a couple of questions

[daniel]

hi there,

i've made some qite good rips of my buffy dvd's, but still i've got a couple of questions:

- sometimes i see movies with a resolution of 512x384 or 576x432 or 640x480; but what is the best choice?
because: once i did 512x384 and bitrate 2000 and the file always was under 310 mb; but with 640x480 i had to set the bitrate to 1000

so: what's the best combination? lower res and higher bitrate or higher res and lower bitrate?

- after merging video and audio in nandub, sometimes (with mp3 yes and ac3 no) the total space has become bigger: 300+40 -> 342
i understand why this happens; but how can i calculate the exact extra space?

- in the new gknot you can choose for a compressibility test
but what is it for?

if you can help me: please do it!

[Acaila]

Hi,

- sometimes i see movies with a resolution of 512x384 or 576x432 or 640x480; but what is the best choice?
because: once i did 512x384 and bitrate 2000 and the file always was under 310 mb; but with 640x480 i had to set the bitrate to 1000

so: what's the best combination? lower res and higher bitrate or higher res and lower bitrate?That's actually the million dollar question. If you use a lower resolution you'll end up with an unsharp video, but completely free of artifacts, if you use a higher resolution you'll end up with a sharp and detailed video but with macroblocks and ringing artifacts.
There is no definate winner, some people like the first, others like the second. To get the best of two world you can use the thing called 'compressibility test' as in your last question.
The compressibility test encodes the movie (a piece of it anyway) at the resolution and other settings you have set. It calculates the quality would get with that resolution and outputs a percentage (the higher the better). You can use this value to adjust the resolution to get an adequate final quality.
A guideline to follow is 60-70%, this will give you excellent quality. So if you end up with a test value of 50% you should decrease the resolution a bit, if you end up with a value of 80% you should increase the resolution a bit. You should first determine which percentage value gives YOU adequate quality though. This will ensure you get a good quality movie at the highest resolution (and thus sharpness/detail) possible.

- after merging video and audio in nandub, sometimes (with mp3 yes and ac3 no) the total space has become bigger: 300+40 -> 342
i understand why this happens; but how can i calculate the exact extra space?Use GKnot, here you can select the audio type and it will calculate the muxing overhead into the final bitrate you should use.

[daniel]

thanks for the clear answers!!

[daniel]

hmm about the compressibility test:

- is it usefull and possible to do a test with 100% of the movie and not the max of 15% in gknot?

- you said between 60% and 70% is giving the best results, but in several guides is mentioned that you must try to keep the value: bits /(pixel * frame) between 0.20 and 0.27

with my dvd's it's impossible to do both: i did a compr test at 15% and only with the resolution 736x400 the percentage is 65; but the other value is 0,136. so a very high resolution and also a very low bits / (pixel * frame) value

if i want 0,2 i must set the resolution to 608x336, but then: 94,5%
so a very high percentage


it's a bit confusing, so what should i do?

[jggimi]

is it usefull and possible to do a test with 100% of the movie ... It's possible, but I wouldn't think very useful. The default test uses "SelectRangeEvery(280,14)" giving you 5% of the entire content, not just one scene. Play your LastCompCheck.avs and you'll see what the SelectRangeEvery filter does....but in several guides is mentioned that you must try to keep the value: bits /(pixel * frame) between 0.20 and 0.27 ... b/p*f is a simple calculation based ONLY on the number of frames and the resolution. It is intended as a guide to use prior to a compressibility test. Once the test has completed, ignore b/p*f, and use the loading factor from your test to set resolution. You should not need to do a second test unless you change filters (type of resizing, noise reduction, et. al.)

[Acaila]

And another thing, NEVER increase the resolution over the original. In your case keep it at 720x.., not 736x..
The 60-70% is just a guideline, if you come out higher be happy it means you'll end up with a higher quality rip :)

The compressibility test value you get from one test is not completely accurate anymore once you change the resolution. So if you get 94% at a width 640 and drag the slider to 720 you need to make a new test for an accurate result.

[jggimi]

...test is not completely accurate anymore once you change the resolution.... I'd thought that was true only if the resizing filter was changed; for example, Neutral Bicubic in the test, Sharp Bicubic after resizing.

Live and learn, I guess.

[daniel]

thank you!

I'd thought that was true only if the resizing filter was changed; for example, Neutral Bicubic in the test, Sharp Bicubic after resizing.

Acaila is right: it does make a difference if you choose another resolution and then redo the test; i tried it myself

[theReal]

Acaila is right: it does make a difference if you choose another resolution and then redo the test; i tried it myself That's right - sometimes the estimation is relatively accurate especially when you only move the slider up one step, but it is never exact.
I just saw it today when I made a test on Kundun with 640x352 and moved the slider to 672x368. The value shown was 69%, but a new test with 672x368 resulted in 73% (all filters were exactly the same).

[ChannelK]

Ok, reffering to the original resolution / bitrate trade off issue, Consider a HEAVILY interlaced PAL dvd at 768*576 (4:3).
I have two thoughts for the particular DVD in question;
1 Encoding at 384*288 effectively deinterlaces the picture (1/2 original resolution) and Comp test results in 79%.

