Patents by Inventor Szabolcs Cséfalvay
Szabolcs Cséfalvay has filed for patents to protect the following inventions. This listing includes patent applications that are pending as well as patents that have already been granted by the United States Patent and Trademark Office (USPTO).
-
Publication number: 20240135139Abstract: Methods and systems for implementing a traditional computer vision algorithm as a neural network. The method includes: receiving a definition of the traditional computer vision algorithm that identifies a sequence of one or more traditional computer vision algorithm operations; mapping each of the one or more traditional computer vision algorithm operations to a set of one or more neural network primitives that is mathematically equivalent to that traditional computer vision algorithm operation; linking the one or more network primitives mapped to each traditional computer vision algorithm operation according to the sequence to form a neural network representing the traditional computer vision algorithm; and configuring hardware logic capable of implementing a neural network to implement the neural network that represents the traditional computer vision algorithm.Type: ApplicationFiled: April 19, 2023Publication date: April 25, 2024Inventors: Paul Brasnett, Daniel Valdez Balderas, Cagatay Dikici, Szabolcs Csefalvay, David Hough, Timothy Smith, James Imber
-
Publication number: 20240135153Abstract: A computer-implemented method of processing data using a Neural Network (NN) implemented in hardware, the NN having a plurality of layers, each layer being configured to operate on activation data input to the layer so as to form output data for the layer, said data being arranged in data channels, the method comprising: for an identified channel of output data for a layer, operating on activation data input to the layer such that the output data for the layer does not include the identified channel; and prior to an operation of the NN configured to operate on the output data for the layer, inserting a replacement channel into the output data for the layer in lieu of the identified channel in dependence on information indicative of the structure of the output data for the layer were the identified channel to have been included.Type: ApplicationFiled: June 29, 2023Publication date: April 25, 2024Inventor: Szabolcs Csefalvay
-
Patent number: 11961286Abstract: A single-instruction, multiple data processor performs object detection in an image by testing for a plurality of object features in a plurality of image regions, the processor comprising: a set of computation units operable to execute a plurality of classifier sequences in parallel, each classifier sequence comprising a plurality of classifier routines, and each classifier routine comprising identical instructions to the other classifier routines in each of the plurality of classifier sequences; wherein each computation unit is configured to independently maintain data identifying an image region and a feature under test on that computation unit, and each classifier routine is arranged to access the data, test the identified feature against the identified image region and update the data such that the computation units are operable to concurrently test different features against different image regions.Type: GrantFiled: February 24, 2020Date of Patent: April 16, 2024Assignee: Imagination Technologies LimitedInventor: Szabolcs Csefalvay
-
Patent number: 11922321Abstract: Methods and systems for identifying quantisation parameters for a Deep Neural Network (DNN). The method includes determining an output of a model of the DNN in response to training data, the model of the DNN comprising one or more quantisation blocks configured to transform a set of values input to a layer of the DNN prior to processing the set of values in accordance with the layer, the transformation of the set of values simulating quantisation of the set of values to a fixed point number format defined by one or more quantisation parameters; determining a cost metric of the DNN based on the determined output and a size of the DNN based on the quantisation parameters; back-propagating a derivative of the cost metric to one or more of the quantisation parameters to generate a gradient of the cost metric for each of the one or more quantisation parameters; and adjusting one or more of the quantisation parameters based on the gradients.Type: GrantFiled: March 9, 2023Date of Patent: March 5, 2024Assignee: Imagination Technologies LimitedInventor: Szabolcs Csefalvay
-
Patent number: 11915397Abstract: A method of rendering an image of a 3-D scene includes rendering a noisy image; and obtaining one or more guide channels. For each of a plurality of local neighbourhoods, the method comprises: calculating the parameters of a model that approximates the noisy image as a function of the one or more guide channels, and applying the calculated parameters to produce a denoised image. Tiling is used when calculating the parameters of the model.Type: GrantFiled: September 30, 2022Date of Patent: February 27, 2024Assignee: Imagination Technologies LimitedInventors: Szabolcs Csefalvay, James Imber, David Walton, Insu Yu
