Final Fantasy VII R4 PAL DVD 2 Disc version - Horrible Transfer :(

[Mr.Bitey]

Hi Guys,

I'd been hanging out for FF7 for a long time only to be utterly dissapointed with R4 PAL transfer - it is horrid :(

It looks like its interlaced and has awful motion - looks like its blending/adding a frame (noticable on all the panning - and there is a lot of panning) and fast motion just turns to blur :(

The stills/slow moving parts look fantastic.

Ive been trying to find a decent decoder/filter to watch this as I _really_ want to watch it on my HTPC resized/denoised/limitedsharpened but its just awful :( - to date windvd7 seems to give acceptable results with deinterlacing over dscaler5 but the motion :(

Any ideas on how to fix? (or is this a restore24 (which I still havent gotten working yet) job?)

I suspect first job will be to deinterlace (running tomsmocomp as a dscaler plugin through ffdshow (realtime) didnt look quite right)

Cheers,
Bitey

[MOmonster]

Sounds as a typical restore24 / Mrestore problem. But without a sample I can´t help you more.

[Revgen]

Buy the R1 version.

[Mr.Bitey]

I turned off limtedsharpenfaster and tried the dscaler plugin DI_VideoWeave.dll - on the default settings it looked pretty good with some combing on very fast motion.. At least watchable!

I'll pull a short clip out when I get home.

Revgen - is the R1 version perfect?! :)

Cheers,
Bitey

[desta]

http://forum.doom9.org/showthread.php?p=825553#post825553

http://forum.doom9.org/showthread.php?p=825907#post825907


If you're wondering why I'm linking those, it's cuz it's the same film (and transfer). I've ordered the R1 version instead now. MOmonster helped me a lot with this, but the dvd on a whole has far too many bad frames.

Mr.Bitey - yeah, the R1 version is much better... just a simple case of ivtc.

[Mug Funky]

columbia tristar should be doing better!

i'm going to assume they did all the PAL discs this way (is there a "language selection" menu before the main menu? if so you can be sure the same disc has gone to all PAL countries, just with changed region codes). not cool.

though i shouldn't complain about blend-style standards conversion. i've done it myself countless times on series anime (no other choice on hybrid content). if the source is pure film though, speedup is the better option, and it should be done for all big name titles.

[desta]

columbia tristar should be doing better!
Well it's not like Sony has been making a great name for itself of late.


i'm going to assume they did all the PAL discs this way (is there a "language selection" menu before the main menu? if so you can be sure the same disc has gone to all PAL countries, just with changed region codes). not cool.
No, there's no language selection until you're actually in the title menu.... well.. that's for the R2 version anyway.


though i shouldn't complain about blend-style standards conversion. i've done it myself countless times on series anime (no other choice on hybrid content). if the source is pure film though, speedup is the better option, and it should be done for all big name titles.
The R1 NTSC version is nice clean progressive content, once the pulldown is removed.

[Mug Funky]

yes, well ideally there'd be a direct-from-HD speedup or even a full re-render (is FF7 3dCG?) in the appropriate resolution.

but it's always the problem that licensors will simply hand over an NTSC master and the authoring house will have to convert. i've even seen this on PAL-country originated movies.

[scharfis_brain]

but it's always the problem that licensors will simply hand over an NTSC master and the authoring house will have to convert.
So where is the problem, that the authring house does a seedup conversion of this NTSC master?

IMO the majority of all big movie titles on R2 PAL disk are taken form a 480 line master.
(How I come to this conclusion?
Lack of detail and leftovers of resizers as well as vertical oversharpening and sometimes orphaned fields and triples fields (Independance Day, Starship Troopers) )

Pure native PAL transfers (from a higer resoluted master or from a new Film-Transfer) are not that common.

ohhh. Again... I am OT ;)

[Mug Funky]

So where is the problem, that the authring house does a seedup conversion of this NTSC master?

no problem... but a lot of places either don't have the facilities or the knowhow.

IMO the majority of all big movie titles on R2 PAL disk are taken form a 480 line master.

i noticed that warner bros tend to do some kind of machine-based speedup. the screen vertically shimmers twice a second, but it's otherwise progressive and quite watchable. but their premium titles tend to be direct to PAL. i guess it's a matter of the DVD budget and a bit of luck as to what master is easier to get to them.

other studios generally have native PAL. i think a lot of film restorations go straight to PAL for restoring (more lines), and convert to NTSC afterward, but that's speculation (the only titles i know of where the restoration was done in PAL and later converted to NTSC are the old Tezuka Osamu series - Astroboy 1960s and Kimba).

the recent Maralyn Monroe DVDs are native PAL to the best of my knowledge, but i haven't actually ripped one yet. just get an intuition from looking at the TV.

