Opened 8 years ago

Closed 7 years ago

#1243 closed enhancement (fixed)

preestChromaPredMode is never called

Reported by: kolya Owned by:
Priority: minor Milestone: HM-16.3
Component: HM Version: HM-13.0
Keywords: Cc: fbossen, ksuehring, davidf, jct-vc@…

Description

code from line 1355 to 1358 of TEncCu.cpp never works

Change History (6)

comment:1 Changed 8 years ago by DefaultCC Plugin

  • Cc fbossen ksuehring davidf jct-vc@… added

comment:2 Changed 8 years ago by fbossen

  • Type changed from defect to enhancement

The exact code is:

  Bool bSeparateLumaChroma = true; // choose estimation mode
  UInt uiPreCalcDistC      = 0;
  if( !bSeparateLumaChroma )
  {
    m_pcPredSearch->preestChromaPredMode( rpcTempCU, m_ppcOrigYuv[uiDepth], m_ppcPredYuvTemp[uiDepth] );
  }

The question is whether there is any benefit in setting bSeparateLumaChroma to false. If there is it may be worthwhile making it a configurable option.

comment:3 Changed 8 years ago by ksuehring

  • Milestone HM-14.0 deleted

comment:4 Changed 7 years ago by karlsharman

This setting was explored in the early days of RExt, having spent a little
while making it work correctly (it had not been maintained/tested in the
main HM branch for a long time)

  • see JCTVC-K0181,

file "ResultsFromConfigurationChangesTo422and444_r4.xlsx",
experiment "03a_encSearchYc"

The BD-rate performance then was:

Format Y U V Enc. Time
4:2:2 0.0%-0.2%-0.3% 141%
4:4:4 +0.2%-0.5%-0.5% 166%

(there is no change in decoding time)

The option seemed to just move the operating point, improving chroma at
the expense of luma, and there was a large increase in the encoding time.

For that reason, the option was not recommended for further RExt
development.

Since then, the option has remained, but the changing of 4:2:2 to
use square blocks rather than rectangular has unfortunately broken
the feature again.

The results using the latest RExt RGB & 4:4:4 test conditions are:

Tier Format Y U V Enc. Time
Main RGB +0.8% 0.0% 0.0% 170%
High RGB +0.5%-0.2%-0.1% 170%
Super-high RGB +0.4%-0.3%-0.2% 170%
Main 4:4:4 +0.2%-0.5%-0.6% 162%
High 4:4:4 +0.3%-0.4%-0.4% 162%
Super-high 4:4:4 +0.3%-0.4%-0.4% 162%

(there is no change in decoding time)

The main 4:4:4 (above) results are similar to the results seen
back in K0181 (top).

Since:

1/ this feature still offers no significant BD-rate improvement
2/ this feature costs a lot in terms of encoding time
3/ it adds complexity to the software
4/ it was not used in the CTC of HM, and it was also broken in HM
5/ the committee decided that RExt didn't need this in the CTC
6/ and it is currently broken for 4:2:2,

... I recommend removing this feature from the code.

comment:5 Changed 7 years ago by ksuehring

Given the performance data, I agree that the code can be removed.

In an inactive state, it is always hard to maintain and easily forgotten with changes in related code.

comment:6 Changed 7 years ago by karlsharman

  • Milestone set to HM-16.3
  • Resolution set to fixed
  • Status changed from new to closed

Removed in r4235.

Note: See TracTickets for help on using tickets.

This list contains all users that will be notified about changes made to this ticket.

These roles will be notified: Reporter, Owner, Subscriber, Participant

  • David Flynn(Subscriber)
  • Frank Bossen(Subscriber, Participant)
  • jct-vc@…(Subscriber)
  • Karl Sharman(Participant)
  • karl.sharman@…(Always)
  • Karsten Suehring(Subscriber, Participant, Always)
  • Nikolay Shlyakhov(Reporter)