CRF OR 2Pass.

[Don_Genaro]

Hi, I was wondering...

If a video encoded with CRF 21 has a final target size of, say, 600 mb, then, ¿if I encode the same video with two pass specific bitrate to achieve the same size (600 mb), wich one would have better quality?

Does two pass yields better quality at a given size than CRF, or it is the same, meaning two pass is kind of obsolete??

Also, it is safe to say that any divx or xvid source can be encoded to 80% of it´s size with x264 without any quality loss?

I´m using Avinapitc to DRF analizye the resulting h264 streams, and I´m confused with some results:

I use the megui crf profile to encode a certain video (CRF 21), then I analize the result and it has an average DRF of about 22, but then I modify the megui profile to have it making more exahustive video analisys (--no-fast-pskip, --subme 7, --me umh) and the resulting stream (at the same CRF 21) has less size and has an average DRF of about 24. What does this means? That with more analisys you get a smaller file with worse quality or a smaller file with the same quality but with a lower DRF average? And why does the megui profile CRF uses just the least pixel refinement posible setting (1) ?

[Dark Shikari]

Originally Posted by Don_Genaro

Hi, I was wondering...

If a video encoded with CRF 21 has a final target size of, say, 600 mb, then, ¿if I encode the same video with two pass specific bitrate to achieve the same size (600 mb), wich one would have better quality?

In most cases there would be hardly any measurable difference in quality.

Originally Posted by Don_Genaro

Does two pass yields better quality at a given size than CRF, or it is the same, meaning two pass is kind of obsolete??

Its basically the same. Twopass isn't obsolete though if you want to specify a specific filesize, such as in filling a DVD/CD.

Originally Posted by Don_Genaro

Also, it is safe to say that any divx or xvid source can be encoded to 80% of it´s size with x264 without any quality loss?

There will always be quality loss; the question is what you will tolerate. If you mean "less additional quality loss than there was encoding from the original source to the Xvid file," you can get well below 80%.

Originally Posted by Don_Genaro

I´m using Avinapitc to DRF analizye the resulting h264 streams, and I´m confused with some results:

I use the megui crf profile to encode a certain video (CRF 21), then I analize the result and it has an average DRF of about 22, but then I modify the megui profile to have it making more exahustive video analisys (--no-fast-pskip, --subme 7, --me umh) and the resulting stream (at the same CRF 21) has less size and has an average DRF of about 24. What does this means? That with more analisys you get a smaller file with worse quality or a smaller file with the same quality but with a lower DRF average? And why does the megui profile CRF uses just the least pixel refinement posible setting (1) ?

The profile uses subme 1 because its stupid, probably. Subme 1 is just atrocious.

Quantizers can vary due to the encoding settings because depending on the source, enabling RDO/similar will either raise or lower the bit cost of encoding, resulting in different quantizer choices to be made.

[Don_Genaro]

Thanks Dark!

I was reaching the same conclution: CRF and twopass are about the same in quality. The DRF analisys of both the two pass and the crf encoding shows almost the same average drf. However in the two pass encode the quantitizers are very normally distributed, and on the crf the quantitizers tend to concetrate on one extreme. ¿Could this normal distribution mean a wiser bit usage than on crf?

[Dark Shikari]

Originally Posted by Don_Genaro

Thanks Dark!

I was reaching the same conclution: CRF and twopass are about the same in quality. The DRF analisys of both the two pass and the crf encoding shows almost the same average drf. However in the two pass encode the quantitizers are very normally distributed, and on the crf the quantitizers tend to concetrate on one extreme. ¿Could this normal distribution mean a wiser bit usage than on crf?

DRF is not a good measure of quality. "DRF" is actually the quantizer. What CRF does is pick a quantizer based on how complex the frame is--more complex frames get higher quantizers while less complex frames get lower quantizers. This generally proves to be pretty effective.

[Sharktooth]

the profile uses subme 1 coz ppl asked for faster 1pass CRF profile. you're free to modify it.