Rate-Distortion and Complexity Analysis of Fast Video Encoders

Hardware accelerators for video processing are always on demand due to the tight real-time restrictions of video applications, so it is important for designers to know which encoding tasks must be prioritized. In this paper, we compare three state-of-the-art fast video encoding implementations by evaluating encoding time, compression efficiency and the video quality of the end product. The fast implementations are called VVEnc, XEVE-EVC and SVT-AV1, and they were selected for their emerging adoption as fast video encoders by the community. Systematic experiments are performed to analyze these codecs in terms of quality, time, and compression efficiency. Our results show that VVEnc is the most efficient fast encoder in terms of compression, but it also takes more time to encode compared to the others. To simulate a realistic scenario, we evaluate how these implementations perform in a multi-core platform. Our experiments suggest that only SVT-AV1 was capable of achieving real-time encoding for HD (
$> 24$
frames per second), and higher frame rates can only be achieved with multiple threads, showing that dedicated hardware solutions are still required for real-time encoding of high-resolution media. We hope that with this work, future researchers will have an easier time identifying the more daunting parts of the video encoding process.