So if you end up with a test value of 50% you should decrease the resolution a bit, if you end up with a value of 80% you should increase the resolution a bit

Thus
2. Increase the resolution a bit but then use a deinterlace filter (fast or normal?).

What, (even generally speaking) would appear to be the best option,
if 1. becuase of the lower res. should neutral or sharp bicubing resizing be used,
2. Because of the higher res. > problem discussed above, use low or neutral bicubic.

Also, does having a b*p/f around 0.35 almost always result in oversize files?

[theReal]

Also, does having a b*p/f around 0.35 almost always result in oversize files?You won't get oversized files, but the quality will eventually be bad if the bitrate is not enough. It depends very much on the source video.

A compress test value of 79% is not necessarily too high. I've been encoding a few DVD's with very high values recently (around 75-80%, one was even 94%) and they all came out perfect in size. I think now that Gknot supports all Divx5 pro features for the compress test, it's not true anymore that over 80% results in undersized files.

However if the resolution is that low, I'd try to use a higher resolution (~512x384) because it will probably look better, even if the comp. test will only give you around 50%.

When you use Fielddeinterlace() (that is "normal" in Gknot, I think), you'll get "about the same" deinterlacing as when resizing to 384x288 (deinterlacing through blending, good compressibility), only in a higher resolution.

To increase the compressibility a little more, you can add convolution3d(0,8,8,8,8,0) or convolution3d(1,8,8,8,8,0) either before or after resizing (depends...), plus maybe a little temporalsmoother, something like (2,1) up to (4,1). Temporalsmoother only if you have a really noisy video, but you can always use convolution3d, it's very subtle (but has a very good effect on compression)
These filters used with neutral bicubic resizing often look better than using bilinear resizing but it results in the same/in better compressibility.

[manono]

Hi-

I might suggest to ChannelK that he run that interlaced PAL DVD through Telecide() without Decimate() to see if it can reconstruct the frames before giving up and just running the whole thing through a deinterlacer or using very low resolution. There's nothing to lose by checking it out for a few minutes in VDub, and a lot to gain in better .avi quality if it works.

[theReal]

That's a good idea, I forgot about that possibility (shame on me! ;))

[ChannelK]

The Real & Manono, thats some of the best feedback ive ever got. thanks!

Manono, im assuming those things you talked about can be found in a virtual dub guide?

[manono]

Hi ChannelK-

If you generate your .avs using GKnot, then it's easy to use Telecide(). Just remove the "#" from in front of the Decomb LoadPlugin line near the top, and again from the Telecide() line further down (making sure you abide by the "Horizontal Crop must be a multiple of 4" rule).

Otherwise, you get the Decomb Filter from Doom9's download page, throw the .dll wherever you keep all the other .dlls, and add those 2 lines to your.avs. You won't find information in a VDub guide, but Doom9 has both a Decomb Guide (http://www.doom9.org/decomb.htm) and a Decomb Parameters Guide (http://www.doom9.org/decomb-params.htm). Also, Decomb itself comes with a very complete Help File. From what I understand (I'm from NTSC land myself), sometimes the PAL DVDs have the fields displaced, and Telecide() can easily put them right again. Quoting from the Help File:

Note that a stream of PAL progressive frames that are shifted by one field will exhibit combing and can be considered a telecined stream for purposes of recovery; Telecide will easily handle this situation.
It doesn't always work (depends on the source), but if it's the right material, it'll work like a charm. Good Luck.

[jggimi]

You may also want to tryTelecide(post=false)which rebuilds progressive frames but does not do any deinterlace postprocessing. If I recall correctly, "Telecide()" will default to post=true.

Results are both subjective and content dependant, so, as Manono recommends, I'd experiment with .avs scripts and previewing in Vdub before doing the encoding.

Good luck!

[ChannelK]

@TheReal, to use convolution3d(1,8,8,8,8,0), do i simply just copy that into the .avs script( like enabling telecide()), also do any more pluggins need to be loaded for this to work? or is removing the # to load decomb.dll the only plugin needed.

To increase the compressibility a little more, you can add convolution3d(0,8,8,8,8,0) or convolution3d(1,8,8,8,8,0) either before or after resizing (depends...),

What does it depend on, or is it simply just trial and error?

Just a couple more questions as well... when does the check box in the final GKnot (encoder) setup page
'IVTC avi> enable frame count'
need to be checked, also, is this about correct,

Telecide for NTSC and PAL (worth experimenting), if using telecide no field deinterlace as telecide() defaults to post deinterlacing.

Purely interlaced PAL > field deinterlace and not telecide.

Decimate() only NTSC 29.97fps.

(... This stuff is starting to make sense)

[jggimi]

... when does the check box in the final GKnot (encoder) setup page...need to be checked...The IVTC process will reduce the framerate from 29.97 to 23.976fps. If you're using Force FILM to IVTC, the framerate will be adjusted automatically by the .d2v project file, and you need do nothing. If you're using any of the .avs script based IVTC processes, either by selecting "IVTC" in the Save & Encode window or by editing the .avs file, you can either manually set the FPS to 23.976 in Gordian Knot's main window, or check that box. I change the framerate....Decimate() only NTSC 29.97fps.I'd use it only on telecined NTSC.