Back to Search Document details
16th Meeting: Geneva, October 2019 2019-09-18 07:11
CE7-2.2: QP dependent binarization in TS residual coding
Authors: Marta Karczewicz Qualcomm H. Wang Qualcomm Yung-Hsuan Chao Qualcomm Muhammed Coban Qualcomm
Abstract
In this proposal, a new binarization method for transform skip residual coding is proposed, where the binarization of absolute coefficient levels is adjusted based on the dynamic range of the quantized residuals. The proposed method is implemented on top of VTM6.0 (CE7-2.2a). For standard QP (22, 27, 32, 37), the results show: Class F: -0.03% AI, -0.22% RA, -0.13% LB BD-Rate, and Class TGM: -0.01% AI, -0.11% RA, -0.09% LB BD-Rate versus VTM6.0. For low QP (2, 7, 12, 17), the results show: Class F: -1.01% AI, -1.04% RA, -1.06% LB BD-Rate, and Class TGM: -5.60% AI, -5.28% RA, -3.56% LB BD-Rate versus VTM6.0.
Meeting Notes
JVET-P0076 CE7-2.2: QP dependent binarization in TS residual coding [M. Karczewicz, H. Wang, Y.-H. Chao, M. Coban (Qualcomm)]

In this proposal, a new binarization method for transform skip residual coding is proposed, where the binarization of absolute coefficient levels is adjusted based on the dynamic range of the quantized residuals. The proposed method is implemented on top of VTM6.0 (CE7-2.2a). For standard QP (22, 27, 32, 37), the results show: Class F: -0.03% AI, -0.22% RA, -0.13% LB BD-Rate, and Class TGM: -0.01% AI, -0.11% RA, -0.09% LB BD-Rate versus VTM6.0. For low QP (2, 7, 12, 17), the results show: Class F: -1.01% AI, -1.04% RA, -1.06% LB BD-Rate, and Class TGM: -5.60% AI, -5.28% RA, -3.56% LB BD-Rate versus VTM6.0.

In addition, the proposed binarization is tested with varying QPs across transform skip coded blocks (CE7-2.2b). For standard QP (22, 27, 32, 37), the results show: Class F: -0.14% AI, -0.28% RA, -0.13% LB BD-Rate, and Class TGM: -0.26% AI, -0.30% RA, -0.14% LB BD-Rate versus VTM6.0 with the same varying QPs applied. For low QP (2, 7, 12, 17), the results show: Class F: -1.46% AI, -1.64% RA, -1.10% LB BD-Rate, and Class TGM: -5.69% AI, -4.62% RA, -3.51% LB BD-Rate versus VTM6.0 with the same varying QPs applied.

References:
JVET-O0559 JVET-O2027
Cited:
JVET-P0027 X JVET-P0485
ARCHIVE v1 :
JVET-P0076-CE7-2.2a-alternative-lowqp.xlsm JVET-P0076-CE7-2.2a-alternative.xlsm JVET-P0076-CE7-2.2a-lowqp.xlsm JVET-P0076-CE7-2.2a.xlsm JVET-P0076-CE7-2.2b-alternative-lowqp.xlsm JVET-P0076-CE7-2.2b-alternative.xlsm JVET-P0076-CE7-2.2b-lowqp.xlsm JVET-P0076-CE7-2.2b.xlsm JVET-P0076-v1.docx JVET-P0076-WD.docx
Decisions
In addition, the proposed binarization is tested with varying QPs across transform skip coded blocks (CE7-2.2b). For standard QP (22, 27, 32, 37), the results show: Class F: -0.14% AI, -0.28% RA, -0.13% LB BD-Rate, and Class TGM: -0.26% AI, -0.30% RA, -0.14% LB BD-Rate versus VTM6.0 with the same varying QPs applied. For low QP (2, 7, 12, 17), the results show: Class F: -1.46% AI, -1.64% RA, -1.10% LB BD-Rate, and Class TGM: -5.69% AI, -4.62% RA, -3.51% LB BD-Rate versus VTM6.0 with the same varying QPs applied.
Citation