JVET-AC0180 AHG 7: Reference frame padding for GDR [T. Poirier, F. Aumont (InterDigital)]
This contribution proposes to pad the dirty area using samples from the clean area instead of setting unavailable samples to 2BD-1 for reference pictures used by CUs in the clean (refreshed) area.
Simulations were conducted with ECM-7.0 under low-delay B configuration specified by AHG 7. The results show a BD-rate gain of 0.2% overall, 0.2% in class C and 0.4% in class E.
Cross checker confirmed the simulation results.
Normative GDR-related aspects in ECM differ from normative GDR-related aspects in VVC as follows:
- Handling of “dirty” and “clean” regions
- Asymmetric deblocking across the dirty/clean boundaries
- Virtual boundary processing
It was commented that the GDR-related deblocking filter adoption from last meeting (JVET-AB0171) has not yet been integrated into ECM7. And it was asked if that aspect would have interaction with the current proposal. The proponent thinks there could be some interactions between the two.
Integration of JVET-AB0171 has encountered the following issue and been delayed:
- when virtual boundary is used (this is outside of CTC), 135-degree SAO would cause the dirty area and clean area to be mixed, the patch to fix the issue has been submitted and merged recently (after ECM7 has been released)
Since then, patch for JVET-AB0171 has been submitted, it is pending review and not yet integrated into ECM. It was confirmed that JVET-AB0171 will be integrated soon and should be available as part of ECM8.
It was commented that the testing of this proposed low-level change would not affect the CTC, and would need to use the GDR-specific conditions defined by AHG7.
It was commented that during AHG report reviews, it was decided to stop AHG7 and put the AHG7 activities requiring normative changes into AHG12.
It was commented that GDR-specific testing conditions have not been documented, and need to be clearly specified in the EE2 descriptions.
It was agreed to investigate this in an EE, using GDR-specific testing conditions.