JVET-AL0127 AHG9: Danmu information SEI message [J. Xu, Y.-K. Wang, L. Zhang (Bytedance)]
Chaired by J. Boyce.
This contribution proposes a new SEI message to convey danmu information, to support interoperability for videos with danmus. A comparison with the MPEG timed text standard is provided.
The contribution asserts that following aspects make TTML unsuitable to represent danmus:
- Attributes are described in plain text, which is not bit rate efficient. It can be a serious issue especially when the number of danmus is large.
- When an attribute is described, it describes the exact information, which does not allow any flexibility for the video displayer.
- To support even simple animation needs much overhead.
A question was raised about persistence and CLVS duration and what the impact of IDRs would be. Can a danmu signalled in one CVLS be displayed in the next CLVS after an IDR is present? No, the persistence would cancel it for the new CLVS. It would be possible to repeat signalling of the Danmu in a new SEI message in the next CLVS.
The archival use case was suggested as being more useful for the proposed SEI than the real time streaming case.
Decision: add to TuC.