compression ratio
|
Bit-rate saving
|
|
Reference
|
Tested
|
Class F
|
4.6
|
4.7
|
−1.4%
|
Class B
|
2.1
|
2.1
|
0.0%
|
SC(GBR)
|
7.5
|
12.0
|
−25.6%
|
RangeExt
|
2.2
|
2.2
|
0.0%
|
Overall (w/o SC)
|
3.2
|
3.2
|
−0.6%
|
Overall (w/ SC)
|
5.3
|
7.6
|
−13.1%
|
Enc Time[%]
|
119%
|
Dec Time[%]
|
87%
|
Closer examination of the results show that the reported gain is predominately derived from the SCC RGB 4:4:4 sequences and that the tool has no effect on natural content sequences (Class B and RangeExt) and very poor performance on Class F sequences. It has previously been noted that the Class F sequences were captured using analogue methods that introduce noise and filtering effects.
The contribution shows that there are additional coding gains when used in conjunction with the SAPHV tool of JCTVC-M0056.
When trans-quant bypass is enabled, each CU has the option of using this mode, an independent palette for each component is transmitted that contains four values. The block is then coded without any prediction or transform and transmits the residual by indexing the palette using run-length coding. For cases where there are insufficient palette entries to code the block, pels may be coded using an escape mode.
Though the initial results are for lossless coding, it is asserted that this tool can be implemented in non-lossless coding situations by using soft matching or palette quantisation. When using the range extensions (non-lossless) common conditions, such an implementation was able to achieve a 20% BD-rate gain in all-intra super-high-tier when using SCC sequences.
Revisit: pPossible CE for further investigation (there is another palette method being presented).
Dostları ilə paylaş: | 
