MPEG2 Encoder Comparisons?

[kiyookasan]

Does it have a 2-pass option?
And how are those encoders in comparison?

My next comment will be considered heretical by many. However, it is a historical fact that 2-pass (full-length-passes) are relics from the days when MPEG-2 encoders were large hardware devices that controlled BETACAMSP decks using RS-serial protocol. In those days, the source material was not all in digital format, and so to do analysis, the encoder had to rewind the source deck and play the material through from start to finish.

I can tell you for a fact, that when the going gets tough for an encoder (when the GOP bitrate is peaking close to the sequence_header.bitrate), the encoder can _at most_ borrow from a bit pool several GOPs ahead or behind without violating the buffering requirements for MPEG. The vbv_buffer is just as important as the sequence_header.bitrate for conformance.

The primary point of VBR encoding is to reduce waste and make files smaller, and in the days of machine controlled tape decks, multi-pass was used to estimate global bitrate allocation to make sure the compression would fit into the media. Case in point: even today, the boards are filled with people using 2-pass VBR and their movies still don't fit into the DVD-R they are targeting and lots of people are having troubles with building and muxing.

Pro Hollywood DVDs are produced on multi-pass interactive systems. Part of the bitrate allocation is determined by the machine, part by the operator. The mastering process (audio,video whatever) is always about trade offs, and I think it takes a pair of eyes to make good decisions. And quantitative tools are helpful as well.

Gen Kiyooka
Digigami

[kiyookasan]

Does it have a 2-pass option?

I can tell you for a fact, that when the going gets tough for an encoder, the encoder can _at most_ borrow from a bit pool several GOPs ahead or behind without violating the buffering requirements for MPEG.

www.scientificatlanta.com/products/customers/white-papers/G1481A_EncodingTechPaper.pdf

Here you go, they're using the '2-pass' terminology, but strictly speaking, they simply have motion-compensated filtering with forward feedback just in front of the encoder. It runs only 1 second in front of the encoder, which if it were DVD encoding, would be about 2 GOPs.

Unfortunately, in this whitepaper, they are pounding the 'I-frame injection' drum, which if you ask me is like saying the sun revolves around the earth.

If you download and analyze the streams SERBIAN's fav hardware encoder (look upstream on the thread) is emitting, they are not doing I-frame insertion and their stuff looks fantastic. I wrote this up in my blog, analyzing the frame size and quality at places where B-frames are found on cut scenes. The entry is a couple pages down.
http://www.digigami.com/megapeg/blog/.

Gen
Digigami

[kiyookasan]

i tried Digigami demo == pile of crap.
macroblocks and banding gradients all over the place... save your money and time.

I thought I'd run a sample to illustrate what is possible for SVCD with MegaPEG.X Pro. I used mpeg2decX to convert LXG into a QuickTime file, and then ran some short tests. I made some small changes to our built-in template. I ran it on the maximum quality setting, which is roughly equivalent to a hardware encoder, which is to say, it uses exhaustive half-pel search. Encoding time fluctuated between 6:1 and 5:1 (ie. the total encode took about 10 hrs, G5 2Ghz). While the encoder was running I was also watching DVDs, browsing the web, etc. NO noticeable change in the response time of the machine.

Here's the bitrate and compression profile for the SVCD. The average bitrate was about 1.5 mbits, well below the SVCD max of 2.6.
http://www.digigami.com/megapeg/screen.php?screen=LXG_svcd
The SVCD file size is about 955 MB.

And here is the bitrate profile for the original DVD.
http://www.digigami.com/megapeg/screen.php?screen=LXG_dvd

I'd be willing to pony up free copies of our pro bundle package for anyone who wants to do some independent tests and write up a tutorial. And if the byproduct is some recommendations that improve our presets, so much the better. If you're interested send me a msg here, or for a bit quicker response mailto:gen@digigami.com.

Gen Kiyooka
Digigami

[geezerbuttz]

HEY..
I e-mailed you about testing the app , but you dont reply .. what gives?

GB

[kiyookasan]

HEY..
I e-mailed you about testing the app , but you dont reply .. what gives?
GB

He's right - I didn't see it. MacMail spam filter probably took it down. I've had the same email address since '95, so I get mammoth amounts of it. Spam that is.

So if anyone else is interested, use the videohelp message system if you're interested.

Gen