Artifacts with divx 5.02 2-pass, but not with 1-pass

[FactorM]

I want to encode Kiki's Delivery Service (Anime), which is NTSC Interlaced. With 1-pass everything looks fine. But after finishing the 1st pass of a 2-pass encoding, i did a little 2nd part sample with 1500 frames. And now there are errors everywhere. i'm using gknot .26b, bitrate is 806, only using b-frames, keyframes at 239.

here is the .avs:

LoadPlugin("E:\dvd\GORDIA~1\mpeg2dec.dll")
LoadPlugin("E:\dvd\GORDIA~1\decomb.dll")
mpeg2source("E:\dvdrip\vts_01.d2v")
Telecide()
Decimate(cycle=5)
crop(21,78,672,339)
BicubicResize(576,320,0,0.5)


i also attached a screenshot.

[manono]

Hi-

The reason is probably that you are encoding a sample from somewhere in the middle, but the log and motion information from the first pass is from the beginning and they don't match up. As far as I know, to test a small part before doing a full encode, you either test from the beginning using the first pass information, or do a 1 pass Quality encode.

Of course, if your 1500 frame test was from the beginning, then that theory goes out the window.

[FactorM]

The tests are from the middle of the movie. And yes if i take a sample from the beginning it works, so you're right. I'm doing the 2nd pass right now, hope it will come out fine.

[Awatef]

of course it will come out fine.
I'm just wondering how you're coming to such pathetic ideas, to do a partial second pass!

[FactorM]

Pathetic or not, but i want to take a look at the quality of the final movie, especially high motions scenes. And don't want to wait hours only for recognizing that the end result looks like crap. And with 1-pass i don't see, how a 2-pass bitrate of 809 look like. btw with divx 4.xx those samples are no problem. but with b-frames and motion vectors these samples are history.

[Awatef]

@ FactorM
what do you think, why there is compressibility tests?! -.-

[FactorM]

I'm sorry for not trusting those numbers. the numbers don't tell me, how e.g. different resize methods look like.

[jggimi]

Just do a 1-pass at the approximate bitrate in the gknot calc. You'll get a "constant" bitrate rather than a variable one, but it should be close enough to show you what it looks like out of the encoder.

I do this quite a bit when I'm working with filters, and want a five-minute encoding instead of a five hour one, for the purpose of evaluating results. I often work with about 30-60 seconds of content for this purpose.

[Beave]

You should be able to encode a part in the middle, if make an .avs file that trims the movie and do both passes on it.

But I would go for compressebility tests after all. So make a test, then encode it and look what's the best looking with what %. I came to a conclusion that around 55% is the best for me. But you should decide for yourself.