Retrieve frame quantisers of encoded DivX/XviD video

[valja]

Hi!

Is there any way to find out from encoded DivX/XviD video, which quantisers were used for encodning frames?

This info is easily available, say, from DivX old style two-pass encoding log file. But is it possible to retrieve the same info from encoded video?

[nixo]

ffdshow --> OSD --> output to file.

Or maybe DivXDRFAnalyzer. Or Avinaptic. Or ffmpeg.

--
Nikolaj

[valja]

Thanx a lot!

I did a small test coding with DivX-5 2-pass and compared logfile with programs' outputs.

ffdshow gives nice list of frames with their quantsers, but b-frame quantisers have doubled value, compared to the DivX-5 2-pass log (6 or 8 vs. 3 or 4 in logfile).

DivX DRF Analyzer did not rekognize b-frames in DivX-5 video, average quantiser was quite close to the DivX-5 logged value.

Avinaptic has nice output, but, as ffdshow, it shows doubled value for b-frames' quantisers.

I do not know reasons of such differences, but now I have the direction, where to dig. :)

[Didée]

Most probably, DivX' logfile does not log the "final" quantizer for B-frames, but the "base" quantizer derived from the surrounding P-frames. The final B-frame quant is Bmultiplicator * basequantizer, with Bmulti defaulting to '2'. (At least, it was like that years ago ... chances are it still is like that.)

In effect, the average quant logged by DivX is also questionable.

[valja]

Most probably, DivX' logfile does not log the "final" quantizer for B-frames, but the "base" quantizer derived from the surrounding P-frames. The final B-frame quant is Bmultiplicator * basequantizer, with Bmulti defaulting to '2'. (At least, it was like that years ago ... chances are it still is like that.)
Thanx for explanation. Actually, I suspected something like this. In XviD such change of B-frame quantisers is clearly visible and configurable.

[valja]

Didée, do you know, if DivX has also some kind of hidden "boost" of I-frames, like Bmulti = 0.7 or something?

I analysed my DivX codings using avinaptic, and noticed, that I-frames have significantly lower average quantiser than P-frames, as example, 2.077 vs 2.838. In the same time, in DivX logfile, I-frames have the same quantisers as P-frames around, no visible "boost".

[Didée]

I would assume that they use something like Keyframe boost - but you better ask someone who is actually using DivX. ;)