Complexity and compression efficiency analysis of libaom AV1 video codec

Area of Science:

• Video compression technology
• Digital media encoding
• Computational complexity analysis

Background:

• The Alliance for Open Media (AOMedia) developed the royalty-free AV1 video format in 2018.
• AV1 incorporates advanced coding tools and partitioning structures, increasing complexity compared to predecessors.
• Understanding AV1's computational demands is crucial for developing efficient codecs.

Purpose of the Study:

• To profile the computational effort of individual AV1 coding steps.
• To analyze the trade-offs between computational cost and coding efficiency for AV1's superblock partitioning.
• To provide recommendations for optimizing AV1 codec development.

Main Methods:

• Profiling the libaom reference software to measure the time spent on each AV1 coding step.
• Conducting experiments to evaluate the impact of disabling various superblock partition types on encoding time and bitrate.
• Analyzing the relationship between computational cost and coding efficiency for different partitioning strategies.

Main Results:

• Inter-frame prediction (76.98%) and transform (20.57%) are the most computationally intensive steps in the libaom encoder.
• Disabling ternary and asymmetric quaternary partitions minimally impacts bitrate (0.25% and 0.22% increase, respectively).
• Disabling all rectangular partitions reduces encoding time by approximately 35%.

Conclusions:

• The inter-frame prediction and transform steps are key targets for optimization in AV1 codecs.
• Strategic disabling of complex partitions, like ternary and asymmetric quaternary, offers significant speedups with negligible quality loss.
• The findings provide a methodological basis for creating faster and more efficient AV1-compatible video codecs.