Lack of detail and leftovers of resizers as well as vertical oversharpening and sometimes orphaned fields and triples fields

you'll be pleased to know i've modded limitedsharpen and dehalo_alpha to only work vertically and it works quite effectively post-lanczosresize to sharpen-up an NTSC -> PAL upsize. of course sometimes the original is too blurry for any benefit, and all you get is a slightly artificial look of edges being forced too close together... this is why i'm not a fan of machine-based conversions - it's hard to get a really good look at things, and nonlinear access is impossible without 4 minute winding times and the possibility of your expensive and precious master being snapped off in an unruly machine... (this has happened twice on the 1 deck and it's frightening because the reel just goes faster and faster after the tape snaps. this is why you should always rewind your tapes to the beginning where there's nothing important that you can't scotch-tape back together).

btw, i'm not tooo worried about this thread going OT... basically the strongest medicine we have is restore24, or getting an NTSC disc.

[Mr.Bitey]

Thanks for all the advice/discussion - very interesting.

Sounds like its the R1 version for me - this is really dissapointing. I watched a trailer for FF7 lastnight (from another disc) while I was doing something else and the trailer was in PAL and was progressive and smooth as silk. Its a 3dcgi movie so there really is no excuse for this awful tranfsfer :(

I'll didnt get around to pulling a clip out last night, but I will so you can all see the awfulness of it.


Cheers,
Bitey

[foxyshadis]

I bet the trailer was sped up. Haha. Drag you in with romises of a great transfer and then shoot you down. =D Good luck on finding the R1 nearby, hope some dealer carries it.

[desta]

Thanks for all the advice/discussion - very interesting.

Sounds like its the R1 version for me - this is really dissapointing. I watched a trailer for FF7 lastnight (from another disc) while I was doing something else and the trailer was in PAL and was progressive and smooth as silk. Its a 3dcgi movie so there really is no excuse for this awful tranfsfer :(

I'll didnt get around to pulling a clip out last night, but I will so you can all see the awfulness of it.


Cheers,
Bitey

I bet the trailer was sped up. Haha. Drag you in with romises of a great transfer and then shoot you down. =D Good luck on finding the R1 nearby, hope some dealer carries it.

I buy my R1 DVD's from Loaded247 (http://www.loaded247.com/s/R1_DVD/id/23765/title/Final_Fantasy_7_-_Advent_Children/), and from that link you'll see it's clearly in stock.

:)

[Mr.Bitey]

Ok, ive upped a few clips to rapidshare.de

Slowpan.mpv - shows the judder
http://rapidshare.de/files/21070907/SLOWPAN.mpv.html

birds.mpv - shows the interlacing - really noticable on the birds in flight
http://rapidshare.de/files/21071007/BIRDS.mpv.html


bad_int_1.mpv - shows a fast panning combination = blurry mess :)
http://rapidshare.de/files/21071928/BAD_INT_1.mpV.html

I'd really appreciate some assitance in fixing this one! if its possible that is.. I can probably live with the judder (although its ugly) but the interlacing is killing me :( if someone could advise how to deinterlace this cleanly with avisynth i'd be really grateful :) - it doesnt need to be in real-time so it doesnt matter if its slow.

Cheers,
Bitey

[desta]

Mr.Bitey, like I said a few posts back, I bought the same DVD (albeit the Region 2, but probably the same transfer) and I couldn't achieve what I considered great results with restore24 or MOMonsters Mrestore... actually I couldn't really get restore24 working properly anyway. MOMonster did help me out considerably, but there were still far too many ugly frames left over (some because of the Mrestore conversion, some that were just there already) and I realised this DVD was a no-hoper.

It dawned on me that I was going to all this trouble to rectify a problem that was made by Sony, and isn't the sort of thing a paying customer should have to do. I've since returned the R2 DVD, and bought the R1 version (from the site I gave in my previous post).... The NTSC version is flawless. It's a simple case of pulldown removal if you're intending to do an encode.

