Question about normal (not alternative) curve compression

[AgeOfPanic]

I am working on a movie now. It will be three CD's with DTS sound. My first encode, based on the settings in the Snowbeaches guide, looks pretty good, but there seems to be some blocking in the low bitrate scenes. I read a lot of topics about the issue of alternative curve system and the consensus seems to be not to use it, unless you want to experiment.
The guide also mentions the normal curve compression, which you can use to redirect some bits to low-bitrate settings. Is this option any good or does it have the same disadvantages as the alternative curve system?
If it is good, what are normal values for these options? And also which should I use: The "High bitrate scenes" or the "Low bitrate scenes" settings.

[Nibor]

Hi!

You can use the 'Low bitrate scenes' setting to help with these scenes.
The normal values are 0. But in your case i suggest to try a value of 10 for the 'Low bitrate scenes'... Try different settings to see what happens!
I often use this setting because it helps the 'bad' scenes and you won't notice much difference in the 'good' scenes :)
Also enable 'Playback with bias' because it also helps the small frames.

Good luck :p

..:: Nibor ::..

[Prettz]

Originally posted by Nibor
Also enable 'Playback with bias' because it also helps the small frames.

I was under the impression that 'Payback proportionally' gave more extra bits to undersized frames, and 'Payback with bias' gave the same amount of extra bits to every single frame regardless of its size.

[Assault]

Payback proportionally distributes the amount of bits proportinally to the size of the frames. So it helps big frames because big frames get more bits than small frames. Payback with bias gives all frames the same amount of bits -> small ones get more bits than when you select proportionally. So payback with bias helps low bitrate frames.

Assault

[Prettz]

Originally posted by Assault
Payback proportionally distributes the amount of bits proportinally to the size of the frames. So it helps big frames because big frames get more bits than small frames. Payback with bias gives all frames the same amount of bits -> small ones get more bits than when you select proportionally. So payback with bias helps low bitrate frames.

Assault
I'm pretty sure that 'Payback proportionally' works the opposite of what you describe; frames large than the average size get the fewest extra bits and frames smaller than the average size get the most extra bits.

[Assault]

@ Prettz

No, you're wrong. Look into the XviD FAQ, it's written there:
Bias bitrate payback merely adds reserve bytes / payback delay to a frame regardless of it's size whereas proportional adds fewer bytes to smaller frames and more bytes to larger frames.
Btw, it wouldn't be proportional if it distributed more bits to smaller frames but it would be inverse proportional(is that the way you call it in english? :rolleyes: ). ;)

Assault

[Prettz]

Originally posted by Assault
Bias bitrate payback merely adds reserve bytes / payback delay to a frame regardless of it's size whereas proportional adds fewer bytes to smaller frames and more bytes to larger frames.

Regarding Bias, that's exactly what I said. Regarding Proportional, that makes absolutely no sense. I guess I'm remembering the 'Payback Proportionally' from Nandub.

[manono]

Hi-

Regarding Bias, that's exactly what I said.

Yes it is. But as I understand what you've written, you then drew the wrong conclusion.

Regarding Proportional, that makes absolutely no sense.

I don't know. Makes perfect sense to me.

Look at it this way. Frame #1 is made up of half the bits, or is half the size, as Frame #2. Payback with Bias gives them each the same number of spare bits. Therefore Frame #1 is "helped" twice as much, relative to its size. Therefore Payback with Bias helps small-framed static scenes more than large, complex scenes. The Bias is towards less complex, relatively static scenes. With Payback Proportionally, Frame #2 gets twice the spare bits as Frame #1. It's twice the size to begin with. Therefore they are both "helped" equally. They are helped proportional to their frame sizes. Using Payback Proportionally, both static and complex action scenes are improved by the same percentage. That's how I understand it anyway.

[LigH]

Let's try some much-too-simple maths to tell absolute from relative gain by comparing the relation between the frame sizes after to those before applying the payback:

There may be 4 units "p" of payback bitrate to spend, and Frame 2 may be 3 times as big as Frame 1 = "f":

a) Biased (rather same absolute gain):

Frame 1 (1 size big) gets 2 units of payback (1f+2p), Frame 2 (3 sizes big) gets 2 units of payback too (3f+2p).

The quotient of (1f+2p)/1f is bigger than the quotient of (3f+2p)/3f.

Biased payback more or less adds a constant everywhere. If you want to draw a graph: Lift the ramping line equally upwards.

Incorrectness here: Most likely, the gain won't be really a constant, but indeed still depend a bit on the frame sizes.

b) Proportional (same relative gain):

Frame 1 (1 size big) gets 1 unit of payback (1f+1p), Frame 2 (3 sizes big) gets 3 units of payback (3f+3p).

The quotient of (1f+1p)/1f is as big as the quotient of (3f+3p)/3f.

Proportional payback multiplies by a factor. If you want to draw a graph: Let the ramp raise steeper.

[Prettz]

Originally posted by manono
Look at it this way. Frame #1 is made up of half the bits, or is half the size, as Frame #2. Payback with Bias gives them each the same number of spare bits. Therefore Frame #1 is "helped" twice as much, relative to its size. Therefore Payback with Bias helps small-framed static scenes more than large, complex scenes. The Bias is towards less complex, relatively static scenes.
I see, I never thought about it that way before. I just thought that because it used a constant bias for all frames I couldn't see how it could be useful, but it makes sense now.


The proportional method that I was talking about (which is the method that Nandub uses, isn't it???) would have acted to smooth out the bitrate curve towards a smooth line. Why isn't this method also an option? Is it just not as good for some reason?


edit: fixed a typo

[visigotik]

Would scenes with fog/smoke/etc be considered high or low bitrate? and wich playback settings would more appropriate for them?

[LigH]

Don't take this opinion for sure - but I would guess that fog scenes are "unfortunately considered as low detail/motion scene", because fog has unsteady motion, and low detail (and therefore not enough detail to detect the subtle motions exactly enough).