STRANGE Footage Problem! (Pictures And Clips Included)

[Melichan923]

You'll see right away there are tons of things I have to work on with this footage and I'll probably end up making a separate thread for those problems, but what I'm most concerned about right now is this blending (or is it?) problem in the footage. The DVDs are the R1 release of Kono Minikuku mo Utsukushii Sekai (This Ugly Yet Beautiful World) by ADV. No matter what de-interlacing/IVTC method I use, I seem to have no luck. It almost seems like this problem is embedded INTO the frames and not related to interlacing at all, but I'm hoping that is not the case and I can use a script to fix it.

Here are some pictures from episode 1:

Spoiler :

Frame 1 (Fine):




Frame 2 (The bad one). You can see parts of the next frame already "blended" into it (The white around his hair and right arm, for example):




Frame 3 (Fine again):

And an even better example from episode 1:

Spoiler :

Frame 1 (Fine):




Frame 2 (The most noticeable horrible one):




Frame 3 (Right before a scene change, so it's kind of bad):

It may help out a lot if you open each picture up in a new tab or save them to your computer and switch back and forth between them. It will be more noticeable that way.

A small 11MB video clip can be downloaded here:
http://www.megaupload.com/?d=XVL5DGR0
If anyone wants more clips to experiment with, just ask.

From what I can tell, not ALL scenes have this. It looks like just the ones that have the same frame repeated and are about to change or in high-action moments. When saving a clip as uncompressed and playing it at normal speed, it is barely noticeable, but if I were to slow clips down, this becomes a huge problem and is very noticeable.

My current AviSynth script:

Code: Select all

MPEG2Source("G:\TUYBW1\VTS_02_1.d2v", cpu=0)
TFM(d2v="G:\TUYBW1\VTS_02_1.d2v",order=-1,mode=5,pp=7,slow=2)
Tdecimate(mode=1)
crop(8,2,-8,-2)
lanczos4resize(848,480)
FluxSmoothST(temporal_threshold=2, spatial_threshold=2)
undot()
limitedsharpenfaster(ss_x=2.0, ss_y=2.0, wide=true, edgemode=1, strength=160)
Converttorgb32()

The problem seems no better with or without those filters.

For people who don't want to download the clip, this DGIndex picture might help a little:
DGIngex Picture

Is there anything I can do to fix this? I'll appreciate any and all suggestions no matter how small because I'm on a very tight deadline with this project. Thanks!

[Scintilla]

Can you put up a lossless video clip, or at least a sequence of PNG images, from the source WITHOUT any AVISynth filters on it?

[Melichan923]

Sure I can. They'll be posted in a few minutes.

Does it matter whether they are HuffYUV, or Lagarith? I usually use Lagarith and it will be better to upload since it's smaller.

[Melichan923]

I forgot to mention that the clip above was a .m2v dvd clip with no filters applied. That was probably confusing since I posted the filtered screenshots, sorry.

Here is a lagarith clip with absolutely no filters applied:
http://www.megaupload.com/?d=Z10YIIJ9

Here are some unfiltered screen shots:

Spoiler :

Frame 1:

Frame 2:

Frame 3:

Frame 4:

If you want to view anything else, I can post it.

[Scintilla]

Ugh. Had a look at the AVI you put up, and I don't think you can solve this problem with normal deinterlacing methods -- it appears that there is blending/ghosting within each FIELD, which you can see if you make a script that just loads the file and then does the following:

AssumeTFF().SeparateFields().BicubicResize(720,480)

Then go to frames 62 and 63. That's one frame of the original source, and it looks pretty normal. So do frames 66 and 67. But frames 64 and 65 are bad -- you can see the ghosting in each one, and they're NOT THE SAME: the top field (64) looks more like the previous frame, while the bottom field (65) looks more like the next frame.

It looks like, somehow, the chroma did not end up moving in synch with the luma... but I'm not positive on that. I have no idea how to fix this, and I'm scared to recommend IVTCing the Y and UV channels separately.

And KonoMini came out AFTER FLCL -- if this is because of a bad master, GAiNAX has no excuse.

[LantisEscudo]

And KonoMini came out AFTER FLCL -- if this is because of a bad master, GAiNAX has no excuse.

