-
公开(公告)号：US20210073614A1
公开(公告)日：2021-03-11
申请号：US17000468
申请日：2020-08-24
Applicant: Imagination Technologies Limited
Inventor： Cagatay Dikici , Paul Brasnett , Muhammad Asad , Stephen Morphet
Abstract: Methods and system for converting a plurality of weights of a filter of a Deep Neural Network (DNN) in a first number format to a second number format, the second number format having less precision than the first number format, to enable the DNN to be implemented in hardware logic. The method comprising: determining, for each of the plurality of weights, a quantisation error associated with quantising that weight to the second number format in accordance with a first quantisation method; determining a total quantisation error for the plurality of weights based on the quantisation errors for the plurality of weights; identifying a subset of the plurality of weights to be quantised to the second number format in accordance with a second quantisation method based on the total quantisation error for the plurality of weights; and generating a set of quantised weights representing the plurality of weights in the second number format, the quantised weight for each weight in the subset of the plurality of weights based on quantising that weight to the second number format in accordance with the second quantisation method and the quantised weight for each of the remaining weights of the plurality of weights based on quantising that weight to the second number format in accordance with the first quantisation method.
-
公开(公告)号：US20190026857A1
公开(公告)日：2019-01-24
申请号：US16138302
申请日：2018-09-21
Applicant: Imagination Technologies Limited
Inventor： Paul Brasnett , Jonathan Diggins , Steven Fishwick , Stephen Morphet
IPC: G06T1/60 , H04N19/426 , G06T3/00 , H04N19/33
Abstract: In an example method and system, image data to an image processing module. Image data is read from memory into a down-scaler, which down-scales the image data to a first resolution, which is stored in a first buffer. A region of image data which the image processing module will request is predicted, and image data corresponding to at least part of the predicted region of image data is stored in a first buffer, in a second resolution, higher than the first. When a request for image data is received, it is then determined whether image data corresponding to the requested image data is in the second buffer, and if so, then image data is provided to the image processing module from the second buffer. If not, then image data from the first buffer is up-scaled, and the up-scaled image data is provided to the image processing module.
-
公开(公告)号：US09922450B2
公开(公告)日：2018-03-20
申请号：US15376323
申请日：2016-12-12
Applicant: Imagination Technologies Limited
Inventor： Stephen Morphet
CPC classification number: G06T15/405 , G06T15/005 , G06T15/40 , G06T2200/04 , G06T2200/28
Abstract: An apparatus and a method for generating 3-dimensional computer graphic images. The image is first sub-divided into a plurality of rectangular areas. A display list memory is loaded with object data for each rectangular area. The image and shading data for each picture element of each rectangular area are derived from the object data in the image synthesis processor and a texturizing and shading processor. A depth range generator derives a depth range for each rectangular area from the object data as the imaging and shading data is derived. This is compared with the depth of each new object to be provided to the image synthesis processor and the object may be prevented from being provided to the image synthesis processor independence on the result of the comparison.
-
公开(公告)号：US20170193631A1
公开(公告)日：2017-07-06
申请号：US15463348
申请日：2017-03-20
Applicant: Imagination Technologies Limited
Inventor： Stephen Morphet
CPC classification number: G06T1/60 , G06F12/02 , G06T11/40 , G06T15/005 , G06T15/80 , G06T17/20 , G06T2200/04 , G09G3/003 , G09G5/363 , G09G5/39 , G09G2310/04 , G09G2360/122 , H04N19/42
Abstract: A memory management system for generating 3-dimensional computer images is provided. The memory management system includes a device for subdividing an image into a plurality of rectangular areas, a memory for storing object data pertaining to objects in the image which fall in each rectangular area, a device for storing the object data in the memory, a device for deriving image data and shading data for each rectangular area from the object data, a device for supplying object data for each rectangular area from the respective portion of the memory and, if the rectangular area contains objects also falling in at least one other rectangular area, also from the global list, to the deriving device, and a device for storing the image data and shading data derived by the deriving device for display. The memory includes at least one portion allocated to each rectangular area and at least one portion allocated as a global list.
-
公开(公告)号：US11587199B2
公开(公告)日：2023-02-21
申请号：US16138302
申请日：2018-09-21
Applicant: Imagination Technologies Limited
Inventor： Paul Brasnett , Jonathan Diggins , Steven Fishwick , Stephen Morphet
IPC: G06T1/60 , G06T3/00 , H04N19/33 , H04N19/426
Abstract: In an example method and system, image data to an image processing module. Image data is read from memory into a down-scaler, which down-scales the image data to a first resolution, which is stored in a first buffer. A region of image data which the image processing module will request is predicted, and image data corresponding to at least part of the predicted region of image data is stored in a first buffer, in a second resolution, higher than the first. When a request for image data is received, it is then determined whether image data corresponding to the requested image data is in the second buffer, and if so, then image data is provided to the image processing module from the second buffer. If not, then image data from the first buffer is up-scaled, and the up-scaled image data is provided to the image processing module.