-
Patent number: 11915396Abstract: A pixel filter has a filter module that performs a first recursive filter operation in a first direction through a sequence of pixels to form a first filtered pixel value for each pixel, and performs a second recursive filter operation in a second direction through the sequence of pixels to form a second filtered pixel value for each pixel, the first and second recursive filter operations forming a respective filtered pixel value for a given pixel in dependence on the pixel value at that pixel and the filtered pixel value preceding that pixel in their respective direction of operation. The filtered pixel value of the preceding pixel is scaled by a measure of similarity between data associated with that pixel and its preceding pixel. Filter logic combines the first and second filtered pixel values formed by the first and second recursive filter operations to generate a filter output for the pixel, for each pixel of the sequence.Type: GrantFiled: June 9, 2022Date of Patent: February 27, 2024Assignee: Imagination Technologies LimitedInventor: Szabolcs Csefalvay
-
Publication number: 20230214659Abstract: Methods and systems for identifying quantisation parameters for a Deep Neural Network (DNN). The method includes determining an output of a model of the DNN in response to training data, the model of the DNN comprising one or more quantisation blocks configured to transform a set of values input to a layer of the DNN prior to processing the set of values in accordance with the layer, the transformation of the set of values simulating quantisation of the set of values to a fixed point number format defined by one or more quantisation parameters; determining a cost metric of the DNN based on the determined output and a size of the DNN based on the quantisation parameters; back-propagating a derivative of the cost metric to one or more of the quantisation parameters to generate a gradient of the cost metric for each of the one or more quantisation parameters; and adjusting one or more of the quantisation parameters based on the gradients.Type: ApplicationFiled: March 9, 2023Publication date: July 6, 2023Inventor: Szabolcs Csefalvay
-
Publication number: 20230177769Abstract: A method of rendering an image of a 3-D scene includes rendering a noisy image; and obtaining one or more guide channels. For each of a plurality of local neighbourhoods, the method comprises: calculating the parameters of a model that approximates the noisy image as a function of the one or more guide channels, and applying the calculated parameters to produce a denoised image. Tiling is used when calculating the parameters of the model.Type: ApplicationFiled: September 30, 2022Publication date: June 8, 2023Inventors: Szabolcs Csefalvay, James Imber, David Walton, Insu Yu
-
Patent number: 11636306Abstract: Methods and systems for implementing a traditional computer vision algorithm as a neural network. The method includes: receiving a definition of the traditional computer vision algorithm that identifies a sequence of one or more traditional computer vision algorithm operations; mapping each of the one or more traditional computer vision algorithm operations to a set of one or more neural network primitives that is mathematically equivalent to that traditional computer vision algorithm operation; linking the one or more network primitives mapped to each traditional computer vision algorithm operation according to the sequence to form a neural network representing the traditional computer vision algorithm; and configuring hardware logic capable of implementing a neural network to implement the neural network that represents the traditional computer vision algorithm.Type: GrantFiled: May 21, 2019Date of Patent: April 25, 2023Assignee: Imagination Technologies LimitedInventors: Paul Brasnett, Daniel Valdez Balderas, Cagatay Dikici, Szabolcs Cséfalvay, David Hough, Timothy Smith, James Imber
-
Publication number: 20230118937Abstract: A method of rendering an image of a 3-D scene includes rendering a noisy image at a first resolution; obtaining one or more guide channels at the first resolution, and obtaining one or more corresponding guide channels at a second resolution. The second resolution may be the same resolution as, or a higher resolution than, the first resolution. For each of a plurality of local neighbourhoods, the method comprises: calculating the parameters of a model that approximates the noisy image as a function of the one or more guide channels (at the first resolution), and applying the calculated parameters to the one or more guide channels at the second resolution, to produce a denoised image at the second resolution.Type: ApplicationFiled: September 29, 2022Publication date: April 20, 2023Inventors: Szabolcs Csefalvay, James Imber, David Walton, Insu Yu
-
