JVET-Q0363 On transform skip residual coding [J. Gan, C. Rosewarne (Canon)]
This contribution proposes removing the second pass of context-coded syntax elements from TSRC. The remaining four context-coded syntax elements are decoded in a single pass, requiring only one CABAC budget check per coefficient. It is asserted that the proposed simplification aligns the CABAC utilisation of RRC and TSRC. Losses relative to VTM-7.0 are 0.01% luma in CTC RA, and 0.37% luma in Class F RA.
The proposal intends to make TSRC more similar to RRC in various aspects. However, a complete unification is not possible, e.g. in terms of scanning order, method of sign coding, etc., such that hardware implementations would likely use separate processing stages anyway. The method particularly loses compression performance in screen content, which is the area where TSRC has most benefit.
No action was taken on this.