Futurama DVDRips

[starfish]

Hi!

I'm trying to encode the 1st season of Futurama released on dvd, but since I have very little experiency with converting interlaced sources, I got some trouble...

The source is 4:3 DVD, PAL, 25 fps and interlaced. This is the info DVD2AVI gives me. I thought "ok, then i'll just have to deinterlace later on i GKnot". So, my first try was to deinterlace with the 'Fast deinterlace' setting. This removed the interlacing, and everything looks allright (not hard to get good results with cartoons from a dvd-source :), but at scenes with much motion, the movement gets blurred. It seems that the interlacing artifacts from the source just got blurred down, but actually it doesn't make it much nicer...

Is this all deinterlacing does? I tried 'Smart deinterlacing' too, with the same results.

I'll bee thankful for every response :)

And thank you theWef for the best program ever!


starfish

[starfish]

Uhh, I just found out that I forgot to set 'Smart Crop All', so I guess I used a illegal resolution, but this shouldn't give these problems?


starfish

[luke]

as you say it's PAL
and in general, PAL stuff don't need deinterlacing : it's just that dvd2avi is wrong.
if i was you i'll try again without any deinterlace

[starfish]

Hi!

Thank your for the information!

But the source sure is interlaced. If I don't check deinterlace, the output is even worse, whit "scanline" edges when hi motion.

Even the quality when deinterlaced is ok, movement looks kinda crap. I've attached a screenshot where the cam follows Fry, and the crowd are passing by quite fast.

I tried to whatch this particular scene straight off the dvd (allthough on my pcdvd), and it wasn't perfect there either. Is this because the original material is done for cabletv?

Is it any filters that can help me?

thanks in advance,

starfish

[hakko504]

Have you tried Don Graft's (http://sauron.mordor.net/dgraft) Decomb package? Just load the plugin in your .avs and use Telecide(). This should take care of the scanlines.

[starfish]

Well, actually the scanlines isn't the problem, the problem is what the scanlines have become, after using deinterlacing in GKnot...

The decomb filter doesn't work that different from the deinterlacer in GKnot (nandub)? But I'll give it a shot anyway!

Thanks for the reply!

starfish

[neuron2]

>The decomb filter doesn't work that different from the deinterlacer in GKnot (nandub)?

Wrong!

Decomb will first recover the progressive frames, then only deinterlace frames that could not be recovered.

Probably you have a simple PAL one-field phase shift. Try:

Telecide(postprocess=false)

If you still get combing, try:

Telecide(postprocess=true)

Finally, when an area is deinterlaced, Decomb can optionally use line interpolation instead of blending. Since you apparently don't like blending, why don't you use interpolation:

Telecide(blend=false)

[starfish]

Thank you! That was a lot of tips! I'll try them all, especially that about interpolation!

I'll get back to you all when I have tried...

starfish

[starfish]

Hmm... Since I am using GKnot, can I dl the 'Smart Deinterlacing' filter for Virtualdub and use it in Nandub, or is the decomb package different?

Sorry, I'm only a newbie to these things :)

starfish

[UHT]

hi

i have just completed the same project you are undertaking, what i did was to use smart de-interlacing, and in the avs script change the true, true, true to true, false, true

looks fine to me :)

[neuron2]

@starfish

Decomb is VERY different. You are missing my main point. If you have a PAL stream that is merely phase-shifted, then you do not want to deinterlace at all (except possibly at bad edit cuts). That is Decomb's entire raison d'etre. It will recover progressive frames through field matching (which corrects the phase shift) and then examine the resulting frames for combing (if postprocessing is enabled). If they are clean then they are not touched; if not, they are deinterlaced adaptively a la Smart Deinterlacer. Smart Deinterlacer will touch every frame.

While it is true that Smart Deinterlacer may give tolerable results in a PAL phase-shift scenario, it is also true that it will not give you the best results that can be achieved. :)

It is up to you to determine the nature of your input material and then apply exactly the correct processing.

[starfish]

... is in process now, so in a few minutes I'll see how that worked! It took me less time than learning decomb.dll and avisynth, but i'll think I'll check out decomb when I have a little more time than almost 20:00 friday evening :) Oooops, have to buy beer within 20 mins!

I'll post if it went ok or not!

Thanks for all help till now!

starfish

[starfish]

i have just completed the same project you are undertaking, what i did was to use smart de-interlacing, and in the avs script change the true, true, true to true, false, true

I tried it, and it really look awesome now! At least compared to how it was... Have started to read through decomb's helpfile, and it really look interresting. Do you think I can improve the result further by using decomb over 'smart deinterlacing' ?

