Scribd
Upload
93 views14 pages

Enter AV1: Alliance For Open Media Codec

The document summarizes test results comparing the new AV1 video codec to H.265 and VP9. Key points: - AV1 is in early stages with limited coding tools enabled in most implementations - Tests by Bitmovin, an AV1-enabled cloud vendor, show AV1 meeting quality targets but with long encode times compared to H.265 and VP9 - Other studies show AV1 can be 20-35% more efficient than VP9 once the codec is further developed and optimized

Uploaded by

Vikram Bhaskaran
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
93 views14 pages

Enter AV1: Alliance For Open Media Codec

The document summarizes test results comparing the new AV1 video codec to H.265 and VP9. Key points: - AV1 is in early stages with limited coding tools enabled in most implementations - Tests by Bitmovin, an AV1-enabled cloud vendor, show AV1 meeting quality targets but with long encode times compared to H.265 and VP9 - Other studies show AV1 can be 20-35% more efficient than VP9 once the codec is further developed and optimized

Uploaded by

Vikram Bhaskaran
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 14

Enter AV1

• Alliance for Open Media codec


• Alliance formed by Google, Cisco, Mozilla, Amazon, Netflix,
Microsoft and Intel
• Goal was open source, royalty free codec
• First introduced in a live setting by cloud
encoding/player/analytics vendor www.Bitmovin.com at
NAB
• Also won best of NAB award
• Very early days of codec
• Unfair fight timewise – very long encode cycles
• Did meet data rate requirements
• Other basic parameters like GOP size
• More technology demonstration than competition
Workflow
• Bitmovin encoded • Parameters used
• Sent WebM/DASH formatted • quality = good
• cpu-used = 0
files to me
• threads = 0
• I decoded into Y4M • lag-in-frames =
• Measured PSNR quality with 25
Moscow University VQMT • kf-max-dist = 60
Tool • kf-min-dist = 60
• undershoot-pct =
• Why no VMAF? 5
• Hybrik didn’t have decoder • overshoot-pct = 5
• Y4M files too big to send • bitrate = 500
Meridian 4K
Sintel 4K
TOSN – 4K
4K Overall
1080p Overall
720p Overall
AV1 vs. x265 – Max Quality
• Re-encode x265 using Slow preset
• Saw before, this was max quality

• Substitute New test clip for TOSN


• All tests at 1080p
• Measure with PSNR and MS SSIM
PSNR Overall
MS SSIM Overall
What’s that Say About AV1?
• Bitstream scheduled to be frozen by end of 2017
• Other results
• Google – AV1 is 35% more efficient than VP9
• Netflix – AV1 is 20% more effective than VP9
• We’re seeing less. Why?
Bitmovin Blog Post (bit.ly/AV1_BM)
• AV1 codebase is VP9/10
• There are 77 additional experimental coding tools
that are under consideration.
• 8 are enabled by default (and by Bitmovin)
• Bitmovin is only AV1 system in production
• More conservative than other experimental systems
• Each user compiles their own version with selected
experiments enabled
• Don’t know how many coding tools included by Netflix
and Google
• Bottom line: results not necessarily inconsistent and will
almost certainly improve at a rapid rate over time
Why So Different from Other Evals?
• Another report concluded “the H.265/MPEG-HEVC
reference software implementation provides significant
average bit-rate savings of 38.4% and 32.8% compared
to AOM/AV1 and H.264/MPEG-AVC, respectively”
• Our results
• Used commercially available HEVC encoder (not reference
encoder)
• Used a much more recent version of AV1 with more experiments
compiled into the code
• Used a version compiled by the company using it (Bitmovin), not
the researcher
• Used encoding parameters for VP9/AV1 (as well as x265/MC) that
were approved by the vendors (not those compiled by the
researcher)

You might also like