Joint Video Experts Team (jvet) of itu-t sg 6 wp and iso/iec jtc 1/sc 29/wg 11
Yüklə 4,04 Mb.
|
Results vs. BMS:
The following table shows additional optional tests (only for BMS):
2.4.1.4.i is modifying the parameters of deblocking (encoder only). Before such a change is done, it should be tested for visual quality. From discussion in track B: In general, we are seeking for complexity reduction compared to BMS-GALF (giving -5.3% compared to VTM) rather than increasing complexity. Methods of increasing complexity (e.g. pixel based classification, multiple classifiers, classification or switching also for chroma) give at most another 0.25% over BMS-GALF Options to reduce complexity: - Simpler classifiers (no results from CE) - Classification block sizes (4x4 loses 0.15% relative to BMS-GALF) - Subsampling in classification - Avoiding pre-defined filters (BMS-GALF has 400, and some CE results reduce them to less or zero) - Filter size (7x7 loses 0.2%), which implies less operations and line buffer reduction. - Multiplication simplifications (e.g. bit shifting loses <0.1%, see JVET-K0215) - Omitting classification (e.g. switching CTU based, 2.4.2.2e loses 2% relative to BMS-GALF) It was suggested to make a subjective comparison of VTM vs. a simplified classification based approach and an approach without classification, but several experts expressed they would not expect that differences would be visible. Breakout activity (L. Zhang) to assess the implementation complexity (memory accesses, memory for storing predefined filters, operations per sample in terms of mul, add, comp, reload operations, etc.) of the CTU based approach and different aspects of the simplified classification based approaches. This should gives us data to assess the complexity aspect versus the performance. Another aspect that requires clarification is the signalling of the filter coefficients. The current BMS-GALF solution of signalling between slice header and first CTU is not desirable. Also the case of multiple slices per picture should be supported. The BoG report JVET-K0521 was presented in track B Sat. 14th 1900. An analysis was done on the algorithmic and memory complexity of different algorithms. The two solutions with lowest complexity are 2.4.1.4.c and 2.4.2.2.e. Both are approximately identical in terms of number of multiplications, significantly reduced relative to BMS-GALF 2.4.1.4.c has more additions and shifts The classification at 4x4 block level does not have high complexity as compared to the filtering itself. It is agreed that the classification based approach provides the best performance (1.5% coding gain). Some concern is however raised with regard to the representation/coding of filter coefficients, in particular concerning the prediction aspects. This should be further studied, to make the representation of ALF parameters more straightforward. Decision(VTM): Adopt JVET-K0371 (based on subtest 2.4.1.4c, 4x4 classification based on Laplacian for luma only, 7x7 luma, 5x5 chroma filters); disable prediction of adaptive filters from fixed filter set; disable temporal prediction; put filter parameters into slice header; Enabling flag at CTU level. Further investigation in ongoing CE: Prediction of filter parameters; enabling at sub-CTU level; other classification approaches from 2.4.1.1 and 2.4.1.2. Also study aspects of 2.4.1.3 that replace multiplication in filtering by shift operations. Was again reviewed later. Draft text was provided (JVET-K0564), containing description of the method described above (syntax, semantics, decoding process).
Some suggestions were made as follows:
Looks generally OK, but probably needs more detailed check by editors. Sub-CE5: Non-local filter
Notes: SVD: singular-value decomposition
Results vs. VTM
Results vs. BMS
2.5.1: Current SVD as used in CE uses floating point implementation. There was a proposal on fixed point implementation by the last meeting, but this was not investigated in CE. Overall, decoder complexity of 2.5.1 is very high. The compression is reduced (but still around 1.5% rate reduction) when combined with GALF (in BMS). The method itself is probably more complex than GALF but provides less gain on top of VTM. 2.5.2/2.5.3 These two approaches provide 1.2% (non-local mean filter) and 1% (Hadamard based noise suppression). This reduces to roughly 0.6% when used in BMS (probably less gain when combined with ALF). Both of these approaches are more complex than e.g. the current design of the bilateral filter from CE2.1, which still gives similar gain (0.5%) for BMS. No action at this moment, further study for significant complexity reduction. Would also be interesting to identify in which cases the non-local filters are able to provide gain that the other loop filters cannot provide. For the SVD based approach, the proponents report that the usage is between 10% and 50%, depending on sequence. As a general statement, VVC should have an overall clean design, and it should be avoided to operate a large number of loop filters sequentially, unless they provide substantial individual gains, and not being overly complex.
