Back to Search Document details
17th Meeting: Brussels, January 2020 2020-01-10 15:06
On modifications of intra prediction process
Authors: A. Filippov Huawei V. Rufitskiy Huawei Biao Wang Huawei S. Esenlik Huawei Anand Meher Kotra Huawei Han Gao Huawei Elena Alshina Huawei Alexey Filippov
Abstract
This contribution proposes to modify intra prediction process. In particular, PDPC scaling is harmonized with intra subpartitions (ISP) coding mode, so that the number of reference samples used is maximized. The simulation results for method 1 reportedly show the Y/U/V BD-rate changes of 0.00%/- 0.02%/0.00%, -0.01%/0.01%/0.03% and 0.04%/0.01%/-0.16% for AI, RA and LDB configurations, respectively. The simulation results for method 2 reportedly show the Y/U/V BD-rate changes of 0.00%/- 0.03%/0.00% and 0.00%/0.06%/0.07% for AI and RA configurations, respectively. More significant coding efficiency improvements (-0.12%/0.57%/-0.27% and -0.10%/-0.32%/0.24% for method 1 and method 2, respectively) are achieved for class E in LDB case.
Meeting Notes
JVET-Q0464 On modifications of intra prediction process [A. Filippov, V. Rufitskiy, B. Wang, S. Esenlik, A. M. Kotra, H. Gao, E. Alshina (Huawei)]

This contribution proposes to modify intra prediction process. In particular, PDPC scaling is harmonized with intra subpartitions (ISP) coding mode, so that the number of reference samples used is maximized. The simulation results for method 1 reportedly show the Y/U/V BD-rate changes of 0.00%/- 0.02%/0.00%, -0.01%/‑0.01%/0.03% and ‑0.04%/0.01%/-0.16% for AI, RA and LDB configurations, respectively. The simulation results for method 2 reportedly show the Y/U/V BD-rate changes of 0.00%/- 0.03%/0.00% and 0.00%/‑0.06%/0.07% for AI and RA configurations, respectively. More significant coding efficiency improvements (-0.12%/0.57%/-0.27% and -0.10%/-0.32%/0.24% for method 1 and method 2, respectively) are achieved for class E in LDB case.

In version 2, a symmetric intra interpolation filter selection is used, named as method 3. This method is independent from the other proposed ones. Compared to the current asymmetric interpolation filter selection approach in VTM7.0, method 3 achieves luma BD-rate changes of 0.00%/0.01%/-0.06% for AI, RA and LDB configurations, respectively. In addition, a combined test of method 1 and 3 are performed, the BD-rate changes of 0.00%/-0.01%/-0.04% for AI, RA and LDB configurations, respectively.

In version 3, a test with minimal changes to the specification (method 0) is carried out. Minor consistent improvement of coding performance on RA is observed for this method. The simulation results for method 0 reportedly show the Y/U/V BD-rate changes of 0.00%/‑0.03%/‑0.01%, -0.02%/‑0.04%/0.01% and 0.00%/-0.08%/-0.14% for AI, RA and LDB configurations, respectively.

It is not obvious that there are problems solved by the various methods. The elements of the proposal do not appear as a simplification, and the compression benefit is low. Further, at least method 3 was added only recently, which is basically disconnected from the original proposal.

No need for action.

References:
JVET-N1010
Cited:
JVET-Q0746 X
PATENTS:
202137043768 0.34 202027056741 0.33 WO/2020/190179 0.32 3891977 0.31 202017055068 0.30 202037057289 0.28 WO/2020/055292 0.27 202037054986 0.26
ARCHIVE v3 :
JVET-Q0464-test0.xlsm JVET-Q0464-test1.xlsm JVET-Q0464-test2.xlsm JVET-Q0464-test3.xlsm JVET-Q0464-test4.xlsm JVET-Q0464-test5.xlsm JVET-Q0464-v2.docx JVET-Q0464-v3.docx
Decisions
No need for action.
Citation