IVTC Problems

[ObiKenobi]

Doing some encodes of my Trigun DVDs and I've just finished encoding the first 3 episodes. The problem is that while episodes 1 and 3 came out perfectly, episode 2 just looks really horrible and I'm not sure what to do.

Here's the script I was using on all three:

LoadPlugin("C:\Program Files\Avisynth 2.5\plugins\DGDecode.dll")
LoadPlugin("C:\Program Files\Avisynth 2.5\plugins\Decomb521.dll")

mpeg2source("E:\TRIGUN_01\VIDEO_TS\thing.d2v")
Telecide(order=1,guide=1,vthresh=100).Decimate(mode=2)
Crop(8,2,-4,0)
LanczosResize(640,480)

I've also attempted using KernelDeint and BlendBob but it still looks horrible with either bad combing or lots of ghosting.

I've posted a small sample clip here (ftp://sample:sample@68.1.228.188 ), since I have no webspace I just used my ftp to host so sorry if it's slow. If anyone can help me I'd really, really appreciate it.

[Valeron]

Hi,I've downloaded your sample clip and have a view at it.
Completely following your scripts but just reduce vthreshold to as small as 10 to remove all the combs in certain frames.
Meanwhile,a certain amount artifacts is produced,but not so easy noticable as combs will be.
But I haven't notice too much ghost as you describe.
Have no more advise this moment.

[Chainmax]

First of all, your vthresh is set too high. The default is 50, and I always use less on my encodes. Keep in mind that vthresh only controls which frames are detected as combed. In order to enhance the deinterlacing of a frame already declared as interlaced you have to lower the dthresh switch. Second, since this is anime, you're more likely to experience bad field matching and therefore combing even after IVTCing, so I'd advice you to use the back and bthresh switches. The line would be something like this:

Telecide(order=1,guide=0,post=2,vthresh=30,blend=false,back=1,bthresh=20)


Third, I saw some rips of this series and most of them had rainbowing. I recommend you that you use sh0dan's DeRainbow (http://forum.doom9.org/showthread.php?threadid=64432&highlight=guavacomb+for+rainbows) script for that, copy the script that has the YUY2 and YV12 versions at the end of the first page. Finally, I noticed some heavy macroblocking on spots which I'm pretty sure is already on the source. I recommend you to use FunkyDeblock (http://forum.doom9.org/showthread.php?threadid=72431&highlight=funkydeblock) and/or Deblock (http://forum.doom9.org/showthread.php?s=&threadid=82015&perpage=20&pagenumber=2) from MVTools. You might also consider sharpening with aWarpSharp, as this anime has thick black lines that would greatly benefit from it IMO.

About the ghosting: I noticed it too, but I don't know other way to try and eliminate it than using TMPG's deghosting.

P.S: Decomb's documentation and the Doom9 article about DVD2AVI, Force Film and deinterlacing are a must read, they are really good and will explain a lot of things.

[ObiKenobi]

Originally posted by Chainmax
First of all, your vthresh is set too high. The default is 50, and I always use less on my encodes. Keep in mind that vthresh [b]only controls which frames are detected as combed.

Yes I understand what vthresh does and I've read the help file numerous time, thank you. The reason I set it to 100 was that with 50 it deinterlaced far too many frames good frames and so thats why I bumped it to 100, which worked for episodes 1 & 3 but not for 2 apparently. And since I was encoding all 3 in a block that is why they all use the same script, which probably wasn't the best idea.

Here's a screenshot from the OP of episode 1 of what happens throughout the episode:

With Vthresh = 50

http://img53.exs.cx/img53/7793/Vthresh50.png

With Vthresh = 100

http://img53.exs.cx/img53/7150/Vthresh100.png

Originally posted by Chainmax
Third, I saw some rips of this series and most of them had rainbowing.

I have the R2 Japanese discs and they don't have the rainbowing that the R1 dvds have, so that's a non-issue. Plus these DVD's are encoded with a much higher bitrate then the R1s, averaging 9.1+ mbit, and don't have any macroblocking problems either :).

Originally posted by Chainmax
Telecide(order=1,guide=0,post=2,vthresh=30,blend=false,back=1,bthresh=20)

I still notice a bit of ghosting and combing, and quite a bit of aliasing (which is the most noticeable and ugly), with this but it does look better then before, I guess I'll just have to accept that episode 2 won't look as good as the episode 1 & 3 encodes. I guess I'll just go back to trying to tweak the thresholds again see if I can't get something better.

