Indexing elements in a source array

Indexing data elements in a source array (112) comprises generating shifted copy arrays (114) based on the source array (e.g. convolving the source array with a set of filter kernels, e.g. sparse filter kernels), each shifted array comprising the source array data elements at a respective shifted po...

Full description

Saved in:
Bibliographic Details
Main Authors Aria Ahmadi, Cagatay Dikici
Format Patent
LanguageEnglish
Published 19.05.2021
Subjects
CALCULATING
COMPUTING
COUNTING
ELECTRIC DIGITAL DATA PROCESSING
IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
PHYSICS
Online AccessGet full text

Cover

More Information
Summary:Indexing data elements in a source array (112) comprises generating shifted copy arrays (114) based on the source array (e.g. convolving the source array with a set of filter kernels, e.g. sparse filter kernels), each shifted array comprising the source array data elements at a respective shifted position. Indices for indexing (130) the source array are received, each index indicating a target location in the source array. For each index, data element(s) are retrieved (e.g. by gating (132) elements based on the index to either select an element or interpolate a data element) from respective shifted copy array(s), wherein the shifted copy array(s) is/are selected according to the index. Therefore, rather than retrieving elements from different locations in a single (source) array (which requires the expensive calculation of a number of memory addresses), multiple shifted arrays are generated, and the elements at the locations in those shifted arrays are retrieved. Expensive memory address calculations are thus avoided. The method may be used in image or feature map warping based on a motion vector field.
Bibliography:Application Number: GB20200007115