Machine-Learning-Based Early-Skip Strategies for AV1 Inter-Frame Prediction

• William Kolodziejski UFPel
• Marcelo Schiavon Porto UFPel
• Luciano Volcan Agostini UFPel

DOI: https://doi.org/10.5753/webmedia_estendido.2026.35012

Resumo

The AOMedia Video 1 (AV1) standard delivers superior compression efficiency for modern video streaming but introduces severe computational effort, particularly in its inter-frame prediction stage. This elevated computational effort restricts AV1’s practical deployment in energy-constrained or real-time scenarios. This extended summary of the author’s master’s dissertation presents a suite of hardware-friendly, machine-learning-based early-skip strategies designed to accelerate AV1 inter-frame prediction during the encoding process, aiming four computationally expensive tools: Interpolation Filter Search (IFS), Local and Global Warped Motion Compensation (LWMC/GWMC), and Overlapped Block Motion Compensation (OBMC). Experimental results implemented in the libaom reference software demonstrate significant execution time reductions, with minimal coding efficiency penalties.

Palavras-chave: Video Coding, Machine Learning, Early Skip, AV1, Inter-Frame Prediction

Referências

AOM. 2015. About AOM. [link] Acessado em: 14 de janeiro de 2026.

Ericsson. 2025. Innovative video formats on the horizon. [link] Acessado em: 14 de janeiro de 2026.

Jingning Han et al. 2021. A Technical Overview of AV1. Proc. IEEE 109, 9 (2021), 1435–1462. DOI: 10.1109/JPROC.2021.3058584

Intel. 2024. Media Capabilities Supported by Intel Hardware. [link] Acessado em: 14 de janeiro de 2026.

William Kolodziejski, Leonardo Braga, Marcelo Porto, and Luciano Agostini. 2026. Hardware-Friendly Machine-Learning-Based Fast AV1 Overlapped Block Motion Compensation. In 2026 IEEE International Symposium on Circuits and Systems (ISCAS). 3231–3235. DOI: 10.1109/ISCAS66217.2026.11562381

William Kolodziejski, Leonardo Braga, Marcelo Rezende, Marcelo Porto, and Luciano Agostini. 2025. Fast AV1 LocalWarped Motion Compensation Using Machine Learning. In Proceedings of the 31st Brazilian Symposium on Multimedia and the Web (Rio de Janeiro/RJ). SBC, Porto Alegre, RS, Brasil, 545–549. DOI: 10.5753/webmedia.2025.14161

William Kolodziejski, Leonardo Braga, Marcelo Rezende, Marcelo Porto, and Luciano Agostini. 2026. DM-FIFS: A Dual-Model Machine-Learning Method for Fast Interpolation Filter Search in AV1 Encoding. IEEE Transactions on Circuits and Systems I: Regular Papers 73, 3 (2026), 1593–1605. DOI: 10.1109/TCSI.2025.3624335

William Kolodziejski, Robson Domanski, and Luciano Agostini. 2025. FastGW: A Machine Learning-Based Early Skip for the AV1 Global Warped Motion Compensation. IEEE Transactions on Circuits and Systems I: Regular Papers 72, 3 (2025), 977–988. DOI: 10.1109/TCSI.2024.3486243