PIC PIC PIC
Joint Collaborative Team on Video Coding (JCT-VC)
of ITU-T SG16 WP3 and ISO/IEC JTC1/SC29/WG11 Document: JCTVC-J0581
10th Meeting: Stockholm, SE, 11 – 20 July 2012 Title:

BoG report: Extended chroma formats


Status: Input Document to JCT-VC
Purpose: BoG report
Author(s):
David Flynn
davidf@rd.bbc.co.uk

Source: BoG

______________________________

Abstract

A break out group on extended chroma formats was held on Tuesday 17th and Wednesday 18th July, covering topics on test material selection, test conditions and a more in-depth presentation of proposals JCTVC-J0191 [1] and JCTVC-J0357 [2] that provide software models for 4:2:2 and 4:4:4 chroma formats.

Test material selection for 4:2:2 and 4:4:4

The BoG reviewed the currently available test material with a view to select material for testing during the next meeting cycle for use by a possible AHG. A viewing session was arranged to examine the JCTVC-0197 [3] sequences.

Each sequence should be 10s (with the exception of Traffic).
4:4:4 shall be in R′G′B′ and Y′CbCr? (or Y′CoCg?) (or constant-luminance Y′CbCr)
4:2:2 shall be in Y′CbCr.

Suggestion: pick material from different productions, rather than different material from within a single production to reduce systematic biases.

Suggestion: concern about lack of diversity if converting from 4:4:4 to 4:2:2.

Suggestion: take two 4:4:4 sequences and reuse as 4:2:2

Sequences should be provided on uni-hannover ftp site in the appropriate format.

Comment: All processing and conversions must be documented.

Comment: Traffic gives 4k content

Suggestion: For any 4k sequence, crop to 2560×1600 as per current Class-A.

How many sequences? Suggestion: 5 – 8

Should the test material exercise something?

Review of new material currently available (4:4:4) for JCT-VC use

Recommendation: Use the following sequences (Table 2): kimono, parkscene, birds-in-cage, ducks-and-legs, traffic, crowd-run, old-town, seeking.

Review of new material currently available (4:2:2) for JCT-VC use

Recommendation: Use the following sequences (Table 2): Kids Soccer, graphics, horse, water rocks close, birds-in-cage, ducks-and-legs, kimono, traffic.

Table 1: Recommended 4:4:4 sequences (before any required conversions)
Sequence
Duration
Resolution
Frame rate
Color space
Bit-depth
Acquisition
Owner
Availability
Contribution
Kimono 10s 1920x1080 24 R′G′B′ 10 Sony HDC-F950 CoSME TBD J0152
Parkscene 10s 1920x1080 24 R′G′B′ 10 Sony HDC-F950 CoSME TBD J0152
Birds-in-cage 10s 1920x1080 60 R′G′B′ 10 HDCAM-SR Plannet inc 1 week J0197
Ducks-and-legs 10s 1920x1080 60 R′G′B′ 10 HDCAM-SR Plannet inc 1 week J0197
Traffic 5s 2560x1600 30 R′G′B′ 10 ? ? 1 week J0197
Crowd-run 10s 1920x1080 50 RGB 16 SVT/EBU 1 week I0108
Old-town 10s 1920x1080 50 RGB 16 SVT/EBU 1 week I0108
Seeking 10s 1920x1080 50 RGB 16 SVT/EBU 1 week I0108

Table 2: Recommended 4:2:2 sequences (before any required conversions)
Sequence
Duration
Resolution
Frame rate
Color space
Bit-depth
Acquisition
Owner
Availability
Contribution
Kids soccer 10s 1920x1080 50 Y′CbCr 10 ArriFlex EBU 1 J0152
Graphics 10s 1920x1080 50 Y′CbCr 10 Sony F35 EBU 1 J0152
Horse 10s 1920x1080 50 Y′CbCr 10 Sony F35 EBU 1 J0152
Water rocks close 10s 1920x1080 50 Y′CbCr 10 Sony F35 EBU 1 J0152
Birds-in-cage 10s 1920x1080 60 Y′CbCr 10 HDCAM-SR Plannet inc 1 week J0197
Ducks-and-legs 10s 1920x1080 60 Y′CbCr 10 HDCAM-SR Plannet inc 1 week J0197
Kimono 10s 1920x1080 24 Y′CbCr 10 Sony HDC-F950 CoSME TBD J0152
Traffic 5s 2560x1600 30 Y′CbCr 10 ? ? 2weeks J0197

Test conditions for a possible AhG

4:4:4

Suggestion: UHDTV defines constant luminance, we should have some constant-luminance sequences with the new chromaticity coordinates.

