All Intra HE
|
All Intra LC
|
|
compression ratio
|
bit-rate saving
|
compression ratio
|
bit-rate saving
|
|
anchor
|
proposed
|
anchor
|
proposed
|
Class A
|
2,21
|
2,46
|
-10,26%
|
2,20
|
2,46
|
-10,48%
|
Class B
|
2,09
|
2,18
|
-4,52%
|
2,08
|
2,18
|
-4,69%
|
Class C
|
1,97
|
2,11
|
-6,49%
|
1,97
|
2,11
|
-6,76%
|
Class D
|
1,83
|
2,01
|
-8,41%
|
1,82
|
2,01
|
-8,67%
|
Class E
|
2,90
|
3,23
|
-10,30%
|
2,89
|
3,23
|
-10,47%
|
Class F
|
4,51
|
5,21
|
-12,23%
|
4,48
|
5,20
|
-12,55%
|
Overall
|
2,58
|
2,86
|
-8,30%
|
2,57
|
2,85
|
-8,53%
|
|
|
|
Enc Time[%]
|
105%
|
105%
|
Dec Time[%]
|
93%
|
93%
|
|
|
Random Access HE
|
Random Access LC
|
|
compression ratio
|
bit-rate saving
|
compression ratio
|
bit-rate saving
|
|
anchor
|
proposed
|
anchor
|
proposed
|
Class A
|
2,72
|
2,80
|
-2,99%
|
2,72
|
2,80
|
-3,07%
|
Class B
|
2,38
|
2,41
|
-0,94%
|
2,38
|
2,41
|
-0,99%
|
Class C
|
2,55
|
2,59
|
-1,69%
|
2,55
|
2,59
|
-1,76%
|
Class D
|
2,67
|
2,73
|
-2,07%
|
2,66
|
2,72
|
-2,13%
|
Class E
|
|
|
|
|
|
|
Class F
|
28,38
|
30,90
|
-7,11%
|
28,24
|
30,82
|
-7,28%
|
Overall
|
7,99
|
8,55
|
-2,85%
|
7,95
|
8,54
|
-2,93%
|
|
|
|
Enc Time[%]
|
102%
|
101%
|
Dec Time[%]
|
100%
|
99%
|
|
|
Low delay B HE
|
Low delay B LC
|
|
compression ratio
|
bit-rate saving
|
compression ratio
|
bit-rate saving
|
|
anchor
|
proposed
|
anchor
|
proposed
|
Class A
|
|
|
Class B
|
2,38
|
2,40
|
-0,77%
|
2,38
|
2,40
|
-0,80%
|
Class C
|
2,55
|
2,58
|
-1,45%
|
2,54
|
2,58
|
-1,51%
|
Class D
|
2,66
|
2,71
|
-1,66%
|
2,66
|
2,71
|
-1,72%
|
Class E
|
4,15
|
4,16
|
-0,28%
|
4,15
|
4,16
|
-0,30%
|
Class F
|
47,02
|
49,50
|
-5,72%
|
46,82
|
49,36
|
-5,84%
|
Overall
|
11,66
|
12,18
|
-2,00%
|
11,62
|
12,15
|
-2,06%
|
|
|
|
Enc Time[%]
|
103%
|
102%
|
Dec Time[%]
|
100%
|
100%
|
|
|
|
|
|
|
|
|
Low delay B HE
|
Low delay B LC
|
|
compression ratio
|
bit-rate saving
|
compression ratio
|
bit-rate saving
|
|
anchor
|
proposed
|
anchor
|
proposed
|
Class A
|
|
|
Class B
|
2,33
|
2,35
|
-1,03%
|
2,33
|
2,35
|
-1,07%
|
Class C
|
2,47
|
2,51
|
-1,50%
|
2,47
|
2,51
|
-1,54%
|
Class D
|
2,59
|
2,64
|
-1,63%
|
2,59
|
2,63
|
-1,68%
|
Class E
|
4,12
|
4,14
|
-0,36%
|
4,12
|
4,13
|
-0,38%
|
Class F
|
47,00
|
49,42
|
-5,64%
|
46,77
|
49,28
|
-5,77%
|
Overall
|
11,61
|
12,12
|
-2,07%
|
11,56
|
12,09
|
-2,12%
|
|
|
|
Enc Time[%]
|
97%
|
100%
|
Dec Time[%]
|
90%
|
92%
|
5.20.3.1.1.1.1.1.2JCTVC-H0055 AHG19: Cross-verification of a method of frame-based lossless coding mode for HEVC JCTVC-H0083 [K. Chono, H. Aoki (NEC)] [late]
(This was initially accidentally uploaded as an "associated resource" instead of as the actual contribution, which led to a lack of awareness of its availability.)
