The Effects of Video Compression on Simulated Time Events

Event-based cameras provide many advantages over conventional frame-based cameras, both in terms of time resolution and motion blur, making them increasingly popular in research and industrial applications, particularly in automotive sensing. However, despite the recent uptake, event-based datasets are still rare. This lack of availability of time-event datasets has led to the development of tools such as v2e, which can simulate event streams from conventional video. Although these tools are incredibly useful for early stage development and testing, little attention has been paid to how the quality of the source video affects these simulated events. This paper presents the first comprehensive quantitative study of the impact of video compression on simulated event streams. We evaluate how lossy compression algorithms affect the accuracy and noise levels of simulated events, revealing that compression artifacts, particularly in darker regions of videos, can produce significant noise in these event streams. We introduce a novel event stream comparison method to determine the fidelity of events simulated from varying video codecs and parameters. Our findings show that compression can severely degrade simulated event quality, and that codec choice and parameters can influence event fidelity significantly. We further benchmark a range of corrective methods, including conventional filters and machine learning-based preprocessing, to mitigate the adverse effects of compression. Although some approaches show some improvements, none fully restore the fidelity of uncompressed video sources. We conclude with practical recommendations for dataset publishers and researchers using event simulation tools, emphasizing the importance of compression-aware practices in dataset preparation and usage.