On Mon, 18 Jun 2007 19:58:33 JST [EMAIL PROTECTED] (Andrew Church) wrote: > >Thank you. Do you know if the first pass data is used for more than > >calculating a bitrate for each set of frames? > > Yes, it is--or to be more accurate, it doesn't calculate the final bitrate > at all during the first pass (though it may calculate bitrate hints to use > during the second pass). The details depend on the particular codec, but > what's usually recorded is complexity information, i.e. a "summary" of each > frame from the encoder's point of view. The encoder then uses that > information in the second pass to guess at how many bits will be needed by > upcoming frames; thus, for example, if the encoder knows that a lot of > simple frames are coming up, it can allocate more bits to preceding frames > (thus improving their quality) without worrying that it'll hit the bitrate > limit later on. > > This, by the way, is why you'll most likely run into problems if you give > different video streams to the two passes. If the complexity analysis from > the first pass says, for example, that frames 100-150 are fairly simple, > the encoder may choose to use up most of its available bitrate in frames > 50-100. But if you apply a filter that makes frames 100-150 harder to > encode, then the encoder won't have enough bits for those frames, and > you'll end up with poor picture quality in that segment. Conversely, if > the analysis says that frames 100-150 are complex, the encoder may choose a > lower picture quality in frames 50-100 to compensate; and even if it turns > out that your filter made frames 100-150 easier to encode, the encoder > can't go back and add more bits to the earlier frames, since they're > already encoded. > > --Andrew Church > [EMAIL PROTECTED] > http://achurch.org/ >
I must say that the detail of information you've given me is great. Thank you so much for that. I'm authoring a dvd encoding frontend and I'll apply your recommendation as default and add an option for faster, less accurate multipass encodes. Regards, Jeff.
