Skip to main content

Minutes IETF97: netvc
minutes-97-netvc-00

Meeting Minutes Internet Video Codec (netvc) WG
Date and time 2016-11-15 06:50
Title Minutes IETF97: netvc
State Active
Other versions plain text
Last updated 2016-11-28

minutes-97-netvc-00
NETVC IETF 97
Date: November 15, 2016
Location: Seoul, Republic of Korea
Area Director: Alissa Cooper
Chairs: Adam Roach, Mo Zanaty
Minutes: Stephen Botzko

----
Requirements & Testing Docs

Is there value in publishing these?

Note IESG guidance not to publish them unless they are useful.

Consensus is that is value in publishing both, and that we will do so.

----
draft-ietf-netvc-requirements-03

Basic requirement - support independently decodable sub-frame units.

QP-points reduced from 12->10, with alignment only done on four points.

MS-SSIM now is luma only.

At least one more revision before publication, intention is to publish after
that version is reviewed.

----
draft-ietf-netvc-testing-03

Metric Updates
FastSSIM dropped in favor of MS-SSIM
VMAF RD uses values directly in RD curves

Range Updates
Indiviual tool testing - 4 quantizers
final testing - 10 qp points

240P and HDR clips need to be added; set not finalized in time for this draft
version. Rate Control testing methodology is also not included.

Likely 3 months away from finalizing this document.

----
draft-fuldseth-netvc-thor-03

- 10 and 12 bit video
- high internal precision regardless of imput
- various optimations

- 8, 10, 12 bit input, with 8, 10, 12 bit internal bit depth.

Luma gains with 8 bit input using 10 bit internal bit depth: ~3.8 db gain at
HD.  Goes up to 4.3 DB at 12-bit internal bit depth. Gains higher for chroma.

some discsussion on whether the gains were perceptual, some suggestion that
chroma gains should shifted to luma.


256-bit generic intrinsics
Not required for standardization, but useful to assess real-world complexity.
added so v128_add_8 could becomd v256_add_16, etc.

----
Daala Update

Fixed Point PVQ completed
Daala Entropy Coder now the default in AV1.

FASTSSIM overflow fixed, original results were somewhat too high (revised
results not quite as good as original)

Deringing filter changes (AV1)
- shorter 5-tap for chrome gave big gains on some sequences
- now works with 4:2:2 by disabling chroma deringing
- simpler threshold for second filter.
- significant speed improvement.

future Plans
- enable perceptual tools in PVQ
- continue port to AV1.


----
PVQ outside of Daala


Bit-rate close to AV1 w/o PVQ
encoding Speed remains a problem
More work needed.