Suggestion: we should look at 10 bit parameters only for now.

Suggestion: can also try 8 bit (for comparison with main/he10) Recommend (initially) to keep current configurations (modified only for chroma format)

QPs: comment: introduce extra QPs

Suggestion: examine bitrate/quality of 10-bit standard QPs with a view to reviewing the test conditions.

Comment: different QPs for 8 and 10bit?

Comment: keeping some overlap between 8 and 10bit will allow for some comparison.

Suggestion: keep 8-bit same, examine 10bit.

Format considerations

R′G′B′, Y′CbCr, Y′CoCg, Y

C’Cb_CCr_C

Suggestion: work on 4:4:4 at greater than 8-bit would be appropriate for still image compression, for lossless compression of still pictures R′G′B′ and Y′CoCg might be used.

Suggestion: Take Y′CbCr as a starting point, make R′G′B′/RGB available for people to investigate and recommend.

Question: if alternative formats are used, will sequences be made available?

Suggestion: Yes, or (ideally, and) provide the software. MD5 sums shall also be provided.

Comment: it would be useful to compare the evaluation method used for 4:4:4, see JVT document.

Common conditions

All intra

Comment about bitrates: Beware when lowering QP, that bitrate doesn’t become insane.

Comment: application areas operate at different bitrates: Very high (low compression ratio) (currently for AVC up to 500Mbps–1Gbps 1080p30 4:4:4), High (currently for AVC  100Mbps)

Comment: benefit of HEVC might be to enable currently “difficult” applications eg 1080p60/120 to use current bitrates.

Comment: be aware that there have been issues with low QPs in the HM

Something else: random access? low delay (B)?

Suggestion: pick only one.

Suggestion: pick low-delay, it allows us to see the performance of inter.

Recommendation: pick low-delay (B)

Comment: Examining re-ordered sequence is not discouraged.

Anchors

Suggestion: test against JM 18.3, as that is the current state of the art

Comment: JM configuration files are required

Comment: there are some problems with JM 18.3 configuration. (with low delay, at 14bit)

Comment: some other proposals may have JM configurations we can reuse

Suggestion for the next meeting

Suggestion: review (and show) all test material that has been used at the next meeting

Problem: critical viewing environment.

Problem: viewing equipment. Consult with Vittorio on facilities.

Comment: reviewing material can be done in 8-bit, which isn’t so critical.

Future test material

Suggestion: Some form of screen content / animated content

Suggestion: sharper material / More vivid (ie, greater chroma excursion)

Suggestion: greater than 10-bit.

Suggestion: linear would be lovely, allows us to do what we want.

Comment: EBU 4k sequences might be available for IBC (September)

Suggestion: Review common test conditions.

Review of JCTVC-J0191 and JCTVC-J0357

Note: A subsequent revision (v3) of JCTVC-J0191 provides software using HM-7.1 as a base. This revision also integrates a number of features in JCTVC-J0357.

Table 3: Comparison of JCTVC-J0191 and JCTVC-J0357

Tool / Feature

J0357

J0191

Text supplied

No

No

Compatibility with HM-7.0

Untested

Standard and non-standard configurations tested (7.0)

Standard configurations tested (7.1)

Chroma formats supported

4:2:2, 4:4:4

4:0:0, 4:2:2, 4:4:4

Input and output file formats

Yes

Yes

plus sample repeat/drop

PU

4:4:4, Introduces NxN split for co-located chroma PU.

4:4:4, Introduces NxN split for co-located chroma PU.

Also allowed in 4:2:2, subject to minimum size limits

Slices/CU

Slices, CU structures are chroma-format independent.

Slices, CU structures are chroma-format independent.

