LCEVC Best Practices

This page gives advice on how to get the very best results when applying LCEVC-enhancement to your encoding workflows.

Introduction

LCEVC is a video enhancement layer which when combined with a base video codec, produces an enhanced video stream. LCEVC data is made up of details which are missing from the base codec video, identified by comparing the base to the source video. The enhancement layer allows a significant reduction in video bitrate to achieve the same video quality or a significant increase in video quality for a similar bitrate or a combination of the two.

It is important to change the way ABR ladders are configured to ensure LCEVC is utilised to its full potential. Ideal ABR ladders depend somewhat on the business priorities of the provider, which may also be informed by end user data such as the average bandwidth available to users. The reduction in bandwidth across the ABR ladder allows more users to consume renditions with higher video quality.

LCEVC also offers video providers the opportunity for savings on their encoding, storage and often most significantly CDN delivery costs. Depending on your objectives, bitrates can either be reduced whilst delivering equivalent quality, the resolution and quality can be increased at similar bitrates or any combination of the two in between can be employed.

How to select optimal bitrates and resolutions for your content

There are a few things to consider when planning optimal ABR ladders and encoding settings for LCEVC.

Which base codec is being used?

Modern base codecs, such as HEVC or VP9, offer more efficient video encoding than h.264, allowing video bitrates to be reduced for similar VQ at a computer processing cost. The addition of LCEVC allows bitrates to be reduced even further whilst also reducing the compute processing cost.

How much detail is in the source?

LCEVC data is the difference in sharp details between the source video and the output base video codec, if the source video is lacking in sharp details, there will be less LCEVC enhancement data.

Any other source configurations?

The amount of motion in the source can also determine encoding settings (more on that later?)

How do you determine the best resolution for LCEVC at any given operating point?

A Convex Hull approach with VMAF can be helpful to determine high-level guidelines for different content or base codecs (always combined with visual inspection, as discussed below), however the chart below provides a guideline, summarising V-Nova's findings across a broad number of contents.

h.264 and LCEVC [h.264] recommended resolutions for different bitrates

Here are HEVC examples up to 4k UHD:

h.265 and LCEVC [h.265] recommended resolutions for different bitrates

Recommended ABR ladders

The table below provides an example. It is taken from a report written by industry analyst Jan Ozer. The h.264 rungs, encoded using x264 (medium preset), are in line with the ladder recommended by Apple in the HLS Authoring Specification. The LCEVC [x264] profiles have been optimised to achieve both a quality uplift and a reduction in the average streamed bitrate.

Example of h.264 ABR ladder and corresponding LCEVC [h.264] ladder

As is often the case, the number of profiles has been reduced from 7 to 6, reducing the total bitrate in the ladder by 27% (whilst achieving a better quality). 1080p resolution has been used for LCEVC down to 2.7 Mbps, and 720p at 1.5 Mbps and even 800 kbps. In the last case 576p would have also been a viable option, prioritizing robustness over sharpness. At 400 kbps 480p was selected, but 432p would have been a reasonable alternative.

Full details and results from the above ladder experiment are available here <ADD LINK WHEN AVAILABLE>. The report demonstrates that the LCEVC ladder saves bandwidth - roughly 19-22% reduction of the average streaming bitrate, assuming market data on the average playback frequency of profiles - and simultaneously improves the overall quality of experience, measures both subjectively via formal Mean Opinion Score (MOS) and with VMAF.