Publication number: 20230114852Abstract: A method of rendering an image of a 3-D scene includes rendering a noisy image and obtaining one or more guide channels. For each of a plurality of local neighborhoods, the method comprises: calculating the parameters of a model that approximates the noisy image as a function of the one or more guide channels, and applying the calculated parameters to produce a denoised image. At least one of (i) the noisy image, (ii) the one or more guide channels, and (iii) the denoised image, are stored in a quantized low-bitdepth format.Type: ApplicationFiled: September 30, 2022Publication date: April 13, 2023Inventors: Szabolcs Csefalvay, James Imber, David Walton, Insu Yu
-
Patent number: 11610127Abstract: Methods and systems for identifying quantisation parameters for a Deep Neural Network (DNN). The method includes determining an output of a model of the DNN in response to training data, the model of the DNN comprising one or more quantisation blocks configured to transform a set of values input to a layer of the DNN prior to processing the set of values in accordance with the layer, the transformation of the set of values simulating quantisation of the set of values to a fixed point number format defined by one or more quantisation parameters; determining a cost metric of the DNN based on the determined output and a size of the DNN based on the quantisation parameters; back-propagating a derivative of the cost metric to one or more of the quantisation parameters to generate a gradient of the cost metric for each of the one or more quantisation parameters; and adjusting one or more of the quantisation parameters based on the gradients.Type: GrantFiled: December 23, 2019Date of Patent: March 21, 2023Assignee: Imagination Technologies LimitedInventor: Szabolcs Csefalvay
-
Publication number: 20220392020Abstract: A pixel filter has a filter module that performs a first recursive filter operation in a first direction through a sequence of pixels to form a first filtered pixel value for each pixel, and performs a second recursive filter operation in a second direction through the sequence of pixels to form a second filtered pixel value for each pixel, the first and second recursive filter operations forming a respective filtered pixel value for a given pixel in dependence on the pixel value at that pixel and the filtered pixel value preceding that pixel in their respective direction of operation. The filtered pixel value of the preceding pixel is scaled by a measure of similarity between data associated with that pixel and its preceding pixel. Filter logic combines the first and second filtered pixel values formed by the first and second recursive filter operations to generate a filter output for the pixel, for each pixel of the sequence.Type: ApplicationFiled: June 9, 2022Publication date: December 8, 2022Inventor: Szabolcs Csefalvay
-
Patent number: 11386528Abstract: A pixel filter has a filter module that performs a first recursive filter operation in a first direction through a sequence of pixels to form a first filtered pixel value for each pixel, and performs a second recursive filter operation in a second direction through the sequence of pixels to form a second filtered pixel value for each pixel, the first and second recursive filter operations forming a respective filtered pixel value for a given pixel in dependence on the pixel value at that pixel and the filtered pixel value preceding that pixel in their respective direction of operation. The filtered pixel value of the preceding pixel is scaled by a measure of similarity between data associated with that pixel and its preceding pixel. Filter logic combines the first and second filtered pixel values formed by the first and second recursive filter operations to generate a filter output for the pixel, for each pixel of the sequence.Type: GrantFiled: September 16, 2020Date of Patent: July 12, 2022Assignee: Imagination Technologies LimitedInventor: Szabolcs Csefalvay
-
Publication number: 20210012466Abstract: A pixel filter has a filter module that performs a first recursive filter operation in a first direction through a sequence of pixels to form a first filtered pixel value for each pixel, and performs a second recursive filter operation in a second direction through the sequence of pixels to form a second filtered pixel value for each pixel, the first and second recursive filter operations forming a respective filtered pixel value for a given pixel in dependence on the pixel value at that pixel and the filtered pixel value preceding that pixel in their respective direction of operation. The filtered pixel value of the preceding pixel is scaled by a measure of similarity between data associated with that pixel and its preceding pixel. Filter logic combines the first and second filtered pixel values formed by the first and second recursive filter operations to generate a filter output for the pixel, for each pixel of the sequence.Type: ApplicationFiled: September 16, 2020Publication date: January 14, 2021Inventor: Szabolcs Csefalvay
-