I'll try it anyway...

starfish

[UHT]

best way to find out is to try it, my rips look fine to me so i dont really fancy doin them again :) might try it on the simpsons box set tho

[starfish]

I may be a total idiot when it comes to things like this, but I cannot get Decomb to work! I'm sure it's only a matter of syntax and the order of the lines in the avs-file, but i've tried so many different ways now...

What I'm trying to do is to get Gordian Knot to use the decomb-filter instead of the 'SmartDeinterlace'-filter. I thougt it only would be to load the filter and change 'SmartDeinterlace' to 'telecide' with the right options, but obviously it's not that simple...

This is one of my .avs-files, edited in GKnot before starting the encode. I have 10-15 different, but they don't differ that much, only a few changes in settings and playing with the AVISource etc.

#
# Created with Gordian Knot
#
# PLUGINS
# get them from http://users.win.be/dividee
LoadPlugin("C:\PROGRA~1\GORDIA~1\mpeg2dec.dll")
LoadPlugin("C:\PROGRA~1\GORDIA~1\decomb.dll")
#
# SOURCE
mpeg2source("E:\!work\futurama\extracts\ep1\vts_01.d2v")
AVISource("E:\!work\futurama\extracts\ep1\vts_01.d2v")
#
# DECOMB
Telecide(postprocess=false,blend=false,debug=true)
#
# CROPPING
crop(11,2,698,572)
#
# RESIZING
BicubicResize(640,480,0,0.5)


I've attached the GKnot log file, if anyone want's to see it...

Please help a dumb newbie out here ;)

starfish

[neuron2]

@starfish

It would really help if you told me *what* wasn't working with Decomb. I never took the graduate course in mind reading. :)

Better yet, make available a short test clip.

[starfish]

Originally posted by neuron2
@starfish

It would really help if you told me *what* wasn't working with Decomb. I never took the graduate course in mind reading. :)


Ohhh, I'm really sorry... That was in my mind, but not in my fingers :)

Ok, what happens is that GKnot invokes Nandub, Nandub launches and quits the exact same moment. GKnot says encoding time 00:00:00, and starts 2nd pass, and 2nd uses exact the same amount of time.

Neither .stats or .ecf's are produced...

Hope I included sufficient info now :)

Thanks!

starfish

[zeronegative]

A little note on the side about the Futurama DVD, you should check the corresponding thread on the Isonews DivX forum...

It basically comes down to this:
Instead of speeding up the 23.976fps source to 25fps (as usually happens with movies and cartoons), the Simpsons and Futurama DVDs are sped up from 23.976 to 24fps only, then have an extra added frame. In other words, every 25th frame is a non-original frame. Instead of duplicating the 24th frame, the 25th frame is frame 24 plus frame 26, which creates some really weird ghosting artifacts.

The best way to rip this DVD is to use a 'decimate' filter, decimating every 25th frame (hm, that wouldn't really count as decimating now would it? :)) and THEN convert it to DivX with your favourite deinterlacing settings.

[meleth]

Originally posted by starfish

I may be a total idiot when it comes to things like this, but I cannot get Decomb to work! I'm sure it's only a matter of syntax and the order of the lines in the avs-file, but i've tried so many different ways now...





What I'm trying to do is to get Gordian Knot to use the decomb-filter instead of the 'SmartDeinterlace'-filter. I thougt it only would be to load the filter and change 'SmartDeinterlace' to 'telecide' with the right options, but obviously it's not that simple...





This is one of my .avs-files, edited in GKnot before starting the encode. I have 10-15 different, but they don't differ that much, only a few changes in settings and playing with the AVISource etc.





#


# Created with Gordian Knot


#


# PLUGINS


# get them from http://users.win.be/dividee


LoadPlugin("C:\PROGRA~1\GORDIA~1\mpeg2dec.dll")


LoadPlugin("C:\PROGRA~1\GORDIA~1\decomb.dll")


#


# SOURCE


mpeg2source("E:\!work\futurama\extracts\ep1\vts_01.d2v")


AVISource("E:\!work\futurama\extracts\ep1\vts_01.d2v")


#


# DECOMB


Telecide(postprocess=false,blend=false,debug=true)


#


# CROPPING


crop(11,2,698,572)


#


# RESIZING


BicubicResize(640,480,0,0.5)








I've attached the GKnot log file, if anyone want's to see it...





Please help a dumb newbie out here ;)





starfish



Have you tried just using

Telecide

with no options? that would probably give you better results. Also you shouldn't type out the options like that Telecide(false,false,true) is how you should put it. And as stated above you should look into using decimate, not only will you get better results but you will also be able to crank up the bitrate since there will be less frames to encode.

[starfish]

Originally posted by meleth

Have you tried just using

Telecide

with no options? that would probably give you better results. Also you shouldn't type out the options like that Telecide(false,false,true) is how you should put it. And as stated above you should look into using decimate, not only will you get better results but you will also be able to crank up the bitrate since there will be less frames to encode.

