JVET-P0517 Support of adaptive color transform for 444 video coding in VVC [X. Xiu, Y.-W. Chen, T.-C. Ma, H.-J. Jhu, X. Wang (Kwai Inc.)]
In this contribution, the adaptive color transform (ACT) tool in HEVC was implemented into the VVC framework, to efficiently reduce the redundancy between three color components of prediction errors in 444 videos. Same as in HEVC, color space conversion in the proposed implementation is carried out in residual domain. Additionally, to better suit the existing VVC tools, two additional changes are proposed: 1) move the ACT enabling/disabling flag from TU-level to CU-level; 2) disable the ACT when the separate-tree partition or intra sub-partition prediction (ISP) mode is applied.
Based on the same test sequences used for HEVC SCC, simulation results reportedly show that when the separate-tree partition is off the proposed ACT implementation provides average {G, B, R} BD-rate savings {9.4%, 3.6%, 4%}, {16.7%, 8.1%, 9.5%} and {23.4%, 10.6%, 12.3%} for AI, RA and LD configurations, respectively. When the separate-tree partition is on, the corresponding {G, B, R} BD-rate savings are {11%, 4.9%, 6.7%} and {20.7%, 9.2%, 11%} for RA and LD configurations.
LMCS is also enabled in both anchor and test, and performed after inverse ACT at decoder. It is reported that LMCS still provides some gain for G. Impact on results seems minor
Including ACT is straightforward, simple computation, and beneficial for case of 4:4:4 RGB coding. For YCbCr, no gain or minor loss is observed.
Decision: Adopt JVET-P0517. SPS flag is included in the spec text
Appropriateness of software and text was confirmed by crosscheckers.