Lower 1/2 pass ratio for divx5 encodes = myth?

[OUTPinged_]

Hello.

For my small testing runs i have not noticed any differences in picture quality at same drf's for divx4 and divx5. there were some minor changes here and there but they are clearly visible only in 4x zoom.

The filesize reduction is about the same as we go up in drf's too.

So, quality and filesize dont differ.

An idea behind lower 1/2 pass ratio in divx4 comparing to divx3 was that divx3 was not handling high drf's well (blocks and disortions).

But that is not an issue with divx4 to divx5 move.

That was how i came to conclusion that 1/2 pass ratio should be the same as for divx4 encode.

But there are numbers of people who claim it is safe to lower it for an encode with b-frames (???)

Does b-frames inclusion change things? Seems ridiculous, as they dont have anything to do with drf/framesize or quality.

Who can confirm that lowering 1/2 pass ratio is(not) safe in divx5 encode with b-frames?

[ProfDrMorph]

What is a 1/2 pass ratio? Never heard about this.

B-Frames DO affect the final file size because B-Frames are even smaller than P-Frames at the same quantizer ( or using Nandub's old expression: DRF ).

[theReal]

I guess he means "the ratio of the second pass compared to the first pass", that is the percentage shown in Gknot.

In that case, if the first pass in Divx5 already makes use of b-frames, to aim for lower numbers here would be wrong. If the first pass doesn't make use of b-frames, you should of course aim for lower numbers.

So the question is: Does a divx5 first pass make use of b-frame compression? I'd say yes, but I'm only guessing...

[OUTPinged_]

thereal: yes it does :-(

it still is not known what is divx5 looking at when it sets drf level of b-frame. but in general it looks like 100% quality based 1-pass encode contains drf4 b-frames.

that means, b-frames arent that good. :-(

[Riman]

I encoded three classic Star Trek episodes (grabbed from cable TV) and there wasn't any significant difference in filesize with or without B-frames. In fact, in some cases the file with B-frames was a bit bigger. Can anybody explain what's wrong?

I usualy use these Virtual Dub settings:

Deinterlace (blended)
Dynamic Noise Reduction (4)
Resize (presize bicubic -1)
Temporal Smoother (4)

Codec settings:

384x288 2-pass, slowest, 500 kbps
576x432 2-pass, slowest, 650 kbps
768x576 2-pass, slowest, 1000 kbps

Theory: too much noise in source?

[theReal]

First I'd say: your bitrate is too small! For a noisy tv-source, I'd use at least 1000-1300 for the 576x432 resolution.

Then, Divx5 files with b-frames don't get any smaller when you are using a fixed bitrate setting. This is what you set the bitrate for: you want to achieve a certain size. It's good that this size is reliable. You might notice a quality increase using b-frames, though.

Your files will only get smaller when you're using fixed quantizers (=fixed quality setting)

btw. have you tried precise bicubic resizing in vdub? It looks quite good and is an effecive noise-filter.

[theReal]

Back to the first question, I think I found the problem:That was how i came to conclusion that 1/2 pass ratio should be the same as for divx4 encode.
But there are numbers of people who claim it is safe to lower it for an encode with b-frames (???)

I think those who said that you can aim for lower numbers meant that if you are doing a compression check with Divx4 or Divx3, but you are then going to encode the file with Divx5 and b-frames, you can aim for lower numbers such as 30%-40%.
That would be because the compression check was done using another codec, one that doesn't use b-frames.

[Riman]

theReal, bitrate is really smaller - I think this is what DivX is made for. To keep highest quality at the given bitrate. I don't want burn one 45-50 minute episode on 1 CD, it's waste of place. (Imagine all the StarTrek episodes á CD...)

B-frames... it was my mistake. Thanks for explaining. I watched all the variants I can't find any quality improvement with B-frames. Yes, it is very subjective...

I usually use Dynamic Noise Reduction and precise bicubic -1 Resize (see my original post) = 2 "noise-killing" filters. Out off topic question: can you give me any other suggestion to make TV source compressions better?

[theReal]

try the avisynth TemporalSmoother, I think it's better than any VirtualDub filter.