5.20.3.1.1.1.1.1.3JCTVC-H0499 Entropy coding for Lossless compression [S.H. Kim, A. Segall (Sharp)] [late]
In this contribution, a CABAC structure for lossless compression was presented. Considering the statistical difference of residual data between lossless compression and common lossy coding, the contribution proposed a simplified CABAC structure without last position coding. The proposed lossless CABAC was asserted to use a significantly reduced number of context models (about 80%) under the lossless coding conditions. It was further reported that the proposed approach reduces the number of coded bits in high efficiency configurations by 1.7% for AI HE encoding, 1.18% for RA HE, and 0.87% for LB HE when coding losslessly.
Suggestions included removing the last coding, and modifying significance map coding.
The gain is interesting, but requires modification of the entropy coder.
No WD text was in the contribution – this was later uploaded after presentation. No cross-check was available. Further study was encouraged.
5.20.3.1.1.1.1.1.4JCTVC-H0528 AHG19: A QP-based enabling method for lossless coding in HEVC [W. Gao, M. Jiang, H. Yu (Huawei)]
It was asserted that lossless coding can be used in many practical applications, such as medical imaging, remote desktop sharing, online gaming, etc. In this contribution, a method was proposed for lossless coding in HEVC. The contribution proposed using QP to signal a HEVC whether a CU employs lossless coding or not. Specifically, if QP=0 for a CU, (inverse) lossless coding is applied, i.e., inverse quantization, inverse transform, de-blocking filter, SAO, and ALF are all bypassed for this coding unit (see JCTVC-G664). Furthermore, the sample-based intra prediction method (SAP), as proposed in JCTVC-H083 for lossless coding, could be used to replace the existing intra prediction method and signalled by this proposal, i.e. selected by QP=0 as well. With this signalling method, the lossless coding mode could be applied to the entire picture or to individual CUs. The HM5.0 software had been modified to incorporate this signalling method. Test results were shown to demonstrate the effectiveness of this signalling method.
The proposal was to use QP=0 to signal bypass of transform, quantization and all filtering in the loop.
Using QP=0 for the entire sequence is identical to the "anchor" used in H0083, but QP=0 could also invoke lossless coding per CU (switching off the in-loop filters would be similar as in IPCM).
5.20.3.1.1.1.1.1.5JCTVC-H0530 AHG19: A lossless coding solution for HEVC [W. Gao, M. Jiang, H. Yu (Huawei), M. Zhou (TI)]
This proposal was combination in a way that QP=0 is used to signal bypass of the transform, quantization and in-loop filtering, and enabling of SAP.
It was commented that it might even be interesting to have separate control over SAP (e.g. making it switchable per PU).
Decision: The lossless mode that comes "for free" (i.e. bypassing transform, inverse quantization, and in-loop filters) was adopted.
One concern that was raised wasa whether using QP=0 is appropriate, as it might happen that QP becomes negative – e.g. when bit depth is increased. This could presumably be done as in AVC where an bit depth compensated QP value determined the selection.
Alternative solutions would be to either define the "lowest available QP value" for CU-based signalling, or have a separate syntax element (similar to I_PCM); with an enabling flag for either of these solutions at slice header / PPS / SPS / APS. This was then discussed in plenary session. Decision: Use an enabling flag in the SPS, which expresses that lowest QP value is used to switch to lossless.
The SAP and entropy coding modification H0499 could be interesting for a future "lossless profile" and should be further studied.
5.20.3.1.1.1.1.1.6JCTVC-H0673 AHG19: Cross-check of H0528 and H0530 on lossless coding [Y. H. Tan, C. Yeo, Z. Li (I2R)] [late]
Dostları ilə paylaş: |