JVET-AM0113 AHG9: Supporting pre-processing aware post processing in the NNPFC SEI message [M. Damghanian, M. Pettersson, R. Sjöberg (Ericsson)]
This contribution proposes to add support for pre-processing aware post-processing to the NNPFC SEI message. The proponents claim that this addition supports usage of the sandwiched codec architecture presented in JVET-AL0060. The followings are proposed for VSEI v4:
Proposal 1
- To add a new nnpfc_purpose entry in Table 20 for pre-processing aware post-processing, with the interpretation that the post-processing corresponds to an applied pre-processing.
- To add a new nnpfc_purpose entry in Table 20 to indicate that the NNPF process is improving visual quality towards a quality metric.
- To add a new syntax element nnpfc_quality_metric_idc where the values of the syntax element describe the type of quality metric the video resulting from the NNPF process is enhanced towards.
Proposal 2
- To add a new syntax element nnpfc_decoded_video_distortion_idc where the value of the syntax element specifies the level of distortion expected due to a pre-processing.
Is being proposed for VSEI v4.
It was suggested that there is not a backwards compatibility issue with making the proposed changes in a future version of VSEI.
It was noted that the EOI SEI message describes the decoded picture. It was suggested that the EOI could communicate that the decoded pictures was not suitable for human viewing, and use the existing NNPF SEI to indicate the post-filter operation.
It was commented that there may be advantages in having a single SEI message address the sandwich codec use case.
It was noted that NNPF already has a nnpfc_for_human_viewing_idc, but it describes the output of the NNPF process rather than the decoded picture.
The proposed wording for the purpose regarding pre-processing has been applied doesn’t seem specific to a sandwich codec, and could apply to other types of pre- and post-processing, e.g. downsampling and upsampling.
It was questioned if the proposal is adding signalling in the most appropriate location to address the described use case.
The sandwich coding use case is seen to be interesting.
Further study is encouraged.
AI usage restrictions SEI message (3)
Contributions in this area were discussed during 1000–1015 on Friday 27 June 2025 (chaired by J. Boyce).