Ben 10 PAL, is there anything I can do?

[ndjamena]

I've encoded Ben 10 before but I'd remembered I'd left a lot of blends behind, since I'd had such luck removing the blends from Turtles Forever with:

sRestore(frate=23.976,speed=1,cache=60,thresh=12)

the second time round I thought I'd go back and see what it did for Ben 10. Unfortunately it's not going to work, it's not immediately obvious since a lot of it is native 12p with doubled frames but as far as I can tell EVERY FIELD is blended in some way. Is there anything I can do to reverse whatever they've done to it?

http://www.mediafire.com/download/7248mvsw29kkvcs/Sample_1.mpg

And since I'm asking about Ben 10, this is from 'Ben 10: The Secret of the Omnitrix' - watch his mouth move. I'm not sure but at a guess I'd say the movie was IVTC'd from the NTSC masters then resized to PAL with speed-up applied but the IVTC was done poorly and left artefacts. Is there any neat trick I could apply to make the damage less obvious without destroying the entire movie? Anti-aliasing maybe? It's apparent all the way through the movie.

http://www.mediafire.com/watch/wdd2dach9habbad/Sample_2.m2v

These are from official retail DVDs by the way.

[lordsmurf]

Pretty much everything is released.
Buy it.

The {C_P} downloads suck. I know that was the source here.
That is the shittiest torrent release group I've ever seen. They butcher everything.
They're morons.

[ndjamena]

Well,



As master Splinter always says, 'the more one learns the less one knows'. Apparently {C_P}'s encoding abilities are on par with Madman Australias. Unfortunately, short of ordering them in from America this is the best I've got top work with. If it's unfixable then that's fine, I'll just stick with what I've got but if there's a way of reversing this I'd like to at least try.

[ndjamena]

Sorry, I thought it would resize it automatically.

[El Heggunte]

Sorry, I thought it would resize it automatically.

You can delete the overbloated images

[jagabo]

SRestore() gets rid of the worst of the blending.

Mpeg2Source("Sample 1.d2v", CPU=6, Info=3) 
Interleave(TFM(field=1),TFM(field=0)) # or your choice of bob filter
SRestore(mode=2, frate=23.976)

[lordsmurf]

Apparently {C_P}'s encoding abilities are on par with Madman Australias.

Yeah, they did that on Inspector Gadget, too.

That is crappy.

The NTSC ones are fine. I bet they did a cheap/lazy conversion from them.

[ndjamena]

Interleave(TFM(field=1),TFM(field=0)), um, IVTC using both top and bottom field first then interleaving the results to 50p, is that preferable to QTGMC(Preset="Placebo") for this job? I ran it through the entire episode and was left with prevalent interlacing in the end result... Unless Mode=2 is the point? I'll try that with QTGMC.

I have Inspector Gadget! Phew, Magna Pacific, not Madman! Hopefully that's a good sign! (although the best sign would be 'This DVD is NTSC Format' like I got with Dilbert :P) I learned too late not to buy American TV shows from Australia (oh crap, I just bought Season 2 of Iron Man: Armoured
Adventures...)

[jagabo]

Interleave(TFM(field=1),TFM(field=0)), um, IVTC using both top and bottom field first then interleaving the results to 50p, is that preferable to QTGMC(Preset="Placebo") for this job?

Use whatever bob you want. I find QTGMC() sometimes creates artifacts in animated material. The dual TFM technique is shitloads faster. But use whatever works.

I ran it through the entire episode and was left with prevalent interlacing in the end result... Unless Mode=2 is the point? I'll try that with QTGMC.

I didn't see any in the short sample. But the full video may have issues. For example, when there is an orphan field you will be left with either comb artifacts or a blend with another field.

The main point was that SRestore does a fair job on the sample. Certainly better than normal field matching or deinterlacing and decimation.

[lordsmurf]

Maybe it was Magna Pacific ... the discs aren't near me right now.
All I know is that they had blending blur.

