Gordian Knot 0.26 DivX5 Comp test and Pro features

[Ewi]

Hi!

At the moment I am trying Gknot 0.26 with DivX5 under WinXP.

But the Comp test of it is irritating me, cause:

1) Where can I set what pro features should be used for my comp test
2) Do the preferences for Divx5 Pro features in the Options Dialog of GKnot affect the comp test?
3) Doom9 says in his new guide that a good comp test value ist 60-70% and that I would get this at around 0.17 for Bits/Pixel*Frame. But with the movie I'm trying (Claws of Steel) I get 60% at around 0.47. Theoretically this could happen but its an elder movie and hence VERY blurry and hence of good compressibility (I would think).

Is this GKnot Comp test feature broken??

Thank you.

[Mac Sidewinder]

If you look at the tabs at the top of gnot you will see one to the right (misc i thing) anyway click on each and on one you will see the default settings for the divx 5 pro features. This will determine which mpeg4 tools are used in the comp test.

Mac

[Ewi]

Ok!

This explains questions number 1 & 2.
Is this for sure or do you only think that GKnot works in that way.

So question 3 remains there. Is it possible that I get 60% at a 0.47 value for Bits/pixel*frame.

Thank you.

[Doom9]

ewi: mac is right.. that's who it works (even asked thewef about it)

as for the other.. the value is pretty darned high though I wouldn't rule out that it's possible.. would be useful if you gave us some info about the movie, the used parameters.. that kind of stuff.
basically it can very well happen that you get outrageous values in the comp check.. once that feature was introduces I was using it briefly until I found some cases where the values just didn't make sense. Since then I only go by bits/pixel values.. they've served me pretty well in the past.. better than the comp check.

[Ewi]

Hi!

Oh, Thank you! That's what I wanted to hear.

The film is a eastern with some heavy action scenes but nothing that
could cause such a need for a high bits/pixel*frame value. Especially the film is really blurry.(It's german,PAL)

I did a test encode at 0.183 bits/pixel*frame (b-frame, GMC, Psycho normal, no deinterlace, simple resize filter[I checked the green line issue in preview before coding]). I looks not so good, but OK (not blocky but I don't like that ringing effect an edges and for my eyes that is to heavy). Using other resize filters seem to debase the bad comp test results (at around 0.20 bits/pixel*frame: simple-31%, neutral-27%)

Tomorrow I will give it a second try at around 0.20/0.21 bits/pixel*frame. I will post my subjective result here, but I expect good result although the comp test gave me 31%.

But I thank you, Doom9; now I can go to sleep without feeling totally stupid.

[Ewi]

At 0.21 bits/pixel*frame the comp test with the options above used gave me 33.5%.

Quality is so good now, that I think could use 0.20 bits/pixel*frame.

Conclusion: The comp test did not work for me with this movie. I will try it again with the next one.

[Adrian]

I know wishlists are frowned upon, but TheWef, in the next update can you please consider including source pre-processing as one of the DivXPro options? I used to use VDub's TemporalSmoother, but now prefer doing it within the codec (it's faster). Purists like Doom9 don't like these filters, but many of us do. I just want to add that I love your program & it's very useful- thanks for all the effort you've put into it. Doom9, I love your site- it's easily the best DivX encoding site. I hope I haven't transgressed your forum rules. :)

[ogo]

Adrian try vdubconf 0.2 it was just made for that :-)

[Adrian]

Ah thanks, I'll give it a go.

[theReal]

60% at 0.47, now that sounds like a DV homevideo (the last one I did was more like 50% at 0.47, though...).

Thanks Mac for the hint on the default Divx5 parameters. I had overlooked these completely and I also wasn't sure if I could activate the Pro features in a compress test at all .