International organisation for standardisation
Yüklə 9,08 Mb.
|
5.4.2Contributions5.4.2.1.1.1.1.1.1JCTVC-F221 CE4 Subtest1: the signalling of minCUDQPsize at LCU level [C. Pang, O. C. Au, X. Wen, J. Dai, F. Zou, X. Zhang (HKUST)] 5.4.2.1.1.1.1.1.2JCTVC-F525 CE4 Subtest 1: Crosscheck of JCTVC-F221 HKUST signalling of minCUDQPsize at LCU level [D. Hoang (Zenverge)] [late upload 07-08] 5.4.2.1.1.1.1.1.3JCTVC-F102 CE4 Subtest 2: Cross-check report of HKUST’s proposal JCTVC-F156 (test 2.1.a) [H. Aoki, K. Chono, Y. Senda (NEC)] 5.4.2.1.1.1.1.1.4JCTVC-F103 CE4 Subtest 2: QP prediction based on intra/inter prediction (test 2.4.b) [H. Aoki, K. Chono, Y. Senda (NEC)] 5.4.2.1.1.1.1.1.5JCTVC-F156 CE4 Subtest2: QP prediction with previous CU (subtest 2.1.a) [C. Pang, O. C. Au, X. Wen, F. Zou, J. Dai, X. Zhang (HKUST)] 5.4.2.1.1.1.1.1.6JCTVC-F159 CE4 Subtest 2: QP prediction based on intra prediction (test 2.3.g) [H. Aoki, K. Chono (NEC), M. Kobayashi, M. Shima (Canon)] 5.4.2.1.1.1.1.1.7JCTVC-F300 CE4 Subtest 2: Delta QP prediction results of test 2.2.b and 2.3.f [M. Kobayashi, M. Shima (Canon)] 5.4.2.1.1.1.1.1.8JCTVC-F332 CE4 Subtest 2: QP prediction from spatially neighboring CUs (test 2.3.b, 2.3.c) [M. Coban, M. Karczewicz (Qualcomm)] 5.4.2.1.1.1.1.1.9JCTVC-F346 CE4 Subtest 2.3: Cross-check of Qualcomm’s proposal (JCTVC-F332) on delta QP prediction [J. Jia, J. Park (LGE)] 5.4.2.1.1.1.1.1.10JCTVC-F619 Crosscheck of JCTVC-F332 CE4 Subtest 2.3.c : QP prediction from spatially neighboring CUs [D. Hoang (Zenverge)] [late upload 07-09] 5.4.2.1.1.1.1.1.11JCTVC-F421 CE4: X-check of 2.2.b and 2.3.f [K. Sato (Sony)] 5.4.2.1.1.1.1.1.12JCTVC-F420 CE4: Result of 2.3.d and 2.3.e [K. Sato (Sony)] 5.4.2.1.1.1.1.1.13JCTVC-F355 CE4 Subtest2: Cross-check report of Sony’s proposal JCTVC-F420 (tests 2.3.d and 2.3.e) by ETRI [S.-C. Lim, H. Y. Kim, J. Lee (ETRI)] 5.4.2.1.1.1.1.1.14JCTVC-F400 CE4 Subtest 2: Delta QP prediction results [M. Kobayashi, M. Shima (Canon)] 5.4.2.1.1.1.1.1.15JCTVC-F640 CE4 Subtest 2: Cross-check report for JCTVC-F400 (test 2.3.g and 2.3.f combined) [H. Aoki, K. Chono, Y. Senda (NEC)] [late reg. 07-05, upload 07-06] 5.4.2.1.1.1.1.1.16JCTVC-F644 CE4 Subtest 2: Cross-check report for JCTVC-F648 (test 2.3.g and 2.3.e.r1 combined) [H. Aoki, K. Chono, Y. Senda (NEC)] [late reg. 07-05, upload 07-07] 5.4.2.1.1.1.1.1.17JCTVC-F661 CE4 Subtest 2: combination of test 2.3.g, 2.3.f and 2.3.e [H. Aoki, K. Chono (NEC), M. Kobayashi, M. Shima (Canon), K. Sato (Sony)] [late reg. 07-06, upload 07-08] For intra CU, the intra prediction direction is used to determine from which CU to predict the QP. For inter CUs, a prediction mode based method (using QP from the left or top CU with same prediction mode) or a mode using average. Gain is 0.4% for intra, 0.2-0.3% for inter. Question: Is it a problem to introduce dependencies between prediction mode and de-quantization? Apparently not, it is stored anyway. 