[Chainmax]

Originally posted by ObiKenobi
Yes I understand what vthresh does and I've read the help file numerous time, thank you.
No need to react like that. I meant to be helpful, not sarcastic. And yeah, encoding 3 eps in a row isn't really a good idea. The only extra tips I can think about are increasing the nt switch by steps of 5 (default value is 10), reducing the dthresh switch by steps of 0.5 (default value is 7.0) and try guide=0. Another alternative would be to set post=1, hints=true in the line I posted and use KernelDeint as a postprocessor. Finally, the last advice is that you try BlendBob again but use this bobbing method: TDeint(mode=1,order=1,mtnmode=1). TDeint can be found here (http://forum.doom9.org/showthread.php?s=&threadid=82264&highlight=TDeint).

[Chainmax]

...or to use TDeint().Decimate(mode=2,quality=3) (credits go to Tritical for making that suggestion :)).

[Nicholi]

I had to post here simply because the world is lacking an awesome Trigun encode. The R1's are indeed quite bad, altogether with rainbowing and impossible IVTC'ing without large amounts of deinterlacing.

Best solution is to manually select which frames need to be postprocessed yourself. This could normally seem like a daunting task but there is a very useful tool which can cut back on tons of time using the ancient method of creating your own override file in notepad. I know chainmax knows about Yatta, not sure why he didn't mention it.

http://yatta.mellbin.org

Basically you can have Yatta mimic whatever normal options you use in your Telecide line in avisynth (guide, back, nt, and such if you use them) but the major benefit is you get an instantly visual result of changing matches and generation of an override file which will be used in the avs script. Mostly though you want to turn off all comb detection in Telecide and mark only the frames which you yourself visually detect as interlaced. This will produce the most optimal results as I see it, mainly because no unnecessary deinterlacing is done at all (as long as you have good eyes). Not to mention you'll have all of the nifty options in Yatta to use to help you in IVTC'ing the rest of the episode. It's sort of an extremely complex and useful app to sum up all at once. But you can drop by #darkhold on irc.chatsociety.net or just use the forums available at the webpage. No known Yatta guide has been created yet, so you sort of have to have an understanding of IVTC then all the different options in Yatta just make sense.

Though what you say about episode 2 is interesting, as I once ran into the R1 source and it was extremely bad. Nothing properly matched in Telecide and it was one of the episodes which required an excessive amount of deinterlacing. I hope you have much more success with the R2's.

[scharfis_brain]

I found this sample being telecined in a very weird way.

the AVISynth-Doc tells this:

PeculiarBlend (clip, int cutoff)

This filter blends each frame with the following frame in a peculiar way. The portion of the frame below the (cutoff)th scanline is unchanged. The portion above the (cutoff-30)th scanline is replaced with the corresponding portion of the following frame. The 30 scan lines in between are blended incrementally to disguise the switchover.

You're probably wondering why anyone would use this filter. Well, it's like this. Most videos which were originally shot on film use the 3:2 pulldown technique which is described in the description of the Pulldown filter. But some use a much nastier system in which the crossover to the next frame occurs in the middle of a field--in other words, individual fields look like one movie frame on the top, and another on the bottom. This filter partially undoes this peculiar effect. It should be used after th before the Pulldown filter. To determine cutoff, examine a frame which is blended in this way and set cutoff to the number of the first scanline in which you notice a blend.

This filter works only with YUY2 input.

I Cannot test it jet myself.

[Chainmax]

Nicholi, I know about YATTA, but I never knew how to do anything else than matching, plus the countless updates and cosmetic changes have made me unable to even start it up :o. That's why I haven't recommended it.

[Nicholi]

@Chainmax

Yeah back then, trying to follow Yatta day to day was almost impossible. I usually made great leaps in versions after new options were out. It's slowed down quite a bit now. I have a feeling only minor adjustments to be made in the future, more filters supported for override functionality and such perhaps (the few filters that use an ovr themselves). Likely no more cosmetic changes though I can foresee besides different kawaii girls as background pics.

6-104 being the latest, has been out for quite some time with everything working almost perfectly, nothing has come up at least.

[Chainmax]

If you don't mind me asking, what are the exact steps that need to be done in order to reach the main YATTA window (including the necessary downloads)? I am going to try TDeint+Decimate on my next encodes, but I'd like to know anyway.

[Nicholi]

@Chainmax
Yeah sure...that split to YMC.exe and Yatta.exe was a bit confusing at first. On another note I am also digging TDeint myself, liking it much more than KernelDeint in certain instances.