[scharfis_brain]

you may try my constant pattern deblending approach which works as follows:

- bob-deinterlace the video
- increase the framerate to 10 times the desired output framerate
- select every 10 th frame with an offset from 0 till 9.

on the scripted basis it looks like this:

xxxsource("blah.xxx")
anybob() #TDeint(mode=1) #is preferred
changefps(10*23.976)
selectevery(10,offset)

replace 'offest' with 0...9
you should get a nice deblended output if the pattern is stable.
But if the pattern breaks, you'll immediatelly get the blends and the stutter back. So you'll need to trim out that scene and apply another offset to it. Like this:


xxxsource("blah.xxx")
x=anybob() #TDeint(mode=1) #is preferred
x0=x.trim(0,100).changefps(10*23.976).selectevery(10,4)
x1=x.trim(100,250).changefps(10*23.976).selectevery(10,9)
x1=x.trim(250,572).changefps(10*23.976).selectevery(10,2)
x0+x1+x2

[Mr.Bitey]

Desta,
Thanks for the advice, I did read your post and subsequently forgot it (too many forums, too many things on the go :) I didnt mean to blatently ignore what you'd taken the time to post :)

scharfis_brain,
Thanks for taking the time to help! - I will give it a go and see how it turns out. I take it I need to experiment with the offset to find the one that? Good job im 'working' from home today :-)

Cheers,
Bitey

[Mr.Bitey]

scharfis_brain,

I attempted to use the example you provided but couldnt get rid of the blends :( I tried offsets 0 through 9 and quite a few 'variations' that seemed like a good idea, but ended up getting me no where :)

Given the efforts by Desta and Momonster - I think its a lost cause (sigh)...

Cheers,
Bitey

[MOmonster]

Restore24 and Mrestore seem to recognize the pattern correct. Only some less blends left with both functions (I don´t see them during playback). The problem is more the bad quality, too strong motionblur and many artefacts from the blended fields left. You can try these functions and decide for your own if the quality is good enough for you.

[scharfis_brain]

you need to deblock (mpeg2source(....,cpu=4) the video before running it through the deblender. Then it will detect the blends more reliable.

[Mr.Bitey]

Thanks guys - i'll have a play around over the weekend after ive finished 'fixing' my abyss special edition..

Cheers,
Bitey

[quake74]

Just for fun, I did some tests too, and the results are here (http://rapidshare.de/files/21325951/test.zip.html).
Actually, I think they both exibit huge judder. But maybe I just don't know how judder looks like, can anyone provide a clip with judder and with the judder removed?

The script for slow.avi is

import("Mrestore.avs")
DGDecode_mpeg2source("D:\SLOWPAN.d2v",info=3,cpu=4)
ColorMatrix(hints=true)
Tdeint(mode=1)
Mrestore(profile=1)
lanczosresize(1024,576)


and slowmegui.avi (the avs was generated by megui automatically)


DGDecode_mpeg2source("D:\SLOWPAN.d2v",cpu=4,info=3)
ColorMatrix(hints=true)
edeintted = last.SeparateFields().SelectEven().EEDI2(field=-1)
TDeint(full=false,edeint=edeintted)
#crop
LanczosResize(1024,576) # Lanczos (Sharp)
#denoise

[MOmonster]

@quake74
Yes, slow pans are a problem for Mrestore. Restore24 will give more fluid results with this small sample. Still Mrestore shows less judder than the simple deinterlacing and your settings are not the right for this source. I can´t test it here, but profile 1 is setted for good quality real movie sources, not for this source here. Profile 2 should for example give a better motion.

[quake74]

@quake74
Yes, slow pans are a problem for Mrestore. Restore24 will give more fluid results with this small sample. Still Mrestore shows less judder than the simple deinterlacing and your settings are not the right for this source. I can´t test it here, but profile 1 is setted for good quality real movie sources, not for this source here. Profile 2 should for example give a better motion.

Ok, new tests here (http://rapidshare.de/files/21328389/Archive2.zip.html), where slow0.avi is the same script above with profile=0 and slow2.avi has profile=2. To me, slow2 looks the best. And I thought this kind of CGI behaves like real movies wrt compression (for example, it has motion blur and the palette of colors is a lot more natural than Pixar movies).

[MOmonster]

Ok, new tests here (http://rapidshare.de/files/21328389/Archive2.zip.html), where slow0.avi is the same script above with profile=0 and slow2.avi has profile=2. To me, slow2 looks the best. And I thought this kind of CGI behaves like real movies wrt compression (for example, it has motion blur and the palette of colors is a lot more natural than Pixar movies).
Yes, I also think that this kind of CGI behaves like real movies, but the quality is not really good and profile 1 is recommed for a good blenddetection and not for a fluid output of pans, so it´s just not recommed for this small sample.