The decomb readme said that this was a alternate method for entering the options, but I haven't tried the other way, I'll try that as well! And if I get that part to work, I'll try to implement a decimate every 25th frame... But, even when you know that there is a 25th frame thats made out of the 24th and 26th, how do you know that it is the 25th frame from the start? Ehh... well, hope it has to be that way...

I'll try and post the results!

Thank you all!

starfish

[phuntyme]

Before you spend too much time... please heed(sp?) what UHT said. Use smartdeinterlace and change the 3 TRUEs to TRUE,FALSE,TRUE. This was what DIVIDEE used originally in his examples. But changed this in his later release. I kept on using the original arguements (true,false,true) and have found it to be better for what I needed.

Hope this saves you a lot of work ^_^

[neuron2]

@starfish

Your method of entering the options is fine. Unfortunately, the symptoms you describe are issues for Nandub or Gordian Knot and I am not familiar with them. Hopefully someone here can help you. I do know that many people have used Decomb with Gordian Knot successfully.

[starfish]

Originally posted by phuntyme
Before you spend too much time... please heed(sp?) what UHT said. Use smartdeinterlace and change the 3 TRUEs to TRUE,FALSE,TRUE. This was what DIVIDEE used originally in his examples. But changed this in his later release. I kept on using the original arguements (true,false,true) and have found it to be better for what I needed.

Hope this saves you a lot of work ^_^

Tried this after UHT's post, and the results were very good compared with the original, but I wanted to see if decomb was even better, everyone says decomb is the best...

But thanks anyway!

starfish

[UHT]

in this instance i found the "middle false" method yielded better results than decomb, that was for me anyway

[starfish]

Originally posted by UHT
in this instance i found the "middle false" method yielded better results than decomb, that was for me anyway

Hmm, maybe I'll go for that method then... But it still bothers me that I can't get the decomb filter to work! I can't see what I'm doing the wrong way!

Anyway, what exactly is it the middle false does? Changes mode from 'Frame-only differencing' to 'Field-only differencing'? This may explain why its a good solution, because it then would detect that the 25th frame is the 24th and 26th combined... And the 'smart deinterlacing' would correctly detect the interlacing artifacts. Or am I completely lost now?

starfish

[Asmodian]

I am not sure but I believe the middle false changes the way smart deinterlace dectects combing artifacts (It notices differences between adjacent fields rather then adajacent frams). Please correct me if I am wrong.

smart deinterlace in a great filter but if decomb is able to reconstruct the progressive stream you will not have to blur the picture at all!

I really would try very hard to get decomb to work (I was using smart deinterlacing but switched to decomb after encountering some difficult, poorly mastarded, anime - I was very impressed) I think the biggest difference between the two is when there is some mastering error that decomb can correct for. (which it sounds like you might have)

you could try the script below in virtual dub (if all the locations are correct you don't need to use Gordian Knot)

LoadPlugin("C:\PROGRA~1\GORDIA~1\mpeg2dec.dll")
LoadPlugin("C:\PROGRA~1\GORDIA~1\decomb.dll")
mpeg2source("E:\!work\futurama\extracts\ep1\vts_01.d2v")
Telecide(postprocess=false,blend=false)
#crop(11,2,698,572)
#BicubicResize(640,480,0,0.5)

I don't think you need the line (maybe it is the problem?)
AVISource("E:\!work\futurama\extracts\ep1\vts_01.d2v")

I would also suggest not using ! in the source names (I used a DOS environment for a long time so I might just be parinoid but it could be the problem)

Hope you figure it out, I do like decomb better then any other deinterlacing technique but I usually use it on 30fps NTSC anime.

ps. If you are watching your encode on a regular TV then there should be little difference between the two deinterlacing methods - smart deinterlace is good enough that the improvements decomb can make are subtle.

[starfish]

Originally posted by Asmodian
I would also suggest not using ! in the source names (I used a DOS environment for a long time so I might just be parinoid but it could be the problem)

Big thanks to Asmodian! I tried without the ! and voila! It worked! I've used DOS and 8+3 for several years too, but obviously I've got too certain that windows95++ will let me use everything :)

Anyway, decomb is now working for me, but it gives me a small headache... I tried a few settings, and found out that none of them was perfect:

My first try was Telecide(postprocess=false,blend=false), as neuron2 suggested some days ago. That gave me this results:
http://129.240.39.38/deFiner/decomb_post_no_blend_false.jpg
You'll see that it still contains interlacing artifacts, and it blends.