Patent number: 10810708Abstract: A pixel filter has a filter module that performs a first recursive filter operation in a first direction through a sequence of pixels to form a first filtered pixel value for each pixel, and performs a second recursive filter operation in a second direction through the sequence of pixels to form a second filtered pixel value for each pixel, the first and second recursive filter operations forming a respective filtered pixel value for a given pixel in dependence on the pixel value at that pixel and the filtered pixel value preceding that pixel in their respective direction of operation. The filtered pixel value of the preceding pixel is scaled by a measure of similarity between data associated with that pixel and its preceding pixel. Filter logic combines the first and second filtered pixel values formed by the first and second recursive filter operations to generate a filter output for the pixel, for each pixel of the sequence.Type: GrantFiled: January 31, 2019Date of Patent: October 20, 2020Assignee: Imagination Technologies LimitedInventor: Szabolcs Csefalvay
-
Publication number: 20200202218Abstract: Methods and systems for identifying quantisation parameters for a Deep Neural Network (DNN). The method includes determining an output of a model of the DNN in response to training data, the model of the DNN comprising one or more quantisation blocks configured to transform a set of values input to a layer of the DNN prior to processing the set of values in accordance with the layer, the transformation of the set of values simulating quantisation of the set of values to a fixed point number format defined by one or more quantisation parameters; determining a cost metric of the DNN based on the determined output and a size of the DNN based on the quantisation parameters; back-propagating a derivative of the cost metric to one or more of the quantisation parameters to generate a gradient of the cost metric for each of the one or more quantisation parameters; and adjusting one or more of the quantisation parameters based on the gradients.Type: ApplicationFiled: December 23, 2019Publication date: June 25, 2020Inventor: Szabolcs Csefalvay
-
Publication number: 20200193208Abstract: A single-instruction, multiple data processor performs object detection in an image by testing for a plurality of object features in a plurality of image regions, the processor comprising: a set of computation units operable to execute a plurality of classifier sequences in parallel, each classifier sequence comprising a plurality of classifier routines, and each classifier routine comprising identical instructions to the other classifier routines in each of the plurality of classifier sequences; wherein each computation unit is configured to independently maintain data identifying an image region and a feature under test on that computation unit, and each classifier routine is arranged to access the data, test the identified feature against the identified image region and update the data such that the computation units are operable to concurrently test different features against different image regions.Type: ApplicationFiled: February 24, 2020Publication date: June 18, 2020Inventor: Szabolcs CSEFALVAY
-
Patent number: 10572756Abstract: A single-instruction, multiple data processor performs object detection in an image by testing for a plurality of object features in a plurality of image regions, the processor comprising: a set of computation units operable to execute a plurality of classifier sequences in parallel, each classifier sequence comprising a plurality of classifier routines, and each classifier routine comprising identical instructions to the other classifier routines in each of the plurality of classifier sequences; wherein each computation unit is configured to independently maintain data identifying an image region and a feature under test on that computation unit, and each classifier routine is arranged to access the data, test the identified feature against the identified image region and update the data such that the computation units are operable to concurrently test different features against different image regions.Type: GrantFiled: December 15, 2017Date of Patent: February 25, 2020Assignee: Imagination Technologies LimitedInventor: Szabolcs Csèfalvay
-
Publication number: 20190354844Abstract: Methods and systems for implementing a traditional computer vision algorithm as a neural network. The method includes: receiving a definition of the traditional computer vision algorithm that identifies a sequence of one or more traditional computer vision algorithm operations; mapping each of the one or more traditional computer vision algorithm operations to a set of one or more neural network primitives that is mathematically equivalent to that traditional computer vision algorithm operation; linking the one or more network primitives mapped to each traditional computer vision algorithm operation according to the sequence to form a neural network representing the traditional computer vision algorithm; and configuring hardware logic capable of implementing a neural network to implement the neural network that represents the traditional computer vision algorithm.Type: ApplicationFiled: May 21, 2019Publication date: November 21, 2019Inventors: Paul Brasnett, Daniel Valdez Balderas, Cagatay Dikici, Szabolcs Cséfalvay, David Hough, Timothy Smith, James Imber