Method and apparatus with neural network performing deconvolution
AssigneeSamsung Electronics Co., Ltd.
InventorsJoonho Song, Sehwan Lee, Junwoo Jang
Life of this patent
Estimating the term…
Abstract
A neural network apparatus configured to perform a deconvolution operation includes a memory configured to store a first kernel; and a processor configured to: obtain, from the memory, the first kernel; calculate a second kernel by adjusting an arrangement of matrix elements comprised in the first kernel; generate sub-kernels by dividing the second kernel; perform a convolution operation between an input feature map and the sub-kernels using a convolution operator; and generate an output feature map, as a deconvolution of the input feature map, by merging results of the convolution operation.
Claims13
An apparatus for processing a neural network, the apparatus comprising:
- a processor configured to perform a convolution operation for a convolutional network using a convolution operator and a deconvolution operation for a deconvolutional network using the convolution operator;
- wherein the processor is configured to perform the deconvolution operation by:
- obtaining an input feature map and a first kernel of the deconvolutional network;
- generating a transpose kernel by performing a transpose operation on the first kernel and generating sub-kernels by performing a split operation on the transpose kernel;
- and
- generating, by using the convolution operator, an output feature map of the deconvolutional network by merging results of a convolution operation between the input feature map and the sub-kernels.
A processor-implemented method for neural network, the method comprising:
- performing a convolution operation for a convolutional network using a convolution operator;
- and
- performing a deconvolution operation for a deconvolutional network using the convolution operator,
- wherein the performing of the deconvolution operation comprises:
- obtaining an input feature map and a first kernel of the deconvolutional network;
- generating a transpose kernel by performing a transpose operation on the first kernel and generating sub-kernels by performing a split operation on the transpose kernel;
- and
- generating, by using the convolution operator, an output feature map of the deconvolutional network by merging results of a convolution operation between the input feature map and the sub-kernels.
Dependencies are inferred from the claim text.