Then I tried with postprocessing, and that gave me this result:
http://129.240.39.38/deFiner/decomb_post_yes_blend_false.jpg
The interlacing artifacts are 100% gone, but now it really blends! It doesn't look that bad "inmotion", but I'm still not quite satisfied :) (Guess I'll never be, but...)

In comparision to UHT's 'true,false,true' smart deinterlacing method, decomb actually is worse. Now you'll be saying that "just go for smart then", but smart deinterlacing introduces a bunch of other artifacts:
http://129.240.39.38/deFiner/true_false_true.jpg
You see that allmost all the interlacing artifacts is gone, but some colors seems to "lag behind". In addition to way too many blocks in the final result:
http://129.240.39.38/deFiner/true_false_true2.jpg

I guess I could run FieldDeinterlace() after SmartDeinterlace, to maually postprocess. No, that was no good. The postprocessing (FieldDeinterlace) only made what was left of artifacts from the smart deinterlacing massive:
http://129.240.39.38/deFiner/true_false_true_post.jpg

Is there a way to remove leftover artifacts and block from smart deint, or a way to force decomb to not blend?

This really makes all the methods too poor for my eye, so what to do?

thanks in advance,
starfish

[neuron2]

@starfish

RTFM!

>Is there a way to force decomb to not blend?

Set the blend parameter to false. I'm very tempted to say "duh!".
The blend parameter is only effective when postprocessing is on, so your Telecide(postprocess=false,blend=false) is silly. You need:

Telecide(postprocess=true,blend=false)

If you still get blends when blend=false, that means your source has blended fields, and there is nothing you can do about it.

>My first try was Telecide(postprocess=false,blend=false), as neuron2
>suggested some days ago.

Read more carefully what I wrote. I never suggested this. I suggested:

Telecide(blend=false)

The postprocessing is enabled by default.

[starfish]

>RTFM!
>
>>Is there a way to force decomb to not blend?
>
>Set the blend parameter to false. I'm very tempted to say "duh!".
>The blend parameter is only effective when postprocessing is on, so >your Telecide(postprocess=false,blend=false) is silly. You need:

I'm sorry I didn't read the manual good enough! I must have missed out that the blend option only is effective when postprocessing...

>Telecide(postprocess=true,blend=false)
>
>If you still get blends when blend=false, that means your source has >blended fields, and there is nothing you can do about it.

This was my second try, and it came out like this:
http://129.240.39.38/deFiner/decomb_post_yes_blend_false.jpg
This was the argument I used: Telecide(postprocess=true,blend=false)
It is possible that I have a blended source, but how could 'smart deinterlace' then manage it? Look at the second last image-link on my previous post.

>Read more carefully what I wrote. I never suggested this. I >suggested:
>
>Telecide(blend=false)
>
>The postprocessing is enabled by default.

I used Telecide(postprocess=true,blend=false), this should give the same result as Telecide(blend=false), shouldn't it?


It may be some more things I have missed out in the manual, if it so I apologize, and ask you to not reply this post. I'll take the hint :)

And I'm still curious about how to get rid of all the artifacts introduced by 'smart deinterlacing'! See preivious post for an example... Anyone?

Thank you all!
starfish

[neuron2]

@starfish

Don't give up so easily! Can you send me a small sample INPUT clip or make it available for download? If SmartDeinterlace can deal with it then so can Decomb, so you may be doing something wrong. If you send me a clip I can tell you exactly what to do! Sorry for my earlier impatience. I am anxious to help you.

[starfish]

@neuron2

Hehe, that's the spirit :)

I've prepared a clip of 72 frames for you. Since the source is DVD the file become really big, so I've made on file with raw data (57mb) and one compressed with lossless huffyuv (33mb).

If you wan't more or less frames, or compressed with another codec, don't hesitate to ask!

http://129.240.39.38/deFiner/pure_avi.avi
http://129.240.39.38/deFiner/huffyuv.avi
ps: i'm fast, so 57mb shouldn't take too long, depends on your connection.

thanks!
starfish

[neuron2]

@starfish

OK. I have had a good look at things. Your clip has a LOT of blended *fields*, but they are limited to the bottom field (except for one). That suggests that Telecide alone should be able to handle the clip. I ran the clip through the VirtualDub Telecide and found it got cleaned up very nicely. So I studied how Decomb's Telecide and VirtualDub's Telecide differ. I made a new version that brings them into agreement and placed it here: http://sauron.mordor.net/dgraft/decomb311.zip. Please watch out if you use positional parameters because there is a new one called "reverse" at the beginning. I suggest using named parameters. So get this version and run the following script:

Telecide(firstlast=true)

It cleans up your test clip very well. If the whole movie has the blended fields limited to the bottom, then this will be your solution. Please let me know.

