Multi-resolution encoding and optimization for next generation video compression

TL;DR

This paper proposes a multi-resolution, multi-bitrate encoding scheme that leverages encoder analysis correlations to significantly speed up video encoding without losing efficiency, validated on HEVC and next-gen codecs.

Contribution

It introduces a novel multi-resolution tier multi-bitrate encoding method that improves computational efficiency by reusing analysis decisions across resolutions and bitrates.

Findings

Achieves up to 2.5x speed-up in encoding times.

Maintains coding efficiency comparable to conventional methods.

Demonstrates hardware-aware optimizations impact on encoding speed.

Abstract

Multi-encoding implies encoding the same content in multiple spatial resolutions and multiple bitrates. This work evaluates the encoder analysis correlations across 2160p, 1080p, and 540p encodings of the same video for conventional ABR bitrates. A multi-resolution tier multi-ABR encoding scheme is modeled and evaluated, which significantly improves the computational efficiency of conventional ABR encoding. Video content is first encoded at the lower resolution with the median bitrate. Encoder analysis decisions, such as motion vectors and CU block structure, are then used in the other encodes in the same resolution tier. The analysis is then extrapolated and refined to be used in higher-resolution encodes. The scheme is validated using x265 HEVC video encoder. The proposed multi-resolution tier multi-bitrate encoding scheme achieves overall speed-ups of up to 2.5x, compared to the conventional single-instance encoding approach. Furthermore, this speed-up is achieved without substantial losses in coding efficiency. SIMD Vector units in CPUs have become the de-facto standard for accelerating media and other kernels that exhibit parallelism. This work also demonstrates the impact of hardware-aware optimizations on the encoding speeds of the next-generation video codecs. The work is evaluated using the Arowana XVC encoder.