Feature Compression for Machines with Range-Based Channel Truncation and Frame Packing

This paper proposes a method that enhances the compression performance of the current model under development for the upcoming MPEG standard on Feature Compression for Machines (FCM).
This standard aims at providing inter-operable compressed bitstreams of features in the context of split computing, i.e., when the inference of a large computer vision Neural-Network-based model is split between two devices.
Intermediate features correspond to tensors that can be reduced and entropy coded to limit the required bandwidth of such transmission.
In the envisioned design for the MPEG-FCM standard, feature tensors may be reduced using NN layers before being converted into 2D video frame to be compressed using existing video compression standards.
This paper introduces an additional channel truncation and packing method which enables the system to preserve only the relevant channels for a given content, while preserving the computer vision task performance at the receiver.
Implemented within the MPEG-FCM test model, the proposed method yields compression gains of 10.59\% on average in terms of bitrate vs. accuracy on multiple computer vision tasks and datasets.