Now, about deinterlacing. I will address that in a coming post. I want to point out that there is no prima facie reason to think that deinterlacing will be a panacea for these blended fields, but in some cases it can be true. I will explain this in detail and make a recommendation for you. The "color lagging" you see is due to the fact that SmartDeinterlacer's algorithm does not look at chroma. For technical reasons that I will explain, FieldDeinterlace won't currently help you even though it has a chroma option. I plan to modify FieldDeinterlace so that it will do what you expect.

Please tell me how this modified Telecide works with your full movie and then we'll address the deinterlacing approach.

[starfish]

Originally posted by neuron2
http://sauron.mordor.net/dgraft/decomb311.zip. Please watch out if you use positional parameters because there is a new one called "reverse" at the beginning. I suggest using named parameters. So get this version and run the following script:

Telecide(firstlast=true)

It cleans up your test clip very well. If the whole movie has the blended fields limited to the bottom, then this will be your solution. Please let me know.


Ok, I've now tested you new filter version a bit. The results are indeed very much better but it seems to still have some issues.

1. The new version introduced the same 'shit' as 'smart deinterlacing'. I got rid of it by turning on 'Anti-shit', with the threshold value of 16dB, but I think it is a bit strange, did you do any changes that could cause this? Take a look at my previous post to see how it looks, the 'shit' from the new decomb filter looks exactly like this screenshot. It isn't really a problem to compress with anti-shit, but it's more timeconsuming, and I think it's strange that it happens now, and not with decomb 3.0...

2. I ran the filter with 'Telecide(firstlast=true)', and it caused some blending. Not as often as before, but still a few frames. Take a look at these examples:
http://129.240.39.38/deFiner/311_blending1.jpg
http://129.240.39.38/deFiner/311_blending2.jpg
So I tried with 'Telecide(firstlast=true,blend=false)' to force interpolation. This removed the blending, but it didn't interpolate, it just doubled up the previous frame. All the frames with blending from first encoding (wo. blend=false) was replaced with the previous frame. This isn't interpolation, isn't it? Please correct me if I'm wrong.

Except from this, the main issue is 100% resolved. There are no interlacing artifacts left :))

Thanks for a good new filter, and I hope these "new" issues are resolvable!

starfish

[neuron2]

@starfish


1. The new version introduced the same 'shit' as 'smart deinterlacing'. I got rid of it by turning on 'Anti-shit', with the threshold value of 16dB, but I think it is a bit strange, did you do any changes that could cause this? Take a look at my previous post to see how it looks, the 'shit' from the new decomb filter looks exactly like this screenshot. It isn't really a problem to compress with anti-shit, but it's more timeconsuming, and I think it's strange that it happens now, and not with decomb 3.0...


I don't have any idea what "shit" is or what "anti-shit" is that you are talking about. Also, you've posted a lot of screenshots; I can't guess which one you mean. Can you please explain what "shit" is and give me a link to a screenshot? Thank you.

Turning off blend just means that the frame is deinterlaced using data from only one field. If the frame has one blended field and one nonblended field then the result will depend on which field you use.
If it happens that the nonblended field is the same picture as the previous frame, then you will get the doubling. You see, you just can't get a perfect world out of these NTSC/PAL conversions. It's sad, but there you are. You have to choose your worst evil.

[starfish]

@neuron2:
>I don't have any idea what "shit" is or what "anti-shit" is that you
>are talking about. Also, you've posted a lot of screenshots; I can't
>guess which one you mean. Can you please explain what "shit" is and
>give me a link to a screenshot? Thank you.

Here is the correct link:
http://129.240.39.38/deFiner/true_false_true2.jpg
When you've saved the .avs-file in GKnot, you can choose 'antishit' to remove green blocks and other common artifacts. This worked out well for the blocks in my screenshot too, so I'll just use this setting and be happy ;)

>Turning off blend just means that the frame is deinterlaced using
>data from only one field. If the frame has one blended field and one
>nonblended field then the result will depend on which field you use.
>If it happens that the nonblended field is the same picture as the
>previous frame, then you will get the doubling. You see, you just
>can't get a perfect world out of these NTSC/PAL conversions. It's
>sad, but there you are. You have to choose your worst evil.

Ok. I believed that 'interpolating' would calculate a new frame from the previous and next frame. Like morphing the previous frame to the next frame with only one picture between. But I guess that's science-fiction :)

I understand that I don't get the quality higher from this level, so I'll put a smile on my face and start encoding the next 12 episodes. Thank you so much for all help and the updated filter, after all the quality is quite awesome now, mostly because of you filters!

b.r.
starfish

[neuron2]

@starfish

Thank you for the explanation about "shit". Decomb is not generating it but the new field matching is giving you different recontructed frames (which replace the blends you got before). That stuff is in the stream and so you do need your "anti-shit" filter.

