Number of Passes?

[willlangford]

hey,
i have always used the original 2 pass. i am looking at doing the nth pass but what i dont understand is the high-motion or low motion. should i do a pass at each or what?? and what kind of quality boost does it give me.

thanks so much,

Will

[bond]

please always use the search function before posting!!!

-> for example look here (http://forum.doom9.org/showthread.php?s=&threadid=52163)

[willlangford]

Originally posted by bond
please always use the search function before posting!!!

-> for example look here (http://forum.doom9.org/showthread.php?s=&threadid=52163)

thanks for that. and i did search and look at that thread. but i didnt understand it. im a bigtime newbie at this stuff. i am just wondering waht the different motions are. basicly a quick and dirty rundown of the nth pass crap.

thanks,

Will

[bond]

high motion are fast scenes (like action)
low motion are still scenes

with the bitrate modulation slider you can adjust the codec in a way so that it will produce more details in high motion scenes OR more in low motion scenes

now the human eye cant really see much details in high motion scenes so imo it is a good idea to put more detail into low motion than in high motion (like the original 2 pass already does)

as a rule of thumb you can say:
"always use the same settings in each pass"

-> if you have a fast cpu i would make 3-passes and move the bitrate modulation to 0.2 or even more to low motion
-> if you have a slow cpu i would use the original 2-pass

[willlangford]

okay thanks!! ill try that tonight and see how it works.

thanks,

Will

[rivers]

as willlangford i always used the 2-pass from the 5.02, and there are things still not clear to me about the n-th pass procedure:

-i've done the 'Multipass 1st pass', now assuming i want to do an adjustment to bm, i must do a third pass?

[temporance]

i've done the 'Multipass 1st pass', now assuming i want to do an adjustment to bm, i must do a third pass?If you've done the 'Multipass 1st pass' you have done one pass. You only need to run one more pass to use the modulation. Run one nth pass encode, setting the bm slider to, for example, 0.25. You can then finish, unless you really want to run a third pass.

[kl33per]

Up to four passes should provide significant quality improvements, not just three. Beyond that though there is pretty much no visible improvement.

[bond]

in my tests i didnt notice a real difference between the 3th and the 4th pass (comparing frame by frame with avscompare)...

and i still recommend using the original 2-pass as it produces imo a better (meaning sharper) picture in low motion scenes (although i used 0.25 bitrate modultion in nth-pass)

[kl33per]

in my tests i didnt notice a real difference between the 3th and the 4th pass (comparing frame by frame with avscompare)...
The difference is less drastic than going from 2 to 3 passes, but it is there. Probably only for the hardcore fans though. Yep, bitrate modulation wll definately improve low-motion scenes, but other things (besides adding nth pass) have changed since the old 2-pass that definately warrant you using the new nth pass.

[bond]

Originally posted by kl33per
The difference is less drastic than going from 2 to 3 passes, but it is there. Probably only for the hardcore fans though.naah, in my tests it was like only one or two very, very, very small pixel every frame which were different (and as i said i compared frame by frame with avscompare at 200% zoom or so)

Yep, bitrate modulation wll definately improve low-motion scenesyup, it improves low motion scenes but still not comparable with original 2-pass

[kl33per]

naah, in my tests it was like only one or two very, very, very small pixel every frame which were different (and as i said i compared frame by frame with avscompare at 200% zoom or so)
It's probably relative to the source to, but as I said the difference in quality is no where near as significant as the jump from 2-pas nth pass to 3. In regards to the Orig. 2-pass vs. nth pass, I'm still leaning towards nth pass. I haven't personally noticed any distinct quality advantages in the way of Orig 2-pass, and I think nth pass gives a more overall balnced quality level, while Orig 2-pass has a tendency to change the average quality level between sequences of film. If there are any significant quality advantageous. in Orig 2-pass, I think there'll be pretty much squashed when Manihi goes final.

[bond]

Originally posted by kl33per
I'm still leaning towards nth passhm, i also think that a good codec should be able to do the work in two passes which is technically possible (dont like this 3 or even 4 pass idea, not everybody has 2000mhz to run an encode the double time if the same quality should be also able to reach with 2 passes)

[Acaila]

Originally posted by bond
i also think that a good codec should be able to do the work in two passes which is technically possibleAnd DivX already does a very good job within 2 passes. Whether or not "very good" is good enough for you is a personal choice, but at least we HAVE a choice. It's just that theoretically 3 passes will always be better than 2, because with 2 the codec still has a lot more guessing (how large a frame will turn out with a given quantizer won't be known until it is actually encoded with it) to perform than with 3.

