Bits and quantizers

[cool9]

I made two samples of a video with sixofnine CQM with Bframes (2/1.5/1.0), q=3@1st pass:

1. Min. quantizers=3. Result: high ave. bitrate/filesize (1.5% lower than target, Ave. bitrate=1644)
2. Min. quantizers=4. Result: low ave. bitrate/filesize (16% lower than target, Ave. bitrate=1390)

Quality is almost indistinguishable between the two.
I wanted to know what occurs when you raise the minimum quantizer(s) resulting in a lower ave. bitrate? Do certain areas get robbed of bits or does the entire video get robbed of bits?

[drcl]

as the target is missed significantly on q4, it seems like quantizers that would be lower than your minimum are clampled to your minimum. there are alot more q3 than q2 in an unlimited encode.


ie. normally would be q2 becomes 3 or 4 repectively.

[Teegedeck]

Such quantizer restrictions don't make sense if you don't hit the aimed-at filesize. Do a filesize prediction with Enc before doing the normal encode if you have to restrict quantizers. I'd advise to use min. quant=2 if you don't want to make filesize predictions.

(See the 'XviD presets' thread for a hopefully reasonable use of quantizer restrictions.)

If you get a lower filesize than aimed for then this of course means quality will be worse. If you don't use curve-compression the degradation should be quite evenly distributed over all kinds of frames.

The difference is almost undetectable in this case because the difference between SixOfNine at quantizer 3 and quantizer 4 is much smaller than the difference between standard MPEG quantizer 2 and 3.

[cool9]

I read thru your Enc instructions. Very easy to use and nice layout but I've never had a lot of luck with it. Sometimes I'll get 47% or 105% or 0.05%. Doesn't make sense. Now I just make some samples using Jaw1 or MPG or Shark LR/HR and take an educated guess at what I should use.

Haven't had luck with curve compression, either. That really throws a monkey wrench into your target bitrate/filesize.

I don't see in what case(s) you would use min. quantizers=4-5? You must really have a good video.

Thanks.

[Teegedeck]

Apparently you're not using Enc in quite the right way. For a filesize prediction check the size prediction option!

Then configure XviD to a constant quantizer and you should get results in Bytes. Devide by 1.048.576 to see the result in MB (well, MiB, to be precise). If you entered your minimum quantizer and got a result smaller than your aimed-at final filesize then you'd know that you'd get undersizing in the 2nd pass.

And you're right; don't use curve-compression if you're unsure what it does!

I don't see in what case(s) you would use min. quantizers=4-5? You must really have a good video.No, that doesn't have anything to do with it. (You've got it completely the wrong way round BTW; higher quantizer=smaller filesize and worse quality.) You don't use any quantizer restrictions except minimum quantizer = 2 (and that does have a very simple reason).

Quantizer restrictions apart from min=2 should not be used. I only use them in the 'XviD presets' in order to prevent people from using a preset in ways they were'nt designed to be used in.