Dynamic dropping/enabling of layers for SVC

[marco99zz]

This is done in webrtc with vp9 by setting target bitrate to 0 for the layer to be disabled, and it can be done anywhere in stream and without key frame. Is this feature supported for SVC?

[aboba]

The target bitrate approach has been discussed, but it isn't clear how difficult this would be to implement, particularly for complex scalability nodes (e.g. K-SVC).

Related to:
Issue w3c/webrtc-svc#4
Issue w3c/webrtc-svc#14

[aboba]

@marco99zz I think this can largely be addressed by adding metadata as well as scalabilityMode configuration. This would make it possible to change the scalabilityMode (e.g. from L1T3 to L1T2). Using metadata, the application can identify layers, and then decide what to do with them. So if an application wants to stop or start sending one or more layers (but doesn't want to change the encoder scalabilityMode), it can identify the layer(s) it doesn't want to send (or wants to start sending again).

Metadata proposal: https://docs.google.com/document/d/1O6kxAQPLWhCRHkH-6zutcqEUOUWRS6NFM2CQAvi1VTg/

[chcunningham]

re: bitrate, the thinking is our top level config.bitrate will describe the base layer. we can revisit bitrate for other layers later, but it seems complicated enough that we may not want it.

[chcunningham]

triage note: marking 'extension'. As noted above, current proposals (to add metadata) are extensions. If we later pursue the ability to control bitrate per layer, this too would be an extension.