I’m using Mac’s VideoToolbox to do decoding, and it requires AVCC instead of AnnexB. The difference between the two is simply the header, so any of the following would work:
- Only output single NALU units (or, only output single NALU units at predictable times
- Have some sort of details about where the NALU units are in a stream
- Use AVCC instead of AnnexB
For (1), it seems potentially possible. From a sample session, I found that the only non-single NALU blob was the first output from NVENC. Can I rely on this being the case? Or, is there a fast way without searching through the whole blob to know if this is the case? I’ve got zero intra-refresh, which could help.