On Intra Video Coding and In-loop Filtering for Neural Object Detection Networks

TL;DR

This paper investigates how traditional video coding optimizations for human viewers translate to neural object detection, revealing that current codecs and in-loop filters impact detection performance and bitrate efficiency.

Contribution

It provides a comparative analysis of HEVC and VVC codecs and the effect of VVC in-loop filters on neural network-based object detection accuracy and bitrate.

Findings

VVC's bitrate savings for detection are lower than for PSNR.

Disabling VVC in-loop filters reduces bitrate while maintaining detection accuracy.

VVC with in-loop filters offers modest bitrate improvements for neural detection.

Abstract

Classical video coding for satisfying humans as the final user is a widely investigated field of studies for visual content, and common video codecs are all optimized for the human visual system (HVS). But are the assumptions and optimizations also valid when the compressed video stream is analyzed by a machine? To answer this question, we compared the performance of two state-of-the-art neural detection networks when being fed with deteriorated input images coded with HEVC and VVC in an autonomous driving scenario using intra coding. Additionally, the impact of the three VVC in-loop filters when coding images for a neural network is examined. The results are compared using the mean average precision metric to evaluate the object detection performance for the compressed inputs. Throughout these tests, we found that the Bj{\o}ntegaard Delta Rate savings with respect to PSNR of 22.2 % using VVC instead of HEVC cannot be reached when coding for object detection networks with only 13.6% in the best case. Besides, it is shown that disabling the VVC in-loop filters SAO and ALF results in bitrate savings of 6.4 % compared to the standard VTM at the same mean average precision.