JVET-Q0166 On the adaptive colour transform [H. Huang, A. K. Ramasubramonian, C.-C. Chen, V. Seregin, T. Hsieh, Y.-H. Chao, W.-J. Chien, G. Van der Auwera, M. Karczewicz (Qualcomm)]
In the 16th JVET meeting, the adaptive colour transform (ACT) for 4:4:4 video coding was adopted in VVC. It is asserted that the following issues are present in the current ACT design:
- The quantization parameter (QP) after adding ACT offset may be out of QP range.
- The chroma ACT QP offset may be added together with regular chroma QP offset.
- The chroma QP used in deblocking may be different from quantization stage due to the ACT offset, which violates the adoption of JVET-P1001.
- The inverse ACT increases the dynamic range of reconstructed residual that may exceed 16 bits. And the input of ACT may exceed 16 bits when the JCCR mode is 2 (i.e., Cr = -Cb).
- The ACT QP offsets are fixed instead of signalled in high level syntax as in HEVC.
- The Cb or Cr ACT QP offset may be applied in JointCbCr modes.
It is proposed to address the above issues by the following:1) add ACT QP offset before clipping of QP; 2) CU-level switching of QP offset values for reconstructing residuals according to the ACT CU-level flag as in HEVC; 3) align the chroma QP at the deblocking stage to that at the dequantization stage as in adopted JVET-P1001; 4) clip the input and output of inverse ACT to 16 bits dynamic range. 5) signal ACT QP offsets as in HEVC; 6) use a separate chroma ACT QP offset for JointCbCr mode 2 to align with chroma QP offset in non-ACT mode.
Item 1: As a solution to the invalid range (which is targeting the same issue as JVET-Q0098, etc.), it is suggested to apply the ACT offset before the original clipping. This however also would use the modified chroma QP value for deblocking. The proposal does this by purpose, but this may not be the right way to do
Item 4: Clipping aspect is not that relevant – JVET-Q0513 has a better solution.