Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion. Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules. |
31st October 2016, 20:20 | #1 | Link |
Registered User
Join Date: Feb 2004
Location: NYC
Posts: 124
|
Is this fixable? ("Field blending"?)
Trying now to process my Tintin DVDs, and dealing with a trouble spot in the first ep. This is not regular combing, it's something else, some kind of weird intermittent aliasing effect on edges (pay attention to windows and window sill in upper-left quadrant). Possibly field-blending? - but this is a mostly 23.976fps source that was produced that way in Canada, and I thought this was more of an issue related to conversion to and from 25fps. Going through the video, the windows and window sills appear to look fine in combed frames but rubbish in the otherwise good ones. Can somebody please explain what's happening here and if this is something that could be dealt with? Thanks.
Excerpt link |
3rd November 2016, 18:12 | #2 | Link |
Registered User
Join Date: Jul 2011
Location: Tennessee, USA
Posts: 266
|
There is no field blending. It uses pulldown on 1 frame every 5 frames to get 29.97fps from film speed animation created at 12fps. You can get 23.976 fps progressive by using inverse telecine (TIVTC), then use sRestore to get the original 12fps video with no duplicate frames. Anyway, that's the way it looks to me. What happened with all the aliasing in the background diagonals near the end of the clip is anyone's guess.
Code:
MPEG2Source("path...........\goons.d2v") AssumeTFF() TFM(pp=0).TDecimate(cycle=5,cycleR=1) sRestore(frate=12) Last edited by LemMotlow; 3rd November 2016 at 22:52. |
4th November 2016, 01:22 | #3 | Link | ||
Registered User
Join Date: Sep 2007
Posts: 5,377
|
Quote:
Quote:
The aliasing at the 4-5 second mark looks like parts of it had been deinterlaced at some point. As to why are only some frames were mishandled in that affected section, I don't know but somebody somewhere goofed up. There is no "easy" fix that is "good" Is this the only scene affected on the disc ? One way is to use strong antialiasing filters , something like QTGMC in progressive mode (input type =1 or 2) or even stacking a few , to stabilize the BG, but that will cause lots of motion artifacts on foreground objects Another approach is to do some manual work with rotoscoping/masks. Since the background is mostly static, and there are a few good frames to take parts from - you can try to make a BG clean plate +/- some antialiasing for foreground objects and composite them . So if it's more than a few scenes it's going to be a lot of work |
||
4th November 2016, 03:14 | #4 | Link | |
Registered User
Join Date: Jul 2011
Location: Tennessee, USA
Posts: 266
|
Quote:
|
|
4th November 2016, 03:23 | #5 | Link | |
Registered User
Join Date: Sep 2007
Posts: 5,377
|
Quote:
Look closely - on frames that are affected , foreground objects are affected too , the 2 guys, and the dog . The BG , car only appear to be affected "more" , because higher proportion of nearly horizontal lines. But the FG objects have stair stepping / broken lines too whenever the BG is affected Anyways, IVTC + QTGMC(inputtype=2 , preset="very slow") is very damaging but stabilizes it somewhat. It can be combined with some rough roto to clean it up decently |
|
4th November 2016, 04:11 | #6 | Link |
Registered User
Join Date: Jul 2011
Location: Tennessee, USA
Posts: 266
|
There's definitely something haywire with those aliased segments. In the original pulldown version the top border flutters up and down on the same frames. The aliasing mostly affects lines at certain angles.
|
8th November 2016, 20:45 | #7 | Link | |
Registered User
Join Date: Feb 2004
Location: NYC
Posts: 124
|
Quote:
Aight, I'll keep that in mind. Thanks. |
|
8th November 2016, 23:14 | #9 | Link | |
Registered User
Join Date: Jul 2011
Location: Tennessee, USA
Posts: 266
|
Quote:
If you had to do a lot of denoisiing, the duplicate frames would be a problem. But you could work around that by separating odd and even frames, then apply filters to each group, then interleave the separated frames together again. |
|
9th November 2016, 00:08 | #10 | Link | ||
Registered User
Join Date: Sep 2007
Posts: 5,377
|
Quote:
It's a big topic to do rotoscoping for the type of accuracy you need in this scenario "properly" or "efficiently" . But the generate idea is you just draw a mask around objects and the mask animates, isolating the object of interest. This way you separate objects into layers which you can composite together. Typically, you would use something like after effects or mocha. There are many tutorials on youtube and similar sites, I don' t know of any good ones offhand. There are some "easy" tools like rotobrush in AE, which is semi automated but not as accurate - but it might be good enough for many cases. There are various tips and tricks that can make roto work go much faster. People tend to start out go frame by frame - that's not the way to do it. It's the most despised gruntwork in compositing and VFX work because it's can become very tedious even for dedicated roto artists with dedicated software and hardware. There are many ways you can generate masks in general in avisynth, but it won't be accurate enough here for what you want to do, and since roto is the most time consuming, it's always the last resort Quote:
The correct way to determine the fps in general is examine the fields for unique motion. Or another way is to bob it to 59.94 to look at the unique frames (maybe with yadif(1,1) if top field first) . I wouldn't IVTC right away first, because you can lose motion samples - if there is a 29.97 fps section for example. It's very common in cartoons, some sections like the intro or credits have a higher content framerate . Or sometimes some sections or even 59.94 parts or sections like overlays |
||
|
|