JVET-P0455 Non-CE8: IBC slice level on/off [X. Xu, X. Li, S. Liu (Tencent)]
In this contribution, it is proposed to control the usage of IBC at slice level, in addition to the current IBC SPS flag. That means, when IBC SPS flag is true, for each slice, an encoder can choose to turn on or off the usage of IBC. The indication of whether IBC is used for a slice is through the value of a slice level variable MaxNumIbcMergeCand. When this variable is set equal to 0, IBC is turned off for the slice. On top of this change, an encoder algorithm is provided to switch IBC on/off using the same logic as disabling MMVD fractional-pel offset, which already exists in the VTM. The intend of the suggested algorithm is that IBC mode is turned off for slices where IBC may not contribute much to the BD rate reduction while IBC mode is turned on for screen content materials. Simulation results report that:
- The BD rate/runtime changes for CTC (class A to E) average are -0.02%/0.02%/0.03% and 103%/100%/100% for AI/RA/LB, separately, when compared to VTM-6.0+IBC=0 anchor;
- The BD rate/runtime changes for Class F average are 0.00%/0.00%/0.00% and 98%/99%/99% for AI/RA/LB, separately, when compared to VTM-6.0+IBC=1 anchor
- The BD rate/runtime changes for Class TGM 1080p average are 0.00%/0.00%/0.00% and 98%/99%/99% for AI/RA/LB, separately, when compared to VTM-6.0+IBC=1 anchor
The control via the MaxNumIbcMergeCand is undesirable. It would be better to have a flag at picture level, if at all. There would be not much harm for an encoder that wants to use IBC only for some pictures, just not using it for the other ones. The penalty for IBC flag if not used should be rather small (proponents mentioned 0.4% for some sequences).
Not evident that this is needed.
In terms of HLS, the MaxNumIbcMergeCand syntax element should be handled like the same element for inter merge an affine merge.