As far as morphing frames is concerned, it is theoretically possible but there are a lot of serious practical issues, such as knowing which are the two clean frames to morph between, knowing when to morph, and doing it without reducing the fps to low single digits.

You were apparently fortunate in that most of your blends were confined to one field. I am working on a solution that will work when the blends appear in both fields. As long as there is one good field in a frame it should be recoverable by matching it to a field in an adjacent frame. The new mechanism will be similar to the "desperation mode" of the VirtualDub version of Telecide.

Finally, your feedback has improved the tool and I'll be releasing that version shortly. Thank you.

[MediaMage]

this sucks
how can they sell this dvds that even show the decomb artifacts on a dvd player.

they should fire the people that encoded the pal version from the ntsc one..

@starfish can u help me, as i also try to rip futurama, but i am not THAT experienced.. i just don´t know where and how to use decomb.dll

i wanted to put it in the gordian knot avs file. but this doesnt work...

u speak german?
im sending a pm to u right now

[starfish]

Do everything as usual in GKnot, but before you hit 'Save & Encode' to add the job in the queue, hit 'Edit' and replace the generated script with this:

LoadPlugin("C:\PROGRA~1\GORDIA~1\mpeg2dec.dll")
LoadPlugin("C:\PROGRA~1\GORDIA~1\decomb.dll")
mpeg2source("E:\work\futurama\extracts\ep2\vts_01.d2v")
Telecide(firstlast=true,blend=false)
crop(11,2,698,572)
BicubicResize(640,480,0,0.5)

Change paths and resize-values to what applies to you...

As easy as it gets :) And the results are very good, but be sure to use the 3.11 version of decomb.dll!

starfish

[MediaMage]

ah thanks!!

[MediaMage]

the entire script???

or just the beginning or so?


got it

ain´t a day where i cant learn something new bout gknot!

[MediaMage]

since decomb 3.8 is out, are there any enhancements that will help with our problem?

[starfish]

Originally posted by MediaMage
since decomb 3.8 is out, are there any enhancements that will help with our problem?

I've tried to follow Donald Grafts enhancements to his filter, but I haven't found anything that would do the Futurama DVD's better. Anyway, our final results got really good!

And ahhem... Correct me if I'm wrong!

starfish

[MediaMage]

yeah they rule!

after encoding all 13 episodes...

i just sold the dvd box, who needs that crappy quality???

its a shame what comes on these 3 dvds...

i just love my divx rips *G

[Arcon]

i have exactly the same problem (same dvd), but with the script posted above i still see some major problems. please take a look at these frames (http://www.geocities.com/futdeinter/frames.zip). i made screenshots of the raw pictures without any deinterlacer and shots with

Telecide(firstlast=true,blend=false).

as you can see, i still get artifacts (file 9271decomb, fry's face) and even those nasty ghosts (frame 9273decomb).

is this the best i can get or have i done something wrong? i tried decomb391, 311 and even greedyhma, but i cant get rid of those ghosts. btw: it works for all the other episodes on the dvd but not for the first one.

please let me know if there is a way to fix this or if i have to encode it this way :(

[starfish]

Hi!

Didn't find your example frames, but I remember I had to use "anti-shit" i GKnot on all the episodes to get rid of annoying 'blocks'. Try that!

Don't know exactly what this option does at all, but it solved my problems :)

starfish

[Arcon]

>Didn't find your example frames

i named the files after the framenumbers, so its frame 9270-9274 of the first episode ;)

>Don't know exactly what this option does at all, but it solved my problems

Q: What exactly does "Anti-Shit" do ?
A: Sometimes the codec can produce errors that really mess up the picture quality in very ugly ways. "Anti-Shit" makes nandub compare each frame that it encodes with the original frame from the frame- server - if it detects differences it re-encodes the frame as a key frame or as a delta frame with a lower DRF - which, if you don't understand what that means, should fix the codec errors. The downside of doing this is that the encoding process takes a lot longer.

i don't think this might help for my ghosts, since i use divx5 and therefore virtualdub instead of nandub and i dont think, that these frames will get any better if i encode the frames as keyframes. the problem seems to be the source, not the encoding-process.

[starfish]

See private message :)

starfish

[Arcon]

if anyone else got trouble downloading the zip-file with my screenshots, please tell me and i'll send it to you.

[Arcon]

i just fiddled around with the options of telecide and found out that i can get a ghost-free frame 9273 by setting the swap-parameter to true. telecide is then able to find the matching top/bottom fields for this frame. the problem is that i get a ghostframe at 9271 if i use swap.

if i understand the comment in the source of the telecide-filter right,

/* Calculate up to six differences: between the top field
of the middle stored frame and the bottom fields of
all the stored frames (3) and (if needed) between the bottom
field of the middle stored frame and the top fields of
all the stored frames (3). We'll pick the best match from
these possibilities. */

the filter searches for every top/bottom field a best matching bottom/top field from the actual frame, the last frame and the next frame. so how can it be that it finds the right one for 9273 only if i swap the lines first?

