial to reduce computational demand of a deep neural network application where the input data set has varying complexity. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and ZHANG et al. (US 20210056357 A1), Paragraph [0054]. THAWAKAR in view of ZHANG fail to explicitly teach perform a steering, accelerating, or decelerating action responsive to the segmentation to avoid an obstacle or hazard in the scene shown by the image. However, LEVI explicitly teaches perform a steering, accelerating, or decelerating action (Fig. 5A, #530 called “CAUSE ONE OR MORE NAVIGATIONAL RESPONSES BASED ON THE MONOCULAR IMAGE ANALYSIS”. Paragraph [0154]-LEVI discloses multiple navigational responses may occur simultaneously, in sequence, or any combination thereof. For instance, processing unit 110 may cause vehicle 200 to shift one lane over and then accelerate by, for example, sequentially transmitting control signals to steering system 240 and throttling system 220 of vehicle 200. Alternatively, processing unit 110 may cause vehicle 200 to brake while at the same time shifting lanes by, for example, simultaneously transmitting control signals to braking system 230 and steering system 240 of vehicle 200.) responsive to the segmentation (Fig. 5A-5F, illustrate performing image segmentation to assist an autonomous vehicle navigation. Paragraph [0150]-LEVI discloses navigational response module 408 may store software executable by processing unit 110 to determine a desired navigational response based on data derived from execution of monocular image analysis module 402 and/or stereo image analysis module 404 (wherein the image analysis modules perform image segmentation).) to avoid an obstacle or hazard (Fig. 5A-5F. Paragraph [0153]-LEVI discloses processing unit 110 may also execute monocular image analysis module 402 to detect various road hazards at step 520, such as, for example, parts of a truck tire, fallen road signs, loose cargo, small animals, and the like. Road hazards may vary in structure, shape, size, and color, which may make detection of such hazards more challenging. In some embodiments, processing unit 110 may execute monocular image analysis module 402 to perform multi-frame analysis on the plurality of images to detect road hazards (wherein hazards are obstacles).) in a scene shown by the image (Fig. 5A-5B. Paragraph [0152]-LEVI discloses at step 510, processing unit 110 may receive a plurality of images via data interface 128 between processing unit 110 and image acquisition unit 120. For instance, a camera included in image acquisition unit 120 (such as image capture device 122 having field of view 202) may capture a plurality of images of an area forward of vehicle 200.) . Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of THAWAKAR in view of ZHANG of an autonomous vehicle, comprising: a camera that captures an image of a scene; a hardware processor; and a memory that stores a computer program which, when executed by the hardware processor, causes the hardware processor to: encode an image using a backbone model to generate a plurality of feature maps; process the feature maps with a dynamic transformer encoder that includes a plurality of layers, exiting the dynamic transformer encoder at a layer identified by the exit point; decode an output of the dynamic transformer encoder to output a segmentation of the image; and with the teachings of LEVI of perform a steering, accelerating, or decelerating action responsive to the segmentation to avoid an obstacle or hazard in the scene shown by the image. Wherein having THAWAKAR’s transformer encoder for segmenting images having perform a steering, accelerating, or decelerating action responsive to the segmentation to avoid an obstacle or hazard in the scene shown by the image. The motivation behind the modification would have been to obtain an image segmenting transformer encoder that enhances segmentation quality and accuracy and reduces the computation load required by the system. Since both THAWAKAR and LEVI are computer vision systems and are systems employable by autonomous vehicles, wherein THAWAKAR the MS-STS VIS 300 addresses these issues by capturing multi-scale spatio-temporal feature relationships, leading to improved mask quality, while LEVI the sheer quantity of data (e.g., captured image data, map data, GPS data, sensor data, etc.) that an autonomous vehicle may need to analyze, access, and/or store poses challenges that can in fact limit or even adversely affect autonomous navigation. Furthermore, if an autonomous vehicle relies on traditional mapping technology to navigate, the sheer volume of data needed to store and update the map poses daunting challenges. Please see THAWAKAR et al. (US 20240161334 A1), Paragraph [0064], and LEVI et al. (US 20220035378 A1), Paragraph [0003]. Conclusion Listed below are the prior arts made of record and not relied upon but are considered pertinent to applicant’s disclosure. KOURIS et al. (US 20230128637 A1) - Broadly speaking, the present techniques generally relate to a method for training a machine learning, ML, model to perform semantic image segmentation, and to a computer-implemented method and apparatus for performing semantic image segmentation using a trained machine learning, ML, model. The training method enables a semantic image segmentation ML model that is able to make predictions faster, without significant loss in accuracy. The training method also enables the ML model to be implemented on apparatus with different hardware specifications, i.e. different computational power and memory, for example… Abstract, Fig. 3A-3C. DAS et al. (US 20230127001 A1) - A method for generating an optimal neural network (NN) model may include determining intermediate outputs of the NN model by passing an input dataset through each intermediate exit gate of the plurality of intermediate exit gates, determining an accuracy score for each intermediate exit gate of the plurality of intermediate exit gates based on a comparison of the final output of the NN model with the intermediate output, identifying an earliest intermediate exit gate that produces the intermediate output closer to the final output based on the accuracy score, and generating the optimal NN model by removing remaining layers of the plurality of layers and remaining intermediate exit gates of the plurality of intermediate exit gates located after the determined earliest intermediate exit gate… Abstract, Fig. 6. Any inquiry concerning this communication or earlier communications from the examiner should be directed to ETHAN N WOLFSON whose telephone number is (571)272-1898. 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, Chineyere Wills-Burns can be reached at (571) 272-9752. 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. /ETHAN N WOLFSON/Examiner, Art Unit 2673 /CHINEYERE WILLS-BURNS/Supervisory Patent Examiner, Art Unit 2673 Application/Control Number: 18/893,037 Page 2 Art Unit: 2673 Application/Control Number: 18/893,037 Page 3 Art Unit: 2673
Prosecution Insights