Imagination Technologies Limited

Imagination Technologies Limted

Paul Brasnett

Jonathan Diggins

Marc Vivet

Stephen Morphet

Szabolcs Cséfalvay

Cagatay Dikici

Steven John Fishwick

James Imber

David Hough

Timothy Smith

Clifford Gibson

Chris Martin

Muhammad Asad

Daniel Valdez Balderas

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.

Upscaling lower resolution image data for processing

A reduced noise image can be formed from a set of images. One of the images of the set can be selected to be a reference image and other images of the set are transformed such that they are better aligned with the reference image. A measure of the alignment of each image with the reference image is determined. At least some of the transformed images can then be combined using weights which depend on the alignment of the transformed image with the reference image to thereby form the reduced noise image. By weighting the images according to their alignment with the reference image the effects of misalignment between the images in the combined image are reduced. Furthermore, motion correction may be applied to the reduced noise image.

Image noise reduction

Hardware implementations of DNNs and related methods with a variable output data format. Specifically, in the hardware implementations and methods described herein the hardware implementation is configured to perform one or more hardware passes to implement a DNN wherein during each hardware pass the hardware implementation receives input data for a particular layer, processes that input data in accordance with the particular layer (and optionally one or more subsequent layers), and outputs the processed data in a desired format based on the layer, or layers, that are processed in the particular hardware pass. In particular, when a hardware implementation receives input data to be processed, the hardware implementation also receives information indicating the desired format for the output data of the hardware pass and the hardware implementation is configured to, prior to outputting the processed data convert the output data to the desired format.

Hardware implementation of a deep neural network with variable output data format

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.

Implementing traditional computer vision algorithms as neural networks

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.

Methods and systems for converting weights of a deep neural network from a first number format to a second number format

An interlaced video signal can include content of different types, such as interlaced content and progressive content. The progressive content may have different cadences according to the ratio between the frame rate of the progressive content and the field rate of the interlaced video signal. Cadence analysis is performed to identify the cadence of the video signal and/or to determine field pairings when progressive content is included. As described herein, motion information (e.g. motion vectors) for blocks of fields of a video signal can be used for the cadence analysis. The use of motion information provides a robust method of performing cadence analysis.

Cadence analysis for a video signal having an interlaced format

Pixel buffering

A data processing system for performing face detection on a stream of frames of image data, the data processing system comprising: a skin patch identifier configured to identify one or more patches of skin color in a first frame and characterize each patch in the first frame using a respective patch construct of a predefined shape; a first search tile generator configured to generate one or more first search tiles from the one or more patch constructs; and a face detector configured to detect faces in the stream by performing face detection in one or more frames of the stream within the first search tiles.

Growing community of inventors

West Molesey, United Kingdom

Paul Brasnett