Method and device for end-to-end feature compression in multidimensional data encoding

Abstract

To provide a method and a device for compressing feature data in an end-to-end network with respect to encoding of multi-dimensional data.SOLUTION: A method uses residual coding to compress feature data by enhancing it by removing redundancy. Enhanced feature data is spatially down-sampled and the number of channels of the enhanced feature data is reduced by applying a 2D convolution operation. The method further generates a heat map based on the reduced enhanced feature data. The reduced enhancement feature data may be scaled using the generated heat map, and the scaled reduced enhancement feature data may be entropy encoded to generate a bitstream.SELECTED DRAWING: Figure 7A

Images