Back to Search Document details
12th Meeting: Macao, October 2018 2018-10-04 03:26
AHG 16: Transform-free coding for 2×N or N×2 chroma blocks
Abstract
In this contribution, no transform/inverse-transform is required for tiny blocks with sizes 2×N or N×2. The 2-point transform could be removed accordingly. Two solutions are provided. In the first solution, tiny blocks are always coded with all residues equal to 0. In this case, the 2×2 coding group scan is also removed. In test 2, tiny blocks are always coded with the transform-skip mode. In test 1, simulation results reportedly show 0.06%/1.55%/1.76% and 0.11%/3.31%/3.57% BD-rate changing on Y/Cb/Cr components for All Intra (AI) and Random Access (RA) configurations, respectively, in average compared with VTM-2.0.1. Meanwhile, it reports encoder running time is also reduced by 2%. In test 2, simulation results reportedly show 0.11%/1.40%/1.58% and 0.13%/2.92%/3.10% BD-rate changing on Y/Cb/Cr components for AI and RA configurations, respectively.
JVET-L0334 AHG 16: Transform-free coding for 2×N or N×2 chroma blocks [K. Zhang, L. Zhang, H. Liu, Y. Wang, P. Zhao, D. Hong (Bytedance)]

In this contribution, no transform/inverse-transform is required for tiny blocks with sizes 2×N or N×2. The 2-point transform could be removed accordingly. Two solutions are provided. In the first solution, tiny blocks are always coded with all residues equal to 0. In this case, the 2×2 coding group scan is also removed. In test 2, tiny blocks are always coded with the transform-skip mode. In test 1, simulation results reportedly show 0.06%/1.55%/1.76% and 0.11%/3.31%/3.57% BD-rate changing on Y/Cb/Cr components for All Intra (AI) and Random Access (RA) configurations, respectively, in average compared with VTM-2.0.1. Meanwhile, it reports encoder running time is also reduced by 2%. In test 2, simulation results reportedly show 0.11%/1.40%/1.58% and 0.13%/2.92%/3.10% BD-rate changing on Y/Cb/Cr components for AI and RA configurations, respectively.

Implicitly enforcing zero coding is not appropriate in particular in the low QP range.

The loss in chroma is relatively large (and even reflected in luma loss, as the bit rate is apparently increased).

Generally, a length 2 transform is not critical in complexity.

No action was taken on this.

References:
JVET-K1010
PATENTS:
US10679415B2 0.34 2020-06-09 AU2015204790C1 0.32 2019-07-04 CN111432220B 0.30 2022-05-06 CN107005717B 0.28 2020-04-07 CN110519604B 0.26 2022-04-01 EP3422724B1 0.24 2024-05-01 CN109644279B 0.22 2023-09-22 CN104380737B 0.20 2018-10-26
Decisions
No action was taken on this.
Citation