SYSTEMS, METHODS AND BITSTREAMS FOR RECONSTRUCTION REFINEMENT IN FEATURE CODING OF VIDEO FOR MACHINES

Abstract:

Systems and methods are provided to improve encoding and decoding of video in a system for feature coding for machines (FCM). At the encoder, a calculate distribution properties process is applied to the feature maps output by a split NN front end to determine at least one probability distribution property of tensors of the feature maps. This process may operate only on detected regions of interest in the input images. The distribution properties may include an RMS determined for the tensors. At a decoder, a reconstruction refinement process uses the probability distribution property encoded in a bitstream to align the probability distribution properties of decoded and restored feature maps with the original feature maps and/or within regions of interest determined by the encoder and signaled in the bitstream. The refined and restored feature maps are applied to the split neural network back end for inferencing.