[ndjamena]

I still had the QTGMC UT Codec lossless processed with QTGMC("Placebo", TR0=1,TR1=1,TR2=0), which SHOULD be more animation friendly. So time and animation NOT a problem.

It's done, oh well, no improvement that I can see, did I see omode=4 at one point or was I imagining it??? Since there's no simple filter to deblend this I guess it's a lost cause. If I cut the right section the main problem with the sample was when the grub falls off the plate, there's five blended fields in a row showing Gwen (24p) and srestore is ineffective against something like that. And no one's even bothered with SotO. If I try again I might try resizing it back to 720x480 and running it through a deinterlacer to see what it spits out. At least Iron Man is fully progressive, and hopefully sped up from a 720p source

I just checked an old Handbrake encode I did from my Inspector Gadget discs and there's definitely blending in there, it doesn't seem as bad a Ben 10. Of course it was done with Yadif so half the fields are missing, I'd need to rerip them with MakeMKV to check properly. End.

[jagabo]

did I see omode=4 at one point or was I imagining it???

Yes, it was in there for a short time until I decided it didn't help.

By the way, samples with more motion would have been better. Especially include a panning shot because those are usually at 24 fps, where the character animation is usually at 12 fps.

[ndjamena]

http://www.mediafire.com/?67xosgkoiztso4q

Actually, most of the 24fps shots ARE character animations - the close-ups. The backgrounds on this show tend to be blurry and indistinct so the blends aren't as apparent. I've been looking for decent scenes to really show the problem but they're all really, really short (people being too miniscule to warrant much of a pan) and the Grub scene is as good as any other. I'm watching it now on my TV and the 'meal worm' scene isn't actually noticeable but I can notice some blended frames when there's a dark object on a lighter background (or vice-versa).

[jagabo]

Actually, most of the 24fps shots ARE character animations - the close-ups.

The character cell animation is 12 frames per second. Of course, each of those frames appears as two film frames at 24 fps. That is why you can get fairly clear results from SRestore -- blends of two identical frames look the same as those frames. So even with lots of blending there are enough frames without blending to get mostly un-blended results.

The panning shot you provided was originally 24 fps where each frame was unique (the cells were moved for each film frame). So there are too many blended frames to get blend-free results from SRestore and/or TDecimate().

Yes, the double TFM bob technique does leave some comb artifacts when there aren't two clear fields to be combined. SRestore removes them when there isn't too much blending. But when there are two many blended fields/frames some of the comb artifacts get through the processing. You can try using different post processing modes in TFM to reduce the combing. pp=2, maybe.

[ndjamena]

Yes, I'm aware that most animation is 12p at this point. Maybe this is unique to Ben 10 and it's sequels but when a character animation takes up too much screen space it switches to 24p (possibly because the movements are larger and more noticeable). The sample I just posted could be considered a panning shot which would explain the 24p, but the first sample with Gwen and the mealworm is definitely NOT a panning shot. Maybe I should have given a shorter sample with just that scene in it but I'd read a post recently where Manono asked the OP to provide a sample encompassing ten seconds before and ten seconds after. Anyway, large objects on screen + blending + a low original frame rate (= large movements between frames) + reducing the frame rate back to 23.976 is a bad combination, especially with high contrast, and some of these blends are really noticeable watching the end encode. If there's nothing I can do to fix it than I'll have to live with it but after wasting all that time trying to remove Dot Crawl from Dilbert the hard way I know now not to make assumptions about what's possible or not.

That double TFM thing is interesting , once with the top field dominant, once with the bottom field - I'd like to find something to use it on but I'll just content myself trying to picture how it works for now.

[jagabo]

If there's nothing I can do to fix it than I'll have to live with it but after wasting all that time trying to remove Dot Crawl from Dilbert the hard way I know now not to make assumptions about what's possible or not.

There might be some way to work FixBlendIVTC, or some of the techniques it uses, into the processing.

[lordsmurf]

I'm not sure new cartoons like Ben 10 are 12fps.
Many new shows use computers to assist, and are actually 24fps when done.