It's not Gainax's fault, it's M.O.F.C.'s (the R1 authoring house). I've had this same chroma/luma desync issue with some of their other DVDs, too. Mine were more minor, though, and mostly cropped up in hard cuts so they were easier to avoid. I'm not sure what to recommend for you, Meli.

[Melichan923]

It looks like, somehow, the chroma did not end up moving in synch with the luma... but I'm not positive on that. I have no idea how to fix this, and I'm scared to recommend IVTCing the Y and UV channels separately.

And KonoMini came out AFTER FLCL -- if this is because of a bad master, GAiNAX has no excuse.

Yikes! I tried out what you said and you are absolutely right. Those frames look horrible... Do you think that method might actually work even if it's risky? If so, please feel free to lead me to a page that might help me attempt to IVTC the Y and UV channels, although I doubt I'll learn how to do it in time. I'm not very knowledgeable in this area and I probably won't understand how to do it.

It's not Gainax's fault, it's M.O.F.C.'s (the R1 authoring house). I've had this same chroma/luma desync issue with some of their other DVDs, too. Mine were more minor, though, and mostly cropped up in hard cuts so they were easier to avoid. I'm not sure what to recommend for you, Meli.

Hm... I wondered if the R2's had the same problem. Does anyone have the original Japanese DVDs who can confirm if they do or not?

I'm not sure what I'll do either Lantis. I really don't want to have to download the raws and I currently don't have the time or spare cash to order new DVDs for this project.

Thanks a whole bunch for your help and time. If anyone has any suggestions or anything I'm willing to try them. I don't want to be stuck editing with this ghosting, lol. Grrr, It will bug me to no end.

[Qyot27]

Normally if I come across frames like that I cut my losses and delete them from the clips I make. However, if this is just way too much to deal with (i.e. that the deletion of those frames would cause motion problems), then maybe it would be possible to separate the luma and chroma, use a deblending filter on the affected channels and then recombine them? Not that I know how to do that or even have tried, although I do have the DVDs in question (I made sure to get the regular edition, not the slimpack - you didn't mention which one you had; it'd only be an issue if ADV screws up even bigger than usual* on their slimpacks and the encodes are completely different in addition to lacking extras).

*Because I seriously doubt that if the encodes differed, that the slimpack versions would get it right.


To do this, you'll need to set up two processing lines, defining which is the luma info and which is/are the chroma info. On the affected channels, use a deblending filter (Cdeblend, for instance, although others exist), and then use Merge (see link) to combine them at the end. If it's necessary to IVTC the footage, you'd use roughly the same method. The page discussing Merge shows examples on how to set the script up properly.

[Melichan923]

Hey Qyot,

I have the regular edition with the DVD extras as well. If you have a few minutes today, could you please put your DVDs in and see if yours have the same problem? I'll appreciate it a lot to find that out. I highly doubt it, but maybe I ended up somehow picking them up from a bad batch or something.

Thank you very much for that info. I'm not sure if I have the time (for this project) to figure out how to do that all though. I will experiment in my extra time today. I really hope I can figure it out within a day or two... otherwise I'll be stuck editing with this problem. I *could* cut out the frames that are just duplicates and replace them with the good frames in Vegas when the problem gets really noticeable... but again, not sure if I have the time to do it with the deadline.

*Crosses fingers she can figure it out in time*

[Qyot27]

I took a look at them, and unfortunately they do exhibit the same issue (I also grabbed the m2v sample and tested that too). I also tried playing around with the suggestions I made, and it doesn't seem to help much, or at all. For exposition's sake, though, this was the script:

Code: Select all

MPEG2Source("tuybwep1_1.d2v", ipp=true, cpu=4)
deint = TDeint(mode=2, mtnmode=3, blim=100, chroma=true, tryWeave=true, slow=2)
TFM(d2v="tuybwep1_1.d2v",order=-1,mode=5,PP=7,field=-1,slow=2, clip2=deint)
clip2 = Blur(0,1)
Overlay(last,clip2,mode="chroma")
TDecimate(mode=1)
AssumeFPS(24)
Crop(8,0,-6,0)
deen("w3d",2,8,8,8,8)
gradfun2db(4.0)
ConvertToRGB24()
Lanczos4Resize(848,480)