Taking both encoding time and quality gain into account it is my opinion that 3 passes should be more than enough for most people.

[SeeMoreDigital]

I've actually just posted a response to this on the DivX Manihi thread but I think it's relevant here too.

For me, the whole question of allowing the 'user' to adjust the 'Bitrate Modulation' is a strange one.

Surley DivX could create an application that can automatically formulate the correct setting(s), during the first pass, and write the necessary information into the 'log/MV file' ready for the second pass?

I don't know about anybody else but I've never seen such a function in any other codec manufacturers software.

In my opinion, this is DivX's achilles heal, especially when encoding at low bitrates!

[temporance]

I think it's nice that the user has control of modulation, though it would be better to have it default to 0.25 instead of 0.0. And it might be nice to be able to use > 0.25.

The setting of this control doesn't depend on the video being encoded, it depends on the preference of the person doing the encoding. Does he want perfect block-free high motion scenes, or perfect block-free low motion scenes? Where does he want the codec to put the blocks, if it has to?

[SeeMoreDigital]

Hi temporance,

Originally posted by temporance
I think it's nice that the user has control of modulation // The setting of this control doesn't depend on the video being encoded, it depends on the preference of the person doing the encoding. Does he want perfect block-free high motion scenes, or perfect block-free low motion scenes? Where does he want the codec to put the blocks, if it has to?

I know what you mean, but a 'user' has very limited control over this function. And it seems a little daft to me that the same 'user' setting should remain the same for the entire duration of the encode.

In my opinion the 'bitrate modulation' should not be set in stone. It should be constantly changing, in unison, with the speed of the source image/action.

And afterall a PC is the ideal tool to quickly analyze and calculate such changes and make the necessary alterations to shuffle the bitrate modulation/blocks accordingly.

[temporance]

In my opinion the 'bitrate modulation' should not be set in stone. It should be constantly changing, in unison, with the speed of the source image/action. I think we have different understandings of what this setting does. IAUI, it sets a bias between low and high motion - something that must remain constant for the whole pass. There is a parameter (e.g. quantizer) that is constantly changing as you describe. Modulation just controls how that parameter changes according to motion etc.

Edit: I'll use an analogy - you setup brightness, contrast etc. on your TV once according to your preference and then don't change it. You don't expect to be adjusting brightness for every different movie you see or every scene or frame even - and you can't buy TV's that automatically do this. Modulation is just another setting like these.

[dTb]

I've seem myself that even on the 4th or 5th pass the codec is refining the quant distribution for frames. The difficult thing is visually perceiving any change particularly when you don't know which frames have changed, even if you do what does it really tell you to observe a particular frame has changed from 2 to 3 say?
Three passes is probably the best compromise between time and quality but there's nothing wrong with a couple extra passes if you've got nothing better to do.

As for the bitrate modulation, like temporance has said it's largely to do with your preferences. From what I understand the codec is still doing largely the same thing, you just have a bit more control directing where the bits go.
These days I notice macroblocking in high motion so I've tended to use a value of -0.02 to slightly improve things at the expense of low motion. Manihi etc. has improved things in the high motion area so I'm beginning to think this won't be necessary anymore and maybe a slight bias towards low motion will suit me better.

Some comments by Gej taken from this thread
http://forums.divx.com/viewtopic.php?topic=47602&forum=23

"Each encoding passes after the 2nd one will reduce the quantizer distribution and will try use lower quantizer when possible, it does this in order to get more consistent quality and to avoid lower quality frames (or sudden quality drop)

The human eye is more sensible to low motion, the original 2pass mode was biased toward low motion, most of the time people perceive the quality to be superior when the bitrate modulation parameter is set a positive value (0.1 to 0.2)."

[kl33per]

Originally posted by bond
hm, i also think that a good codec should be able to do the work in two passes which is technically possible (dont like this 3 or even 4 pass idea, not everybody has 2000mhz to run an encode the double time if the same quality should be also able to reach with 2 passes)
You're right, not everybody does have 2GHz machines to do encodes. I'm testing 4-pass Manihi on my PII-350 cause thats the best I've got (and yes it's really slow). As has been said by a dTb and others, nth passes aren't a bad idea and have potential to improve quality drastically (particularly from 2-passes to 3-passes). MPEG2 has used Multi-passes for years and this has really helped it with encoding.

[bond]

ok, guys you convinced me about nth-pass (joke :p )

no serious, i will test it again when the final divx5.0.6 (or even divx6 ?) comes out, other users experiences sound very reasonable...
hm, cant wait anymore for this release :D