5.4.2.1.1.1.1.1.18JCTVC-F707 CE4 Subtest 2: Cross check report of JCTVC-F661 for MinCUDQPSize = 32x32 [M. Coban] [late reg. 07-11, upload 07-12] 5.4.2.1.1.1.1.1.19JCTVC-F689 Cross-check of CE4 Subtest 2, Combination of 2.3.g, 2.3.f and 2.3.e (JCTVC-F661) [Rickard Sjöberg (Ericsson)] [late reg. 07-08, upload 07-13] 5.4.2.1.1.1.1.1.20JCTVC-F394 Cross-check of MV coding JCTVC-F661 [T. Yamamoto (Sharp)] [late upload 07-12] 5.4.2.1.1.1.1.1.21JCTVC-F705 CE4 Subtest 2: Spatial QP prediction: combination of test 2.3.g with 2.3.b/2.3.c [M. Coban, M. Karczewicz] [late reg. 07-11, upload 07-11] Both combinations give approx. 0.4% for intra, 0.2/0.1% for inter. 5.4.2.1.1.1.1.1.22JCTVC-F732 CE4 Subtest2: Verification result of Qualcomm's Spatial QP prediction (JCTVC-F705) [M. Shima (Canon)] [late reg. 07-13, upload 07-16]
This joint contribution asserts that TM5-based subjective quality adaptive quantization is a reasonable method for evaluating QP prediction and delta-QP entropy coding technologies. Commercial video encoders reportedly use similar subjective quality adaptive quantization schemes. The TM5-based subjective quality adaptive quantization was used not only for CE4 Subtest2 but also for new QP prediction and delta-QP entropy coding technologies such as JCTVC-F046, JCTVC-F174, JCTVC-F422, JCTVC-F499, and JCTVC-F577. In CE4 discussions, it had been agreed to adopt the TM5-based subjective quality adaptive quantization into HM software. It was suggested that this implies that the TM5-based subjective quality adaptive quantization is endorsed for evaluating QP prediction and delta-QP entropy coding technologies and that CE4 Subset2 results obtained by using the software were successfully evaluated. In addition, as described in JCTVC-F756, the QP coding BoG participants reportedly agreed that use of TM5-based subjective quality adaptive quantization is realistic enough and should be sufficient for the text condition of the next CE on QP coding. QP prediction schemes were asserted to be orthogonal to other technologies such as intra prediction, in-loop filtering, etc., and it was reportedly expected that a QP prediction scheme (2.3.g+2.3.f+2.3.e) that is suggested by CE4 summary for HM adoption will perform the best among spatial QP prediction proposals of CE4 Subtest2 also in the next version of HM. It was proposed that the 2.3.g+2.3.f+2.3.e proposal (JCTVC-F661 and JCTVC-F705) be adopted into HM4.0. A concern was raised looking at the fact that the current dQP implementation in software is not in the best state (see plenary discussion of Tuesday), and introducing new elements with even more dependencies might cause even more problems (with relatively small gain). The CE should target a consistent bundle of dQP prediction and entropy coding, which is cleaner. It was suggested that the “Most recent QP dependency” should be removed. No action was taken on this. General note: To assess the benefit of more compressive coding of dQP, other more realistic application cases of dQP adaptation (see e.g. JCTVC-F492) would be helpful.
Yüklə 9,08 Mb. Dostları ilə paylaş:
|