Arm Limited

Anoto Ab

Tomas Fredrik Edsö

Erik Persson

Ola Hugosson

Andreas Björklund

Dominic Hugo Symes

Petter Ericson

Fredrik Peter Stolt

Torbjörn Gärdenfors

Derek David Whiteman

Patrik Andersson

A method for compressing activation data of a neural network to be written to a storage is provided. The activation data is formed into a plurality of groups and a state indicator indicates whether there are any data elements within each group that have a non-zero value. A second state indicator indicates, for groups having a non-zero value, whether sub-groups within the group contain a data element having a non-zero value. A sub-group state indicator indicates, for each sub-group having a non-zero value, which data elements within that sub-group have a non-zero value. Non-zero values of data elements in the activation data are encoded and a compressed data set is formed comprising the first state indicators, any second state indicators, any sub-group state indicators and the encoded non-zero values.

Compression and/or decompression of activation data

In a data processing system that comprises a memorycomprising N memory banks, a memory controller is configured to store one or more N data unit&#xd7;N data unit arrays of data in the memorysuch that each data unit in each row of each N&#xd7;N data unit array is stored in a different memory bank of the N memory banks, and such that each data unit in each column of each N&#xd7;N data unit array is stored in a different memory bank of the N memory banks

Storing arrays of data in data processing systems

A video processing system includes a video processing unit (VPU) and one or more display processing units, all having access to external memory. Video data representing frames to be displayed is generated. The VPU generates pixel data representing the frames and stores it in memory. The display processing units then read the pixel data to display the frames. The VPU is configured to generate and store in memory pixel data representing reference frames for the sequence of video frames at the full resolution of the reference frame and also at at least one lower resolution to the full resolution.

Video processing system

A video processing system comprises a video processor and an output buffer. When a new frame is to be written to the output buffer, the video processing system determines () for at least a portion of the new frame whether the portion of the new frame has a particular property. When it is determined that that the portion of the new frame has the particular property (), when a block of data representing a particular region of the portion of the new frame is to be written to the output buffer, it is compared to at least one block of data already stored in the output buffer, and a determination is made whether or not to write the block of data to the output buffer on the basis of the comparison. When it is determined that the portion of the new frame does not have the particular property (), the portion of the new frame is written to the output buffer.

Data processing systems

A data processing system comprises a video processor (). The data processing system is configured to, when a new frame () is to be encoded by the video processor (), determine for a sub-region of a set of plural sub-regions that the new frame () is divided into, whether the sub-region has changed from a previous frame (), and to control the encoding operation for the new frame () on the basis of the determination, e.g. to avoid performing motion estimation.

A data processing apparatus is provided which is configured to receive a down-sampled source block and a down-sampled reference frame portion. The data processing apparatus comprises interpolation circuitry configured to interpolate between pixels of the down-sampled reference frame portion to generate a set of interpolated down-sampled reference frame blocks. Cost function calculation circuitry calculates a cost function value indicative of a difference between the down-sampled source block and each interpolated down-sampled reference frame block. Minimisation circuitry identifies the lowest cost function value and estimation motion vector generation circuitry generates an estimate motion vector independence thereon.

Motion vector estimator

A video processing apparatus and method are provided, the video processing apparatus comprising first stage video processing circuitry and second stage video processing circuitry. The first stage video processing circuitry receives input video data and performs one or more processing operations on the input video data to generate an intermediate representation of the input video data. The intermediate representation comprises first and second separate data portions, with the first data portion containing transient data derived from the input video data and the second data portion containing long term data derived from the input video data. Transient data is only required for processing of a single video frame, while the long term data is required for processing of at least two video frames. The first stage video processing circuitry is arranged to output the first and second separate data portions for storing in a buffer, and the second stage video processing circuitry then retrieves the first and second data portions from the buffer and performs one or more further processing operations on those data portions in order to generate output video data. The transient data is compressed prior to being stored in the buffer, and then decompressed when retrieved from the buffer by the second stage video processing circuitry. Such an approach enables the operations of the second stage video processing circuitry to be decoupled from the operations of the first stage video processing circuitry, while reducing the storage capacity requirements of the buffer.

Video processing apparatus and a method of processing video data

A position code is recorded by capturing a sequence of images while writing on the base with a user unit. The position code may be decoded to a sequence of absolute positions reflecting the movement of the user unit over the base. If the position code in any image is not sufficiently recorded, e.g. because it is obscured by other graphics on the base, the spatial relationship of the image to another image in the sequence is determined. If the position code in this other image in the image sequence is decodable to an absolute position, the absolute position of the image in which the position code is not decodable could nevertheless be determined by using the spatial relationship. In this way the movement of the user unit can be recorded even though the position code is obscured, which could be the case when the same printing ink is used for both the position code and other graphics on the base.

Method, apparatus, computer program and storage medium for recording a movement of a user unit

A method, in an apparatus for position decoding, operates to extract a sequence of positions from a sequence of images of a position-coding pattern, as recorded by a sensor. The method, which may be embodied as a computer program on a computer-readable storage medium, involves the steps of retrieving at least one reference position (P, P); and extracting the sequence of positions solely by matching information obtained from each of the images with a corresponding pattern reference area (MA), which represents a known subset of the position-coding pattern with a given, direct or indirect, spatial relation to the reference position(s).

Methods, apparatus, computer program and storage medium for position decoding

A method aims to identify a virtual raster pattern in an image of a surface that is provided with a plurality of position-coding marks. Each mark is associated with a respective intersection of raster lines belonging to the raster pattern. By means of the method, the virtual raster pattern is identified via Fourier analysis of the image. A computer program product and a device for position determination are also described.

Growing community of inventors

Lund, Sweden

Tomas Fredrik Edsö