Back to Search Document details
11th Meeting: Ljubljana, July 2018 2018-07-14 09:09
CE6-Related : Matrix multiplication based NSST with reduced memory map
Authors: Mehdi Salehifar LGE Moonmo Koo LGE J. Lim LGE S. Kim LGE
Abstract
In JVET-J0017 [1] LGE’s response to the call for proposals (CFP), a new Non-Separable Secondary Transform (NSST) called Reduced Secondary Transform (RST) was proposed and this RST is being investigated in CE 6.2.6. In this contribution, more detailed analysis on test results of CE 6.2.6 is reported and further updated results is shown based on proposed NSST kernel mapping for memory reduction. Specifically the following three tests are investigated;
Meeting Notes
JVET-K0100 CE6-Related: Matrix multiplication based NSST with reduced memory map [M. Salehifar, M. Koo, J. Lim, S. Kim (LGE)]

A non-separable secondary transform (NSST) called reduced secondary transform (RST) was proposed and investigated in CE 6.2.6.

A direct matrix multiplication NSST for 4x4 NSST (16x16 direct matrix multiplications) and 8x8 NSST (16x64 direct matrix multiplication) is introduced and investigated in this contribution. relative to a full secondary transform, this reduces the multiplication and multilayer complexity. Also results with memory reduction also reported.

This uses 16 secondary transform kernels instead of ~100 as used in the CE test.

Ordinarily, implementing a secondary transform larger than 4x4 has high complexity. This proposal use a sparse matrix decomposition to simplify the computation. The number of transform kernels is also reduced.

References:
JVET-J0017 JVET-J1010 JVET-J1026
Cited:
JVET-K0026 JVET-K0440 X JVET-K1026
PATENTS:
EP3723373B1 0.34 2023-06-07 EP3806475B1 0.32 2023-05-10 EP3723372B1 0.30 2025-11-05 US20230319311A1 0.28 2023-10-05 US20210211727A1 0.26 2021-07-08 US11818352B2 0.24 2023-11-14 US11601679B2 0.22 2023-03-07 US11178417B2 0.20 2021-11-16
ARCHIVE v4 :
JVET-K0100.docx JVET-K0100_r1.docx JVET_K0100.pptx VTM_1_1_RST4x4Only 6Sets_fullSeq.xlsm VTM_1_1_RST4x4Only_eff19Sets_fullSeq.xlsm VTM_1_1_RST4x4RST8x8_6Sets_fullSeq.xlsm VTM_1_1_RST4x4RST8x8_eff19Sets_fullSeq.xlsm
Decisions
Ordinarily, implementing a secondary transform larger than 4x4 has high complexity. This proposal use a sparse matrix decomposition to simplify the computation. The number of transform kernels is also reduced.
Citation