-
Streaming Media East – Jan Ozer talk on H.264 – my brief notes
Streaming Media East – Jan Ozer talk on H.264- my brief notes
Jan often does extensive direct testing before making observations.
These are points I thought were most interesting.
CABAC vs CABVLC. He nearly always uses CABAC as there’s generally very little difference in decode ability for most computers.
Microsoft Silverlight 3 will be extension agnostic. It will even handle .mov
Currently no difference with extensions and use in Flash but Adobe says they may implement .f4v specific metadata in the future.
H.264 (typical web video settings) is generally easier to decode than VP6 (despite a certain company’s claims to the contrary).
Free cross platform utility provides codec details used in H.264 (and other) encodes
https://mediainfo.sourceforge.net/en
He noted that YouTube is using CABAC and High Profile on its 720p encodes.
Apple’s online video uses CAVLC and Main he suspects probably due to limitations in Compressor.
Interesting article relating
https://www.streaminglearningcenter.com/articles/18/1/YouTube-does-720P-HD-using-H264/Page1.htmlOnce implemented, he believes H.264 Scalable Video Coding will dominate as a standard multi bit rate delivery method but it’s about 2 years away (compared to Adobe Dynamic Streaming, Microsoft Smooth Streaming, Move Network Adaptive Streaming)
He said Dicas is about as good as Main Concept (MC had slight but insignificant edge), Apple’s is way behind. Generally this is most noticeable at lower bit rates). Squeeze has fixed issues with it’s implementation of Main Concept in recent updates.
He mentions H.264 can look good at 0.1 bits per pixel or less. 0.15 would be conservative. Anything more would be unnecessary.
He went over various compression apps (Carbon Coder, Compressor, Squeeze, Episode). He did mention he sets slices to 1 instead of Automatic (1 per CPU) since, although it encodes slower, because of fewer issues when CPUs only examine a given slice (Basically some compressions apps like Episode can distribute parts of an image to different CPUs in a multicore system).
He’ll make the PDF of the presentation available at Streaming Media web site.