JVET-AG0107 AHG9: On phase indication SEI message [T. Chujoh, T. Ikai (Sharp), K. Kawamura (KDDI)]
VVC can change the picture resolution in a sequence by using RPR; however, most video applications display decoded pictures with a constant resolution, even with decoded resolution changes dynamically. When the same phase information for the resampling process of RPR is not identified between the transmitting and receiving sides, pixels of the picture may shift in time when the resolution changes dynamically in the sequence. To avoid this phenomenon, the phase indication SEI message has been adopted in VSEI to share the phase information between the transmitting and receiving sides. The remaining problem is that the bitstreams, defined as the phase information implicitly but without the SEI, are not interoperable. This contribution proposes the following:
- The phase indication SEI message is moved to the VVC text, which is made as a normative SEI message.
- Syntax elements of the phase indication SEI message have the inferred values. Their values mean the co-sited positions. Then, even if there is no phase indication SEI message in the bitstream, the phase information is defined.
These proposals do not change the conformance bitstreams and the reference software.
It was commented that it does not make a difference whether something is in VSEI or VVC. It was further commented that VUI may be a better place. It also seems to be more a display issue which is out of scope of our standards.
It was asked why the co-sited position should be defined as default when the SEI is not present? It would still have the undesired effect when the downsampler generates a different position. However, it might be assumed that an implement would use a downsampler that matches the inferred value when no SEI is present. This could be augmented with an informative note for the downsampler in case of RPR.
It was asked how serious the problem is? For some sequences, it may not even be visible. Further, other effects such as loss of structure is happening at the switching point.
Further study was recommended on the possibility of defining a default phase position when the SEI message is not present, and adding an informative remark about what an encoder should do.