| 摘要 |
A method including receiving a set of input data in a first matrix format. The method further includes compressing the set of input data to obtain a first set of compressed data in a second matrix format, where compressing the set of input data includes using a quantization equation, the quantization equation including Yq(i,j)=[(Y(i,j)+offset)<<n]/qs, where Yq(i,j) represents a coefficient in a matrix of the first set of compressed data having a coordinate (i,j), Y(i,j) represents a coefficient in a matrix of the set of input data having the coordinate (i,j), offset is an integer, << is a first bit-wise shift operator, n is an integer, qs is a real number. The method also includes sending the first set of compressed data to a first destination.
|
