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. |
15th June 2004, 20:35 | #21 | Link | |
ffdshow/AviSynth wrangler
Join Date: Feb 2003
Location: Austria
Posts: 2,441
|
Quote:
Of course, better wording might have been a good idea... np: Underworld - Pearl's Girl (Underworld 1992-2002) |
|
16th June 2004, 14:32 | #23 | Link |
Guest
Join Date: Jan 2002
Posts: 21,901
|
I thought of an idea that may help us out of our dilemma.
The basic problem is that after decimation we are missing some unique frames from the input. We can't put them back without messing up audio. Or can we? What if we put them back as blends: before: A A A B B C D D D E E normally C could disappear, but if so we render: A B+C C+D E F I believe this may be a solution for decimating these clips while retaining smoothness and audio sync. I'll perform some experiments and report back. |
16th June 2004, 15:51 | #24 | Link | |
I'm afraid we've to stop
Join Date: Mar 2003
Location: Amongst mad people
Posts: 5,398
|
Quote:
__________________
Gordian Knot Family: Gordian Knot: website, download Auto Gordian Knot: Website and download, tutorial, FAQ |
|
16th June 2004, 17:55 | #26 | Link |
brainless
Join Date: Mar 2003
Location: Germany
Posts: 3,653
|
erhmmmmm..
dropping or blending an orphan frame? why this? wouldn't it be much cleverer, delaying the video by, lets say up to 3 frames, until frames get dropped or blended? maybe sometime later in the video we have the chance getting some more duped frames that are compensating orphaned frames. or maybe accept a slight AV async, until a scenechange occurs (dropping frames directly before/after scenechanges is not noticeable, while dropping frames within a scene IS noticeable)
__________________
Don't forget the 'c'! Don't PM me for technical support, please. |
16th June 2004, 19:57 | #27 | Link |
Guest
Join Date: Jan 2002
Posts: 21,901
|
In practice, it doesn't happen that way. And the av sync mismatch is not slight. But hey, go ahead and implement it for a test of concept if you really think it is viable.
The only thing that really would be cleverer in my view would be to insert blank audio whenever required to retain orphans. I don't know how random 33ms audio silences would sound and in any case I'm not motivated enough by the problem to go down that route. But adding the blend mode to FDecimate() is not too difficult. Last edited by Guest; 16th June 2004 at 19:59. |
17th June 2004, 03:21 | #28 | Link | |
Registered User
Join Date: Jul 2002
Posts: 587
|
Quote:
to center via an audio stretch with pitch save. zep ducks... |
|
17th June 2004, 10:44 | #29 | Link | |
I'm afraid we've to stop
Join Date: Mar 2003
Location: Amongst mad people
Posts: 5,398
|
Quote:
__________________
Gordian Knot Family: Gordian Knot: website, download Auto Gordian Knot: Website and download, tutorial, FAQ |
|
17th June 2004, 11:20 | #30 | Link |
brainless
Join Date: Mar 2003
Location: Germany
Posts: 3,653
|
for sure, I cannot implement this mzself, I just had some thoughts...
so, why is it not possible to accept a slight A/V async to achieve that only frames on scene changes (and not within a scene) are getting dropped? Most of the FIlms have very much cuts, so it should be okay... I do not mean the 24 / 30 hybrid problem here. I think about weird patterns....
__________________
Don't forget the 'c'! Don't PM me for technical support, please. |
17th June 2004, 12:17 | #31 | Link |
Registered User
Join Date: Apr 2002
Location: Germany
Posts: 5,389
|
I would prefer to post some praises about fdecimate... but alas, I found a strange thing about it, that hinders me in evaluating it for a specific case:
It seems that fdecimate is not playing nicely together with scripts that use AviSynth's conditional environment. When I exchange SmartDecimate (within Restore24) with fdecimate, then the script goes totally goofy. I discovered the following: when calling fdecimate(metrics=true), then - the clip is still decimated to 23,976 fps (the docs say it won't decimate at all when metrics=true) - much more serious: some variables in Restore24 will "freeze". Restore24 uses a good amount of global variables. And at least one of them, initialized with a value of 1.0, will freeze at 0.999..., instead of getting updated on every frame. Probably even more variables are affected, I can't tell for now. The script's flow is basically quite simple: [clip /w blended fields] -> [bob] -> [conditional filtering] -> [clip /w dups] -> [decimation] Obviously, decimation is the very last step. Without fdecimate as decimator, everything is fine. When calling fdecimate(metric=true), the mentioned freeze of one or more global variables will occur, so that the blend replacement doesn't work any more at all. Lastly, when simply calling fdecimate(), then the freeze of variables seems to NOT occur. But still, the decimation goes glorily wrong, then. Donald, I know this is a pretty muddy bug report :| But I have no clue what really is going on here. Only thing I know - something IS going on. Perhaps you have an idea where to look for the cause. I'll be happy to provide more information, if you tell me which. Regards Didée