So for Yatta to work now you need 3 essential things, because it is no longer packaged with the dll's (current Yatta 6-104).

AviSynth 2.5X
decomb521.dll
mpeg2dec/mpeg2dec2/mpeg2dec3/dgdecode.dll

Any of the above mpeg2decs or dgdecode will now work in Yatta. All the dll's just need to be in the folder with Yatta, and then the magic begins.

YMC.exe is now the app to create the .yap file and all such metrics. Run YMC.exe, 'Add' your d2v. Be sure the proper dll is being used to open the d2v on the right (mpeg2decs). You only really need Telecide metrics calculated, the Decimate ones are quite useless in my opinion and slow down .yap generation alot. If you need to configure your Telecide options be sure to first click on the .d2v in the list on the left, then be sure to select Telecide so it is also highlighted and click 'Configure'. After that just click 'Start', and the .yap generation is underway, after its done it will immidiately open Yatta.exe with your .yap for you. From then on just run Yatta.exe and 'Open' your .yap file. Whenever you update Yatta however, or move files and lose the d2v, Yatta will then lose track of what .yap goes with what .d2v, however then after opening the .yap it will simply ask 'Where is the d2v" and you select it to load.

Thats about it. Not much has been moved INSIDE yatta. Most all the old options are there in the same places, however there are ALOT of new ones. Takes some getting used to, I'd recommend reading over the whole "Changes.txt" as it has lots of useful information even I have yet to discover.

[ObiKenobi]

Originally posted by Nicholi
I had to post here simply because the world is lacking an awesome Trigun encode. The R1's are indeed quite bad, altogether with rainbowing and impossible IVTC'ing without large amounts of deinterlacing.

....

Though what you say about episode 2 is interesting, as I once ran into the R1 source and it was extremely bad. Nothing properly matched in Telecide and it was one of the episodes which required an excessive amount of deinterlacing. I hope you have much more success with the R2's.

Well its only episode 2 that gives me any problems at all. The other episodes I've done 1,3,4,5 have IVTCed almost flawlessly (an occasional mouthflap comb but that's about it). I'm not sure what is up but episode 2 seems to have been done completely different then the rest of the episodes in that it has a bit of blue-green tint to it, it doesn't have near as much edge enhancement as all the other episodes and the telecine is completely odd. I've also tried like you mention to set a pattern for Telecide but the telecine is so screwy I can't get a consistent pattern which is the problem, it fluctuates so much that any pattern I feed it fixes some combs but adds problems in other areas. Yeah I've heard of Yatta and it's not bad, though I still prefer to do it manually.

Also, I've downloaded a few old encodes done a few years ago from the R1s for comparison and I see that I'm not the only one that has had problems as they have just as bad combing/ghosting/aliasing that I'm getting (I guess they just gave up and released it as it was rather then spend forever trying to cleanly IVTC). I'll keep trying a bit more to see if I can get a clean IVTC but I may end up just leaving it as is and having my friend use a deinterlacer in ffdshow to catch the remaining combs.

And yes the R1s are hideous in every respect to video quality that's why I got the R2s in the first place :). Rainbows and dot crawls are almost 99.5% nonexistant where the R1 had them and not to mention the higher bitrate encoding that yields no real noticeable blocking. Add with the R2s recently having been re-released I'd recommend anyone who hates the quality of the R1s to pick up the R2s, though they are a bit expensive at 5800 yen a pop for 7 discs.

[Chainmax]

ObiKenobi: if the aliasing is so noticeable, give TDeint a try. It worked wonders on my Spawn DVD in that aspect. The defaults are good, and you only have to follow if with Decimate in order to perform IVTC.

Nicholi: thanks, I don't remember why I couldn't even setup the telecide options before. I think it had soemthing to do with a sclavc dll but I'm not sure. Anyway, after creating the override file, is using Telecide(ovr="whatever") ok or do I have to put the settings (guide, back, etc) on it as well?

[Nicholi]

Ahh yes, ever since Yatta went somewhere into the 60 #'s, a COMPLETE override is created now. And I mean complete. Before of course every frame was forced into the match in Yatta. Now every frame is also specifically turned on or off for PostProcessing. So really guide, back, post, vthresh, anything having to do with detecting what match should be made, or if a frame is combed or not will have no effect on the outcome of your encode because all that has already been chosen in the ovr file. Only options you truly need in Telecide now are

Telecide(order=?,hints=?,ovr="filepath\filename")

Otherwise yeah you just about only have to load the override file and thats it.