公开(公告)号：US20220012574A1

公开(公告)日：2022-01-13

申请号：US17327154

申请日：2021-05-21

Applicant: Imagination Technologies Limited

Inventor： James Imber

IPC: G06N3/063

Abstract: A method of determining a number format for representing a set of two or more network parameters of a Deep Neural Network “DNN” for use in configuring hardware logic to implement the DNN. The method includes: determining a sensitivity of the DNN with respect to each network parameter in the set of network parameters; for each candidate number format of a plurality of candidate number formats: determining a quantisation error associated with quantising each network parameter in the set of network parameters in accordance with the candidate number format; generating an estimate of an error in an output of the DNN caused by quantisation of the set of network parameters based on the sensitivities and the quantisation errors; generating a local error based on the estimated error; and selecting the candidate number format of the plurality of candidate number formats with the minimum local error as the number format for the set of network parameters.

公开(公告)号：US10885427B2

公开(公告)日：2021-01-05

申请号：US16181104

申请日：2018-11-05

Applicant: Imagination Technologies Limited

Inventor： James Imber

IPC: G06F11/00 ,  G06N3/063 ,  G06N3/08 ,  G06F7/483 ,  G06F17/11 ,  G06N3/04

Abstract: Methods for determining a fixed point format for one or more layers of a DNN based on the portion of the output error of the DNN attributed to the fixed point formats of the different layers. Specifically, in the methods described herein the output error of a DNN attributable to the quantisation of the weights or input data values of each layer is determined using a Taylor approximation and the fixed point number format of one or more layers is adjusted based on the attribution. For example, where the fixed point number formats used by a DNN comprises an exponent and a mantissa bit length, the mantissa bit length of the layer allocated the lowest portion of the output error may be reduced, or the mantissa bit length of the layer allocated the highest portion of the output error may be increased. Such a method may be iteratively repeated to determine an optimum set of fixed point number formats for the layers of a DNN.

公开(公告)号：US20190236449A1

公开(公告)日：2019-08-01

申请号：US16181147

申请日：2018-11-05

Applicant: Imagination Technologies Limited

Inventor： James Imber

IPC: G06N3/08

CPC classification number: G06N3/08 ,  G06N3/0454 ,  G06N3/063

Abstract: Methods for selecting fixed point number formats for representing values input to and/or output from layers of a DNN which take into account the impact of the fixed point number formats for a particular layer in the context of the DNN. The methods comprise selecting the fixed point number format(s) used to represent sets of values input to and/or output from a layer one layer at a time in a predetermined sequence wherein any layer is preceded in the sequence by the layer(s) from which it depends. The fixed point number format(s) for each layer is/are selected based on the error in the output of the DNN associated with the fixed point number formats. Once the fixed point number format(s) for a layer has/have been selected any calculation of the error in the output of the DNN for a subsequent layer in the sequence is based on that layer being configured to use the selected fixed point number formats.

公开(公告)号：US20170116755A1

公开(公告)日：2017-04-27

申请号：US15333475

申请日：2016-10-25

Applicant: Imagination Technologies Limited

Inventor： James Imber ,  Adrian Hilton ,  Jean-Yves Guillemaut

IPC: G06T7/40 ,  G06T7/00

CPC classification number: G06T7/90 ,  G06T15/506 ,  G06T2207/10024 ,  G06T2207/10028 ,  G06T2207/20028

Abstract: An image processing system and method for determining an intrinsic colour component of one or more objects for use in rendering the object(s) is described. One or more input images are received, each representing a view of the object(s), wherein values of the input image(s) are separable into intrinsic colour estimates and corresponding shading estimates. A set of surface normals for the object(s) of the input image(s) is determined. In accordance with the values of the input image(s) and the determined set of surface normals, a global lighting estimate is determined which provides consistent corresponding intrinsic colour estimates for a plurality of regions of the object(s) from the input image(s). The intrinsic colour component is determined in accordance with the values of the input image(s) and the determined global lighting estimate. The determined intrinsic colour component of the object(s) and the determined set of surface normals for the object(s) are stored for use in rendering the object(s).

公开(公告)号：US20160042556A1

公开(公告)日：2016-02-11

申请号：US14544122

申请日：2014-11-26

Applicant: Imagination Technologies Limited

Inventor： James Imber ,  Adrian Hilton ,  Jean-Yves Guillemaut

IPC: G06T15/50 ,  G06T11/60 ,  G06T7/00 ,  G06T15/10 ,  G06T11/00

CPC classification number: G06T7/90 ,  G06T7/507 ,  G06T11/001 ,  G06T11/60 ,  G06T15/04 ,  G06T15/06 ,  G06T15/10 ,  G06T15/50 ,  G06T15/80 ,  G06T17/00 ,  G06T2207/20224 ,  G06T2215/16

Abstract: Relightable free-viewpoint rendering allows a novel view of a scene to be rendered and relit based on multiple views of the scene from multiple camera viewpoints. Image values from the multiple camera viewpoints can be separated into diffuse image components and specular image components, such that an intrinsic colour component of a relightable texture can be determined for a specular scene, by using the separated diffuse image components. Furthermore, surface normals of geometry in the scene can be refined by constructing a height map based on a conservative component of an initial surface normal field and then determining the refined surface normals based on the constructed height map.

Abstract translation: 可靠的自由观点渲染允许基于多个摄像机视点的场景的多个视图来渲染并再现场景的新颖视图。 来自多个摄像机视点的图像值可以被分离成漫射图像分量和镜面图像分量，使得可以通过使用分离的漫射图像分量来确定镜面场景的可重构纹理的固有颜色分量。 此外，场景中几何的表面法线可以通过基于初始表面法线的保守分量构建高度图，然后基于构造的高度图确定精确的表面法线来进行细化。

公开(公告)号：US20140354645A1

公开(公告)日：2014-12-04

申请号：US14016622

申请日：2013-09-03

Applicant: Imagination Technologies Limited

Inventor： James Imber ,  Adrian Hilton

IPC: G06T15/60

CPC classification number: G06T15/60 ,  G06T15/04

Abstract: A model of a scene of an image (e.g. a frame of a video sequence) is generated from one or more views of the scene captured from one or more different camera viewpoints. An initial texture for applying to the model is derived from the one or more views of the scene. The initial texture is separated into a lighting estimate and a colour estimate, which may be orthogonal and which may be processed independently. The lighting estimate is filtered with a high-pass filter to thereby determine shadow regions of the scene which are regions of detailed shadow which are likely to be caused by ambient occlusion in the scene and which are therefore retained when the texture is relit for rendering the image. A shadow-detail estimate (or “dark map”) is provided which indicates one or more shadow regions of the texture which are to remain in shadow when the image is rendered.

Abstract translation: 从一个或多个不同的摄像机视点捕获的场景的一个或多个视图生成图像场景的模型（例如视频序列的帧）。 用于应用于模型的初始纹理从场景的一个或多个视图导出。 初始纹理被分为照明估计和颜色估计，其可以是正交的并且可以被独立地处理。 用高通滤波器对照明估计进行滤波，从而确定场景的阴影区域，该阴影区域是可能由场景中的环境遮挡引起的详细阴影区域，并且当纹理依赖于渲染 图片。 提供阴影细节估计（或“暗图”），其指示当呈现图像时将保持在阴影中的纹理的一个或多个阴影区域。