TU

TU may be 4x4 for chroma 4:4:4

TU may be 4x4 for chroma 4:4:4.

Option to permit 4:2:0 style rules (eg force min chroma size as 8x8)

TU structures

Non-square TU (4:2:2)

Non-square TU (4:2:2)

Quantizer (4:2:2)

introduces scaling factor to compensate for non-square transform

Configurable options:

  • QP+3
  • QP-3
  • QStep Table mod to QP±3

NSQT

Extended to 4:2:2 case

(might be same as 191)

Configurable options:

  • 4:2:2 maintains 4:2:0 splitting rules (16×4 luma ⇒ 4×8 chroma).
  • 4:2:2 allow 8×4 chroma when 16×4 luma.

Extended to 4:4:4

Transform skip

TBC

4:4:4 Option for J357 method (single transform skip for all channels)

Intra-prediction angular prediction

Uses same angle, but pads reference samples

4:4:4 = 2 modes (DM, LM)

The 4:2:2 algorithm has been modified to account for the change in sample aspect ratio for the chroma channels.

  • Option to force DM chroma only
  • Option to import J357 method (v3)

Intra reference sample filtering

No change

4:4:4 = on/off (chroma)

4:2:2 = off/on(v)/on(h,v)

TU edge filtering for H, V and DC modes

No change

4:4:4 = on/off (chroma)

4:2:2 = off/on(v)/on(h,v)

LM_Chroma

4:4:4 uses adjacent col

Options:

  • 4:4:4 uses adjacent col
  • 4:4:4 uses non-adjacent col (as per 4:2:0 method)

MDDT

No change

Option for MDDT to be applied to chroma 4x4 TUs for 4:4:4.

Motion interpolation filters

4:2:2 use luma (v)

4:4:4 use luma (h+v)

4:2:2 Options:

  • use luma (v)
  • no change

4:4:4 Options:

  • 4:4:4 use luma (h+v)
  • 4:4:4 no change

Scaling lists

TBC

For 4:2:2, a larger scaling list is used and sub-sampled.

Options:

  • Use 32x32 chroma scaling list (imported J357 method in v3)
  • Use 32x32 luma scaling list for chroma as luma.

Chroma QP clipping table

No change

Unchanged from HM7.0, but there is the option to use a different clipping table for 4:2:2 and 4:4:4.

Chroma CBF Context Variable (CV) derivation

No change

Unchanged from HM7.0, but the CBF derivation for chroma may configured to use a non-depth based selection criteria.

Chroma CV sets and rules

No change

Option:

  • Rules for chroma CV derivation may be forced to be same as luma
  • The CVs for Cb and Cr may be separated
  • A single set of CVs for Y, Cb and Cr

MDCS

No change

MDCS is only used for luma in HM7.0, but it may be optionally applied to chroma. Plus flexibility on direction mapping

Significance map coding

Doubled height of map (replicating rows)

Doubled group size

Options:

  • The significance map given in JCTVC-I0521 may be used, instead of the default neighbour-based algorithm, for the 4x8, 8x4, 8x16 and 16x8 4:2:2 chroma TUs.
  • Maintain group size 16, using 2x8 coefficient groups for vertical scan in 4:2:2
  • J357 method (in v3)

Deblocking filter

No change

No change

ALF

No change

No change

SAO

No change

No change

RDO mods

Modified RDO to handle RGB. (applies to all 4:4:4)

Option for J357 method (v3)

References

[1]   P. Silcock, K. Sharman, N. Saunders, and J. Gamei, “AHG12: Extension of HM7 to Support Additional Chroma Formats.” document JCTVC-J0191, July 2012.

[2]   K. Kawamura, T. Yoshino, and S. Naito, “AHG12: 4:2:2/4:4:4 chroma format extension for HEVC Version 2.” document JCTVC-J0357, July 2012.

[3]   K. Sugimoto and S. Sekiguchi, “AHG12: 4:4:4 test sequences for professional extension development.” document JCTVC-J0197, July 2012.

[4]   T. Suzuki, P. Silcock, K. Sharman, N. Saunders, J. Gamei, and A. Tabatabai, “Proposals on HEVC professional extensions.” document JCTVC-J0152, July 2012.