We have terms: tier, layer, operating point, and we have tracks.
Layers – defined by the coding system. Tier – a ‘delta’ that is a group of layers. Operating point – an accumulation of tiers that represents, something that can be both decoded and output. Track – a convenient grouping of data to make e.g. decoding easier.
Does every tier represent a new operating point, or does an OP aggregate many tiers, or can tiers be used in multiple OPs? Each tier with its dependent tiers does represent a potential operating point, but the writer is not required to declare every OP as being intended or useful. A tier is always contained in exactly one track.
Is it true that we need tiers because we want several layers in one track? Yes. We note that we even have sub-layers.
(We note that HEVC uses the word ‘tier’ now, which is unfortunate.)
Does the ‘tier’ concept buy us anything? Perhaps sub-layers? We note that the operating point here talks about a set of layers, in any order, and not the ‘top’ of a tier or tiers.
We agree that the design seems to have become complex. Can we take a step back, look at use cases and operations that we want to support, and maybe simplify? We note that extractors give you ‘cookbook’ extraction (do this and get the result you want), whereas the sample groups support ‘table driven’ flexible extraction.
We need a better sample map group which has an explicit mapping-type (the type of the groups mapped to).
We would like to revisit this (or a revision of it) at the next meeting. Ye-Kui in particular volunteers to help co-author a revision.
Dostları ilə paylaş: | 