@neuron2: i'm sure you gould give me a hint, even if its just a RTFM (if the answer is in TFM ;) ). i really tried to find the reason for my problem and from my understanding of your source-comment it should be impossible to get the ghost-effects i get at all :(

the top-field of 9273 matches the bottom-field of 9272,
the bottom-field of 9273 matches the top-field of 9274,

but Telecide(firstlast=true,blend=false) does not choose this combination.

[neuron2]

@Arcon

Ummm, well. You are quoting the source code comment from the VirtualDub version of Telecide() but you are using the Avisynth version. The Avisynth version does not currently support the 6-way comparison.

I suppose your next question will be whether I intend to support it. I've not given it much thought recently. Quantify things for me. If you have just a few such frames, is it really a big deal, especially when you can't notice them when played at normal speed?

Maybe you can use the VirtualDub version for this application?

[Arcon]

The Avisynth version does not currently support the 6-way comparison.

ok, this solves all my questions from above at once ;)

If you have just a few such frames,

the ghosts appear every ~50 frames (in the scene from the screenshots 3 times in 10 frames, so this depends), the artifacts every time a scene changes or something moves rapidly.

is it really a big deal, especially when you can't notice them when played at normal speed?

for me it is, but maybe my demands are to high. at least i did notice them, since i first used greedyhma and found the same ghosts while watching the encoded material. i then decided to try telecide to get rid of them. so if you intend to adapt your filter to the fox way of storing pal-dvds in the next few months i'll definitely wait for that before i encode the whole dvd-box :)

Maybe you can use the VirtualDub version for this application?

if this works it would be perfect, too. but right now i'm not sure if i use it the wrong way or if it doesn't help either for my material. i removed all plugins from my avs-file, opened the avs in vd, selected the codec, loaded the telecide-filter and encoded the frames. but the result doesn't look quite different. the frames look pretty much like the ones made with the avs-version, only that i can still see the comb-lines :(

if it would be usefull for you i could cut this scene from the vob-file and send it to you. it's just 30 frames long.

[neuron2]

>if it would be usefull for you i could cut this scene
>from the vob-file and send it to you. it's just 30 frames long.

OK. If possible include a portion that has blends in both fields.

[Arcon]

thanks for looking at my problem :)

the link to the sample has been sent via pm.

[neuron2]

@Arcon

You wrote in your PM:i'm not quite sure what you mean with 'include a portion that has blends in both fields' since i thought blending means to combine two fields.If you do a SeparateFields() on your clip and view the individual fields, you will see that some of them are each blends of two pictures. One field that has two pictures blended is called a "blended field". If you see such fields both in the top and the bottom, I say that you have blends in both fields.

Furthermore, if you study the field sequence of the clip, you will realize that there is simply no way to correctly match the fields so as to produce good progressive frames (even with six-way comparison). The clip has been ruined by the mastering process. My guess is that it was done as part of one or more format conversions.

As I describe in the Decomb help file, there is nothing you can do to get progressive frames out of such clips. The best you can do is to use blend deinterlacing as postprocessing for Telecide(). Usually these blends are not objectionable when the clip is viewed at normal speed. Naive viewers don't notice anything wrong.

[Arcon]

The clip has been ruined by the mastering process. My guess is that it was done as part of one or more format conversions.

hm, ok. at least i can encode it now knowing that it can't be fixed. thanks for your help anyhow :)

but just one last question: why did some of the problem frames became better when i used swap=true? ok, others went worse, but at that time i hoped that some magic 'smartswap' function could help in this case to combine the best of both settings.

[neuron2]

It depends on the specific sequence of fields. If the blend is in the bottom and Telecide() is matching on bottom, then it can't find a match. If you change the reverse option (not swap) it may be OK because it matches on the top. See the theory of operation description in the Decomb help file.

The problem is that you have frames that have blends in both the top and the bottom. There is no way to match these. There are other situations, too, that leave you with no match. If you read the theory of operation and then look at the field sequence, it should all become very clear.

Your question also tells me you didn't understand what I've said about 6-way comparisons, etc. Did you read the Decomb help file, specifically the Appendix?

[Arcon]

Did you read the Decomb help file, specifically the Appendix?

nope, i only read the part about blending. i read the appendix just after your post.

[Valky]

I also want to share my efforts with this DVD. I have ripped many movies myself, but since this one is definetely the hardest so far I help if it's still needed.