-
Patent Agency Ranking
公开(公告)号：US20220067497A1
公开(公告)日：2022-03-03
申请号：US17524606
申请日：2021-11-11
Applicant: Imagination Technologies Limited
Inventor： Cagatay Dikici , Paul Brasnett , Muhammad Asad , Stephen Morphet
Abstract: Methods and system for converting a plurality of weights of a filter of a Deep Neural Network (DNN) in a first number format to a second number format, the second number format having less precision than the first number format, to enable the DNN to be implemented in hardware logic. The method comprising: determining, for each of the plurality of weights, a quantisation error associated with quantising that weight to the second number format in accordance with a first quantisation method; determining a total quantisation error for the plurality of weights based on the quantisation errors for the plurality of weights; identifying a subset of the plurality of weights to be quantised to the second number format in accordance with a second quantisation method based on the total quantisation error for the plurality of weights; and generating a set of quantised weights representing the plurality of weights in the second number format, the quantised weight for each weight in the subset of the plurality of weights based on quantising that weight to the second number format in accordance with the second quantisation method and the quantised weight for each of the remaining weights of the plurality of weights based on quantising that weight to the second number format in accordance with the first quantisation method.
-
公开(公告)号：US11217008B2
公开(公告)日：2022-01-04
申请号：US15891555
申请日：2018-02-08
Applicant: Imagination Technologies Limited
Inventor： Stephen Morphet
Abstract: An apparatus and a method for generating 3-dimensional computer graphic images. The image is first sub-divided into a plurality of rectangular areas. A display list memory is loaded with object data for each rectangular area. The image and shading data for each picture element of each rectangular area are derived from the object data in the image synthesis processor and a texturizing and shading processor. A depth range generator derives a depth range for each rectangular area from the object data as the imaging and shading data is derived. This is compared with the depth of each new object to be provided to the image synthesis processor and the object may be prevented from being provided to the image synthesis processor independence on the result of the comparison.
-
公开(公告)号：US11188817B2
公开(公告)日：2021-11-30
申请号：US17000468
申请日：2020-08-24
Applicant: Imagination Technologies Limited
Inventor： Cagatay Dikici , Paul Brasnett , Muhammad Asad , Stephen Morphet
Abstract: Methods and system for converting a plurality of weights of a filter of a Deep Neural Network (DNN) in a first number format to a second number format, the second number format having less precision than the first number format, to enable the DNN to be implemented in hardware logic. The method comprising: determining, for each of the plurality of weights, a quantisation error associated with quantising that weight to the second number format in accordance with a first quantisation method; determining a total quantisation error for the plurality of weights based on the quantisation errors for the plurality of weights; identifying a subset of the plurality of weights to be quantised to the second number format in accordance with a second quantisation method based on the total quantisation error for the plurality of weights; and generating a set of quantised weights representing the plurality of weights in the second number format, the quantised weight for each weight in the subset of the plurality of weights based on quantising that weight to the second number format in accordance with the second quantisation method and the quantised weight for each of the remaining weights of the plurality of weights based on quantising that weight to the second number format in accordance with the first quantisation method.
-
公开(公告)号：US10109032B2
公开(公告)日：2018-10-23
申请号：US13798934
申请日：2013-03-13
Applicant: Imagination Technologies Limited
Inventor： Paul Brasnett , Jonathan Diggins , Steven Fishwick , Stephen Morphet
IPC: H04N19/33 , G06T1/60 , G06T3/00 , H04N19/426
Abstract: In an example method and system, image data to an image processing module. Image data is read from memory into a down-scaler, which down-scales the image data to a first resolution, which is stored in a first buffer. A region of image data which the image processing module will request is predicted, and image data corresponding to at least part of the predicted region of image data is stored in a first buffer, in a second resolution, higher than the first. When a request for image data is received, it is then determined whether image data corresponding to the requested image data is in the second buffer, and if so, then image data is provided to the image processing module from the second buffer. If not, then image data from the first buffer is up-scaled, and the up-scaled image data is provided to the image processing module.
-
公开(公告)号：US11240406B2
公开(公告)日：2022-02-01
申请号：US14823629
申请日：2015-08-11
Applicant: Imagination Technologies Limited
Inventor： Steven John Fishwick , Stephen Morphet
Abstract: There is provided a method and apparatus for motion estimation in a sequence of video images. The method comprises a) subdividing each field or frame of a sequence of video images into a plurality of blocks, b) assigning to each block in each video field or frame a respective set of candidate motion vectors, c) determining for each block in a current video field or frame, which of its respective candidate motion vectors produces a best match to a block in a previous video field or frame, d) forming a motion vector field for the current video field or frame using the thus determined best match vectors for each block, and e) forming a further motion vector field by storing a candidate motion vector derived from the best match vector at a block location offset by a distance derived from the candidate motion vector. Finally, steps a) to e) are repeated for a video field or frame following the current video field or frame. The set of candidate motion vectors assigned at step b) to a block in the following video field or frame includes the candidates stored at that block location at step e) during the current video field or frame The method enables a block or tile based motion estimator to improve its accuracy by introducing true motion vector candidates derived from the physical behaviour of real world objects.
-
-
-
-
-
-
-
-
-
Search Results
25
records
(took 0.017 seconds)
