2019 MSU Study Names Aurora1 Top AV1 Encoder Implementation

Aurora1 proved to be the best choice for the critical evaluation standard of encoding speed compared with efficiency and quality.
  • The Visionular Aurora1 AV1 encoder proved much more efficient than x264, requiring nearly half the encoding bitrate to reach the same YUV-SSIM metric. This was achieved with an additional compute time of just 35%. 
  • With the Aurora1 AV1 encoder, services can be confident to move to the advanced AV1 codec standard without incurring overly burdensome operational costs.
  • Aurora1 enables services to save 50% on their delivery bitrate while maintaining the same visual quality. 

challenge.

The annual MSU codec study has become an important competitive marker for video codec and encoding technology developers. Though most all commercial video service companies and operators have internal teams that conduct regular evaluations and comparisons, the role of an independent third party in benchmarking video encoders is critical.

In this MSU report, objective assessment methods were used to compare the encoding quality of recent HEVC and AV1 encoders. To evaluate codec performance, this effort employed 11 video sequences at 4K resolution.  

the comparison consisted of two use cases. For each use case codec developers were able to provide encoding parameters that had to satisfy a minimum speed requirement for the respective use case:

  • Fast—20fps
  • Universal—1fps

All comparisons were run with a computer configured using an Intel Core i7-8700K (Coffee Lake) processor operating @ 3.7GHz with 32 GB of RAM running Windows 10.

For objective quality measurements the YUV-SSIM metric was chosen.

solutions.

Video encoders tested:

  • Visionular Aurora1 AV1 encoder
  • SVT-AV1, SVT-HEVC, SVT-VP9 Intel Open Visual Cloud encoders
  • MulticoreWare x265 HEVC encoder
  • Huawei Technologies HW265  HEVC encoder
  • Bytedance V265 HEVC encoder
  • MainConcept HEVC encoder
  • Nanjing Yunyan sz265 HEVC encoder
  • Google WebM VP9 encoder
  • SIF encoder

Baseline comparison was H.264 using x264 Developer Team core: 157 r2969 d4099dd

results.

Judging from the mean quality scores, MSU noted that the Visionular Aurora1 achieved first place in the quality competition followed by HW265, MainConcept HEVC, Bytedance V265 Encoder, and finally the Intel Open Visual Cloud SVT-AV1.

To understand how the results were tabulated, following is an explanation. View a free abridged version of the report by clicking here.

RD Curves

The main charts in this comparison are classic RD curves (quality/bitrate graphs) and relative-bitrate/relative- time charts. Additionally, MSU used bitrate-handling charts (the ratio of real to target bitrates) and per-frame quality charts.

The RD charts show variation in codec quality by bitrate or file size. For this metric, a higher value presumably indicates better quality.

Relative-Bitrate/Relative-Time Charts

Relative-bitrate/relative-time charts show the average bitrate’s dependence on relative encoding time for a fixed quality output. The y-axis shows the ratio of a codec’s bitrate under test to the reference codec’s bitrate for a fixed quality.

A lower value indicates a better-performing codec. For example, a value of 0.7 means the codec can encode the sequence using 30% fewer bits than the reference codec (x264) produced.

The x-axis shows the relative encoding time. Larger values indicate a slower codec. A value of 2.5 means the codec works 2.5 times slower, on average, than the reference codec (x264).

By the numbers, the Visionular Aurora1 AV1 encoder achieved:

  • 58% reduced bitrate at the same quality over x264 using subjective evaluation
  • 44% reduced bitrate at the same quality over x264 using SSIM
  • Best performing codec for quality/speed (high quality use case)

Graphs

According to subjective quality scores, the best encoder for ripping (high quality), universal (balance of efficiency and quality), and fast (low-latency) use cases is the Aurora1.

The following graph compares encoders based on their ability to reach the same quality rate (BSQ) as measured subjectively.

The following graph compares encoders based on their ability to reach the same quality rate using the YUV-SSIM metric for the 4K universal use case.

This chart shows Aurora1 clearly outperforming all the other codecs based on bitrate/quality using subjective comparison.

The following chart demonstrates the ability for Aurora1 to trade-off speed and quality in a superior manner by achieving the highest efficiency of all the codecs evaluated at a speed faster than x264.

presentation. Aurora1 + ML for VOD & RTC applications.

Zoe Liu Visionular Aurora1 AV1 ML presentation at BAV
how can we help you?

Click the button below so that we can help you make the best codec decision for your application.

At Tsingyuan Ventures, we believe in startup companies that ride the transformative power of major technology shifts such as deep learning in computer vision. Visionular’s founders are world-class technologists in their field of video codec and AI-driven optimization. We feel privileged to support their adventure with our resources and experience.

Dr. Xuhui Shao
Managing Partner, Tsingyuan Ventures

We'd love to tell you more about our AV1 encoder. Hit the button so we can chat.