Note: This document was withdrawn by mistake. Proponents were asked to register it again under a new number. – Add this number JVET-K0176 CE2: SAO modification (CE2.3.4) [J. . Chen, K. . Choi, C. . Kim (Samsung)] JVET-K0192 CE2-3.3 SAO Palette results and discussion [P. . Bordes, F. . RacapeRacapé (Technicolor)] JVET-K0215 CE2: ALF with Multiplication Replaced by Bit-Shifting (Test 4.1.3) [S. . Esenlik, Z. . Zhao, J. . Chen (Huawei)] JVET-K0231 CE2.1.2: Bilateral filter - spatial filter strength adjustment [Y.-C. Su, C.-Y. Chen, Y.-W. Huang, S.-M. Lei (MediaTek)] JVET-K0232 CE2.2.1.3: Long deblocking filters [C.-M. Tsai, T.-D. Chuang, C.-W. Hsu, C.-Y. Chen, Y.-W. Huang, S.-M. Lei (MediaTek)] JVET-K0233 CE2.3.1: SAO with EO sign constraints removal and more EO patterns [C.-Y. Chen, C.-Y. Lai, Y.-W. Huang, S.-M. Lei (MediaTek)] JVET-K0235 CE2.4.2.2: CTB-based ALF with slice filter sets [C.-Y. Chen, Y.-C. Su, Y.-W. Huang, S.-M. Lei (MediaTek)] JVET-K0236 CE2.5.2: Non-local mean in-loop filter [C.-Y. Lai, C.-Y. Chen, Y.-W. Huang, S.-M. Lei (MediaTek)] JVET-K0285 CE2.4.2.1: Multiple-feature based adaptive loop filter [W.-Q. Lim, J. . Erfurt, M. . Siekmann, H. . Schwarz, D. . Marpe, T. . Wiegand (HHI)] JVET-K0307 CE2-2.1.1: Long deblocking filters and fixes [K. . Andersson, Z. . Zhang, R. . Sjöberg (Ericsson)] JVET-K0315 CE2: Deblocking Improvements for Large CUs (Test 2.1.7) [W. . Zhu, K. . Misra, A. . Segall, P. . Cowan (Sharp)] JVET-K0324 CE2: Tests on SAO design in CE2.3.2 [A. . Gadde, D. . Rusanovskyy, M. . Karczewicz (Qualcomm)] JVET-K0327 CE2: Adaptive Loop Filter Simplification (Test 2.4.1.2) [R. . Vanam, Y. . He, Y. . Ye (InterDigital)] JVET-K0334 CE2: Tests on long deblocking (CE2.2.1.4) [D. . Rusanovskyy, J. . Dong, M. . Karczewicz (Qualcomm)] JVET-K0371 CE2.4.1.4: Reduced filter shape size for ALF [M. . Karczewicz, N. . Shlyakhov, N. . Hu, V. . Seregin, W.-J. Chien (Qualcomm)] JVET-K0384 CE2.1.3: In-loop bilateral filter [A. . Gadde, V. . Seregin, W.-J. Chien, M. . Karczewicz (Qualcomm)] JVET-K0386 CE2-2.2.2: Luma-adaptive deblocking filter [S. . Nemoto, S. . Iwamura, A. . Ichigaya (NHK)] [late] JVET-K0393 CE2: Extended Deblocking Filter (CE2.2.1.2) [Kyohei Unno, Kei Kawamura, Yoshitaka Kidani, Sei Naito (KDDI)] [late] JVET-K0435 Crosscheck for CE2-5.1 [W. . Zhang (Hulu)] [late] Yüklə 4,04 Mb. Dostları ilə paylaş:
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||