The only working script is this mentioned above, so made 'name.avs' file with notepad and copy paste this in it:




LoadPlugin("C:\PROGRA~1\GORDIA~1\mpeg2dec.dll")
LoadPlugin("C:\PROGRA~1\GORDIA~1\decombLegacy.dll")
mpeg2source("E:\F_CD1\3\3.d2v")
Telecide(firstlast=true,blend=false)
crop(11,2,698,572)
BicubicResize(544,416,0,0.5)


Load this to virtua-dub (v1.4.13).
See that you have the latest decomb plugin in the correct place (mine is default installation root for GordianKnot).
My goal was to fit 4 episodes to one CD and that means 175mb filesize for each episode. Since vbr 128kbs mp3 file for each AC3 audio was 18mb-18.6mb so that the correct filesize for each in xvid 2-pass int was: 159744
I use koepis xvid build 9-12-02. This version hasn't gave me any errors with virtua-dub like others after that. My settings were 1 b-frame/150 and only qpel. And 10% frame boost for the second pass.

These settings and codec gave me perfect video without any artifacts or ghost images!! And the filesize was just 175mb on almost every episode after I muxed the audio with Nandub.

First I made these episodes with DivX 5.02 Pro with b-frames and qpel and normal psycho. Quality was pretty good althought all of them were undersized. They were about 150-166mb each.
I just wanted to test this XviD-codec with cartoons before I decided to sell this DVD-box and the quality with XviD was much more satisfying than the episodes encodec with DivX 5.02!!

[starfish]

Originally posted by Valky
These settings and codec gave me perfect video without any artifacts or ghost images!! And the filesize was just 175mb on almost every episode after I muxed the audio with Nandub.

First I made these episodes with DivX 5.02 Pro with b-frames and qpel and normal psycho. Quality was pretty good althought all of them were undersized. They were about 150-166mb each.
I just wanted to test this XviD-codec with cartoons before I decided to sell this DVD-box and the quality with XviD was much more satisfying than the episodes encodec with DivX 5.02!!

Hmm... This seems odd to me. I was (I think, it's been a while since I coded these dvd's) unable to get rid of all the ghosting, but the 'shit' was avoided with the 'antishit' option.

It I'm not all wrong, the ghosting has nothing to do with the codec, since it's decomb's output that contains the ghosts... I guess xvid maybe could eliminate the 'shit', but it would be impossible for the xvid-codec to decide if the ghost images are intended or not.

It you really got true progressive (or at least a file with no ghosting) I have to congratulate you :) I wasn't able to do this, but the decomb-filter has developed a lot since my rips, and I haven't been much into the changes lately, so I guess this is your main advantage over my rips, not the divx3.11->xvid change...

But, as I said, I'm not 100% updated anymore, so I might be wrong :)

b.r.
starfish

[Valky]

Well..I have been comparing these three sources..dvd, divx 5.02 and xvid files. And I dont find anymore those annoying ghost-images that were so obvious in that dvd-disc. But I also think that codec didn't do it. It is that DecombLegacy.dll.

But overall that XviD codec gave me much more sharpen and clearer picture than Divx 5.02. Maby I was just lucky :)

[ermannob]

IMO ghost effects will never go away completely.:mad:
i tried different setup of decomb filter for days... but i never got perfect results.

I'm trying to encode these dvds in these days. This thread game me big help. thank you everybody!;)


erm

[cowlemon]

so what is the best way to encode the series? can someone post the script?

[SpAwN_gUy]

wELL, recentely i was RiPPiNG Azumanga DVDs.. ^_^ the quality was(for XviD) ~110%... yeah,.. i Kown :).. and i DO NOT have Scripts,.. but i do have GKnot ini file. and while i was loocking through several Forums found Interesting Results. So, Here is the Deal(c) HACKERS
Oh, yeah, Input was NTSC Partially Interlased Material (why Partially? .. Intros and Outros were NonInterlased, and the whole Serie WAS.. and TomsMoComp gave me really Wierd Results... weah I had SEX wis My PS :) ).. SO:
Telecide(order=1,post=2).Decimate(5,0,0.0,3.0,2,"",false) [Decomb]
or Just
Telecide(order=1).Decimate() [Decomb]
then CROP...
... and then One of the Following Filters, depending on PCPower you Have..(Didn't noticed any great Differents betwen Those..)
Deen("a3d",2,5,5,5) [Deen]
Deen("a3d",1,5,5,3) [Deen]
Convolution3D("animeHQ") [Convolution3D]
..then, Started using this Filter recently
ColorMatrix("Rec.709->Rec.601",false) [ColorMatrix]
..and then
--> RESIZE

well it was GOOD for ANiME Encoding.. proove me wrong,.. and ANY Comments would be Great..

Sorry 4 Ma BAD Inglish :)