JVET-P0485 AHG18: Performance of CE7-1.2d and CE7-2.2 on lossless coding [Y.-H. Chao, A. Nalci, H. Wang, M. Coban, M. Karczewicz (Qualcomm)]

This document reports the transform skip modification proposed in CE7-1.2d-alternative and the transform skip binarization proposed in CE7-2.2a-alternative for lossless coding. Two test scenarios are tested: In the first scenario, the anchor base uses only regular transform residaul coding (TR) for lossless coding with BDPCM support in luma and the test base uses only transform skip (TS) residual coding with BDPCM support in luma and chroma. In the 2^nd scenario, the anchor base is AHG18 lossless branch and the test base is the method 1 in JVET-P0082, which enables transform (TR) residual coding, transform skip (TS) residual coding, and BDPCM for lossless coding. The results show:

• 1^st test scenario (TR only v.s. TS only)
  • CE7-1.2d-alternative:

Overall: -1.64% AI, 0.40% RA, xxx% LB

Class F: -2.92% AI, -1.39% RA, -2.37% LB

Class TGM: -7.03% AI, -9.45% RA, xxx% LB

• 2^nd test scenario (AHG18 v.s. TS + TR)
  • CE7-1.2d-alternative:

Overall: -2.72% AI, -1.28% RA, -1.51% LB

• • CE7-2.2a-alternative:

Overall: -3.30% AI, -1.69% RA, -1.92% LB

Note: xxx = values not provided in contribution.

No detailed presentation was needed, as this was not relevant in context of the new TS residual coding method.

There is a new document JVET-P1028, which combines this with the new TS residual coding.