divx 5.2 Quantization type

[sheyac]

Hello

This release came just in time as I'm about to do a Matrix Revolutions encode. Quick question though. Xvid's strong point seems to be in it's ability to preserve detail, is this due to the usage of the MPEG quantization type, instead of the H.263 used by DivX?

Anyway, as I understand I'll be able to pick in 5.2 which one to use, and I'm currently going to be using a bitrate of about 1440 kbits/s. Which quantization type do you recomend i use given the movie?

Will MPEG always yield a sharper, more detailed picture than H.263, or is this largely dependant on the source, as well as bitrate?

Thanks

Sheyac

[bond]

Originally posted by sheyac
Will MPEG always yield a sharper, more detailed picture than H.263, or is this largely dependant on the source, as well as bitrate?as you already know yourself the new mpeg mode in divx5 is the same technology as used in xvid (defined in good old mpeg-4), so you will find the answer to your question by using search (for example in the xvid forum), as what goes for xvid's mpeg mode also goes for divx5's (and any other mpeg-4 compliant codec) :)

[DigitAl56K]

We strongly recommend H.263. MPEG-2 was a nice add, but in most of the content I have tested, H.263 looks much better.

[sheyac]

@DigitAl56K

Is this using high bitrates (e.g. doom9 guide for xvid says that going for 2 cd's MPEG is better)?

Thanks

Sheyac

[DigitAl56K]

Try a subjective test - encode a few sample clips, turn of post-processing in the decoder, and judge for yourself.

I personally had to use extremely high bitrates before MPEG-2 quantization came on par with H.263.

[Soulhunter]

IMO DivX's MPEG QM behaves different than XviD's MPEG QM !!!


What I can tell atm...

DivX's with MPEG QM produces less details but more blocking than XviD with MPEG QM !!!

Think I gonna invest some more time for further testing... ;)


Bye

[Gannjunior]

I'm doing a complete comparsion test between 5.2 and the other codecs.
I've met 2 stranges things.At this moment I'm doing the test at Q=2. first I'm not able to achieve a logical SSIM value if I check multiple bf: i obtatin a 3.71%, but the quality and the size are almost the same as the test with single classic bf....
Second strange problem: the MPEG-2 matrix seems to do the opposite work it should do: I'm doing the test at Q=2, and I've also tested at 1.5 and 1, but I get also a size drastically smaller than h263 in the same condition of about the 60%.An the SSIM values drop obviously drastically.Visual quality confirm this, even if is not too much bad to see the mpeg2 test.( it risults smooth).Later,I'll do test in 2pass at 25-50 and 75% of saturation, and probably, MPEG-2 will give us interesting result....
Anyone confirm my " problem"?

ciao!!

[Manao]

I've met 2 stranges things.At this moment I'm doing the test at Q=2. first I'm not able to achieve a logical SSIM value if I check multiple bf: i obtatin a 3.71%, but the quality and the size are almost the same as the test with single classic bf....When you're doing your comparison, the clips aren't in sync, hence the bad SSIM. B-frames unpacked create an avi clip which has a lag of 1 or two frames.

[Gannjunior]

Hi Manao,
the first thing I've done when i've seen the strange result it has been to check sync between source and encoded file in viruladubmod vertical display, and they were sync..:( now i'm going on with my tests and every time i check SSIM in encoded file by multiple bf, i obtain always the same value: 3.71% and it doesn't matter the setting i've choosen...
The others test instead are coerent....

ciao!

[len0x]

Originally posted by Gannjunior
Anyone confirm my " problem"?


Yes, I have the same issues... I don't think MPEG quant was implemented correctly (see my other thread in this forum).

[len0x]

interestingly enough - looking at registry keys for DIVX 5.2 I can see "Intra quant matrix" and "Inter quant matrix" keys which are filled with zeros. Does this mean there will be support for custom matrixes soon?

[Manao]

len0x : if MPEG quantization type is supported, there is almost nothing to do to support custom quant matrices : you just have to write it into the bitstream. But do they want to add it ?

[Sharktooth]

They will ... or i'll kill'em all... muhahahahah :D

[bond]

why does dxn labell its mpeg mode as "mpeg2"?
afaik the default matrix used in mpeg-2 is different from the one used in mpeg-4 (not sure tough)

[len0x]

I also wonder if I select mpeg matrix and fill in those values in registry with custom xvid matrix if that's gonna make any difference to just mpeg matrix (maybe custom matrix are already supported but there is no GUI to change them.)

[Sharktooth]

Originally posted by bond
why does dxn labell its mpeg mode as "mpeg2"?
afaik the default matrix used in mpeg-2 is different from the one used in mpeg-4 (not sure tough)
Because it is the quantization "type"...

[bond]

Originally posted by len0x
I also wonder if I select mpeg matrix and fill in those values in registry with custom xvid matrix if that's gonna make any difference to just mpeg matrix (maybe custom matrix are already supported but there is no GUI to change them.) hm from what i see, it doesnt insert any values of the default mpeg matrix, so i doubt that the codec would use changed values (correctly)
but this can only be found out with trial and error :D

i tried enabling some options (like data partioning, drop frames aso...) via the registry, but i couldnt get it to work :(

still i wonder why, if these options would actually be working, they are not choosable via the gui (for example via an nero-like "expert mode")?

Originally posted by Sharktooth
Because it is the quantization "type"... no clue what you mean with this?

[Sharktooth]

It's a "Like in MPEG-2" quantization mode.

[bond]

Originally posted by Sharktooth
It's a "Like in MPEG-2" quantization mode. well i dont think so

thats why i posted "afaik the default matrix used in mpeg-2 is different from the one used in mpeg-4" above

[Sharktooth]

It's not about the matrix! it's about the quantization type! :devil:

[bond]

well the quant type is also defined in mpeg-4, so calling it mpeg-2 might be misleading
but well, its not really a big issue...