JVET-R0309 [AHG16] Clean-up on palette predictor update for local dual tree [H. Jang, J. Nam, S. Yoo, N. Park, S. Kim, J. Lim (LGE)]
This contribution proposes to clean-up regarding predictor palette update for local dual tree structure. Three alternative clean-up methods are suggested as described below:
- Clean-up 1. There is mismatch between SPEC and SW. Here, propose to fix SPEC to align with SW regarding local dual tree predictor palette update process.
- Clean-up 2. Update predictor palette entry only for luma component but not for chroma component in local dual tree structure.
- Clean-up 3. Remove palette predictor update process of local dual tree structure.
The suggested clean-up method 2 shows {0.00%, 0.00%, -0.06%} and method 3 shows{0.00%, 0.07%, 0.10%} coding performance respectively for average of ClassF/SCC under CTC with palette on. It is asserted that proposed clean-up methods save H/W power consumption with closed to zero coding loss and also cut worst case decoding latency by disabling predictor palette update for local dual tree.
Presented Thursday 9 April at 1427 UTC (chaired by JRO)
For “clean-up 1”, spec should be aligned with software. This is also included in JVET-R0333. See further notes there.
Items #2 and #3 try to simplify the interaction of palette with local dual tree structures. It is agreed that there is no problem in the spec such as unspecified decoder behaviour. Also in local dual tree the total predictor table size is identical with single tree. No need for action on these items.