__________________
- We´re at the beginning of the end of mankind´s childhood - My little flickr gallery. (Yes indeed, I do have hobbies other than digital video!) Last edited by Didée; 17th June 2004 at 12:19. |
17th June 2004, 13:10 | #32 | Link |
Guest
Join Date: Jan 2002
Posts: 21,901
|
@Didee
That's really strange, especially the claim that it still decimates with metrics=true! What happens if you save the output of the clip minus the final FDecimate() as an AVI and then apply FDecimate() to that. Is it the same? Anyway, please PM me with the script and some instructions for recreating the issue and I'll look at it. @scharfis_brain I'm not saying your idea is without merit, but with the two clips I have there are lengthy scenes that drift terribly out of sync. So I don't think it is reliable enough to handle all instances but it may allow one to avoid some blends. But it would be a very difficult thing to implement and may well require two pass operation. Last edited by Guest; 17th June 2004 at 13:12. |
17th June 2004, 16:38 | #33 | Link |
Registered User
Join Date: Apr 2002
Location: Germany
Posts: 5,389
|
intermediate report:
When first saving the de-blended 50fps clip to HD, and then feeding that to fdecimate, then the decimation is handled correctly. (Phew ) But even when doing it this way, setting metrics=true gives me a strange output. fdecimate then delivers a clip with 23.976fps, and 'correctly' reduced framecount. But the actual series of frames is : new-dup-new-dup-new-dup-dup-new... I'll provide you tomorrow with test material. - Didée
__________________
- We´re at the beginning of the end of mankind´s childhood - My little flickr gallery. (Yes indeed, I do have hobbies other than digital video!) |
17th June 2004, 22:07 | #34 | Link |
Guest
Join Date: Jan 2002
Posts: 21,901
|
I know what's going on. I'll fix that in the next release, which will also add the singleton blend mode.
I forgot to make the fps and frame count settings conditional on the metrics setting. I'd like to get the test materials anyway, to address the other problem you reported. |
18th June 2004, 03:17 | #35 | Link |
Guest
Join Date: Jan 2002
Posts: 21,901
|
Status report: Found some bugs that could explain the problems. Fixed them. YUY2 was causing memory overwrites.
I've completed the code that detects when a skip occurs and it is working well. Now I just need to add the frame blend code stolen from Decomb and then we'll know whether it was all worthwhile or not. I'll do that tomorrow. BTW, my skip detection code shows that I accumulate as many as 10-15 skips before a scene change in many cases. That would be a large desync and then I'd have to lop off 10-15 frames at the scene change. So it's a good idea in theory but not in practice. Last edited by Guest; 18th June 2004 at 03:23. |
18th June 2004, 05:52 | #36 | Link | ||
Registered User
Join Date: Jul 2002
Posts: 587
|
Quote:
Quote:
down the hall way for like 1 minute without a scene change?(just curious) |
||
18th June 2004, 12:38 | #37 | Link | |
Guest
Join Date: Jan 2002
Posts: 21,901
|
Quote:
|
|
18th June 2004, 16:14 | #38 | Link | |
Registered User
Join Date: Jul 2002
Posts: 587
|
Quote:
i love to test on because lots of movement (the pillars) makes for easy spotting of singletons and i saw plenty of them. (I never counted how many though) yes it is the singleton cuts that are really obvious in that scene. |
|
19th June 2004, 13:28 | #39 | Link |
Guest
Join Date: Jan 2002
Posts: 21,901
|
Blend results are disappointing
I've completed my experiments with the blend idea. They are very disappointing. The skips are just as obvious if slightly attenuated. The improvement is nowhere near enough to justify itself.
So I just went ahead and fixed the existing version, resulting in version 1.0.1: http://neuron2.net/fdecimate/fdecimate101.zip For those wondering how FDecimate() differs from ChangeFPS(), it can preferentially deliver unique frames, avoiding duplicates where possible. I don't see a fully acceptable solution to Zep's problem short of a coordinated adjustment of the audio. @didee I tried to run your restore24 scripts but I got an error that the script wasn't returning a video clip and I didn't have time to to track that down. Also, there was a syntax error on a levels call, so apparently this script had never run. Anyway, please try the fixed version and report back. Thank you. |
19th June 2004, 17:26 | #40 | Link | |
Registered User
Join Date: Jul 2002
Posts: 587
|
Re: Blend results are disappointing
Quote:
Do you even want to try to adjust audio or is that too much of a pain to get into? i wish i had to a way to do the whole clip like i did in my tests without doing it manually just to see how good it comes out on a full encode. Are there any free compilers that can do avisynth filter dev with? I have some ideas but i can't ask you to keep messing with this. It is time i get my hands dirty and i always wanted to get into filter tests so now is a good as time as any Last edited by Zep; 20th June 2004 at 05:26. |
|
Thread Tools | Search this Thread |
Display Modes | |
|
|