Implementing Traditional Computer Vision Algorithms as Neural Networks

Patent metadata

Field	Value
Publication number	US-2019354844-A1
Application number	US-201916418322-A
Country	US
Kind code	A1
Filing date	May 21, 2019
Priority date	May 21, 2018
Publication date	Nov 21, 2019
Grant date	—

Abstract

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.

First claim

What is claimed is:

1 . A method of implementing a traditional computer vision algorithm as a neural network, the method comprising:
receiving a definition of the traditional computer vision algorithm that identifies a sequence of one or more traditional computer vision algorithm operations which form the traditional computer vision algorithm; 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.

2 . The method of claim 1 , wherein at least one of the traditional computer vision algorithm operations is a histogram operation and the histogram operation is mapped to a convolution primitive, an activation primitive and a pooling primitive.

3 . The method of claim 2 , wherein the convolution primitive is configured to convolve an input to the histogram operation with h 1×1×1 filters wherein h is a number of bins in the histogram.

4 . The method of claim 1 , wherein at least one of the traditional computer vision algorithm operations is a dilation operation and the dilation operation is mapped to a convolution primitive and an activation primitive.

5 . The method of claim 1 , wherein at least one of the traditional computer vision algorithm operations is a dilation operation with a square structuring element and the dilation operation with a square structuring element is mapped to a pooling primitive.

6 . The method of claim 1 , wherein at least one of the traditional computer vision algorithm operations is an erosion operation and the erosion operation is mapped to a convolution primitive and an activation primitive.

7 . The method of claim 1 , further comprising training, using one or more neural network training techniques, the neural network representing the traditional computer vision algorithm prior to configuring the hardware logic to implement the neural network.

8 . The method of claim 1 , wherein the mapping is automatically performed based on a library that comprises a mapping of traditional computer vision algorithm operations to mathematically equivalent sets of one or more neural network primitives.

9 . The method of claim 1 , wherein the traditional computer vision algorithm is a BRISK descriptor algorithm and the neural network comprises a single fully connected primitive.

10 . The method of claim 9 , wherein the fully connected primitive is configured to perform a matrix-vector multiplication between a matrix of weights and a vector of intensity values.

11 . The method of claim 10 , further comprising determining the weights of the matrix using one or more neural network training techniques.

12 . The method of claim 1 , wherein the hardware logic capable of implementing a neural network comprises a neural network accelerator.

13 . The method of claim 12 , wherein the neural network accelerator is embodied in hardware on an integrated circuit.

14 . A system for implementing a traditional computer vision algorithm as a neural network, the system comprising:
hardware logic capable of implementing a neural network; and a converter configured to:
receive a definition of the traditional computer vision algorithm that identifies a sequence of one or more traditional computer vision algorithm operations which form the traditional computer vision algorithm;
map 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;
link 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
configure the hardware logic capable of implementing a neural network to implement the neural network that represents the traditional computer vision algorithm.

15 . A neural network accelerator configured to implement a neural network that represents a traditional computer vision algorithm, the neural network having been generated by mapping each traditional computer vision algorithm operation forming the traditional computer vision algorithm to a mathematically equivalent sequence of one or more neural network primitives.

16 . A computer-implemented automated tool for forming a neural network, the automated tool having access to a library of mappings from traditional computer vision algorithm operations to mathematically equivalent sets of one or more neural network primitives, wherein the automated tool is configured to:
receive a definition of a traditional computer vision algorithm that identifies a sequence of one or more traditional computer vision algorithm operations which form the traditional computer vision algorithm; use the library to map 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; link the one or more network primitives mapped to each computer vision algorithm operation according to the sequence to form a neural network representing the computer vision algorithm; and output a definition of the neural network for use in configuring hardware logic to implement the neural network.

17 . A non-transitory computer readable storage medium having stored thereon computer readable instructions that, when executed at a computer system, cause the computer system to perform the method at set forth in claim 1 .

Assignees

• Imagination Tech Ltd

Inventors

• Brasnett Paul
• Balderas Daniel Valdez
• Dikici Cagatay
• Cséfalvay Szabolcs
• Hough David
• Smith Timothy
• Imber James

Classifications

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title

• G06N3/02 Primary

  Neural networks · CPC title

• G06F8/40

  Transformation of program code · CPC title

• G06F17/16

  Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)} · CPC title

• G06F17/15

  Correlation function computation {including computation of convolution operations (arithmetic circuits for sum of products per se, e.g. multiply-accumulators G06F7/5443; digital filters, e.g. FIR, IIR, adaptive filters H03H17/00)} · CPC title