Improved version of Telecide filter for VirtualDub
Donald Graft
for Avery Lee's VirtualDub: Telecide, a telecine killer. The
new version adds a high-quality mode that significantly
reduces incidence of false frame output.
I would be very interested to receive feedback about this
filter.
Thank you,
Donald Graft
Filters for Avery Lee's VirtualDub
http://www.geocities.com/Colosseum/Pressbox/8699/index.html
Legend65
wrote:
>I have just uploaded a new version (beta 3) of my latest filter
>for Avery Lee's VirtualDub: Telecide, a telecine killer. The
>new version adds a high-quality mode that significantly
>reduces incidence of false frame output.
>
>I would be very interested to receive feedback about this
>filter.
Looks like the archive is corrupt on the server. I've downloaded 3
times and can't extract the .vdf, although the .txt does.
-Kevin
Legend65
wrote:
>Looks like the archive is corrupt on the server. I've downloaded 3
>times and can't extract the .vdf, although the .txt does.
Whoops, nevermind :) Didn't see virtualdub was still open in the
background.
-Kevin
Donald Graft
The new version seems to be a really big improvement.
I captured 1200 frames off Cartoon Network and the filter
emitted only 2 minor falsely aligned frames that I couldn't
even notice at normal playing speed.
By the way, here is a nice tip for Telecide users.
Due to the one-frame delay it introduces, the audio and video
will become desynchronized by one frame (33 ms at 30 fps).
But do not despair! Just use VirtualDub's Audio/Interleaving/
Audio skew correction dialog to set a 33 ms correction. QED!
Avery is a God.
Don
"Legend65" <lege...@dontemailme.com> wrote in message news:5udt3tsthbck2qehl...@4ax.com...
Cart Wheel
> I captured 1200 frames off Cartoon Network and the filter
> emitted only 2 minor falsely aligned frames that I couldn't
> even notice at normal playing speed.
>
> By the way, here is a nice tip for Telecide users.
> Due to the one-frame delay it introduces, the audio and video
> will become desynchronized by one frame (33 ms at 30 fps).
> But do not despair! Just use VirtualDub's Audio/Interleaving/
> Audio skew correction dialog to set a 33 ms correction. QED!
Thanks for the tip (and the filter), Don. I'm probably gonna run a
comparison in a few days between VirtualDub's built-in pulldown removal
and your Telecide filter. I'll let you know what the results are.
By the way, how exactly do you recommend removing the duplicated frames?
Cart
http://www.geocities.com/lukesvideo/index.html
Donald Graft
"Cart Wheel" <n...@spam.com> wrote in message news:Qrz%5.9695$ZC4.3...@typhoon.mw.mediaone.net...
> > The new version seems to be a really big improvement.
> > I captured 1200 frames off Cartoon Network and the filter
> > emitted only 2 minor falsely aligned frames that I couldn't
> > even notice at normal playing speed.
> >
> > By the way, here is a nice tip for Telecide users.
> > Due to the one-frame delay it introduces, the audio and video
> > will become desynchronized by one frame (33 ms at 30 fps).
> > But do not despair! Just use VirtualDub's Audio/Interleaving/
> > Audio skew correction dialog to set a 33 ms correction. QED!
>
> Thanks for the tip (and the filter), Don. I'm probably gonna run a
> comparison in a few days between VirtualDub's built-in pulldown removal
> and your Telecide filter. I'll let you know what the results are.
Thank you, I will be very curious to see the result. Of course, VirtualDub is
limited to 3:2 pulldown, whereas Telecide can handle any kind of telecined
progressive frames.
> By the way, how exactly do you recommend removing the duplicated frames?
Good question. I am working on a solution to that as we speak. Stay tuned.
Aside: the output of Telecide for 3:2 pulldown is quite watchable even without
removing the extra frames, but I understand the purist's desire to remove them.
Don
Cart Wheel
VirtualDub is
> limited to 3:2 pulldown, whereas Telecide can handle any kind of
telecined
> progressive frames.
If you don't mind, I have another question. Would the Telecide filter
also handle telecined PAL video where within every 24 source frames
there were 2 fields that were duplicated to produce a 25 fps video? For
example, the odd/even fields might look something like this on account
of the telecining:
AA BB ... LL LM MN NO ...VW WX XX (pattern starts over again here from
AA)
Would it be possible to restore something like this back to the original
24 fps? (Note: I only work with NTSC video, and I don't even know if
this pattern is actually used. However, I know someone who could be
helped by the restoration of this kind of telecining.)
> > By the way, how exactly do you recommend removing the duplicated
frames?
>
> Good question. I am working on a solution to that as we speak. Stay
tuned.
> Aside: the output of Telecide for 3:2 pulldown is quite watchable even
without
> removing the extra frames, but I understand the purist's desire to
remove them.
Although it's probably true that I am a purist, I think it's important
for anyone to remove those duplicated frames. One of the key benefits
of pulldown removal is a decrease in file size for the resulting video.
Cart
http://www.geocities.com/lukesvideo/index.html
Donald Graft
"Cart Wheel" <n...@spam.com> wrote in message news:uiA%5.9740$ZC4.4...@typhoon.mw.mediaone.net...
> If you don't mind, I have another question. Would the Telecide filter
> also handle telecined PAL video where within every 24 source frames
> there were 2 fields that were duplicated to produce a 25 fps video? For
> example, the odd/even fields might look something like this on account
> of the telecining:
> AA BB ... LL LM MN NO ...VW WX XX (pattern starts over again here from
> AA)
Telecide as is will certainly produce correct progressive frames, but the
output will still be 25fps with one dup'ed progressive frame per sec. The solution
I am working on will remove these extra frames. The hacky idea is to
have Telecide generate a cut list and then have a hacked up aviedit32
read the cut list and delete the frames. That's because I've never written
a VFW app but am good at kluging things together. :-) Of course, Avery
will eventually release his new generation VirtualDub, which will allow for
filters to delete frames. That will be the cat's meeow.
> Would it be possible to restore something like this back to the original
> 24 fps? (Note: I only work with NTSC video, and I don't even know if
> this pattern is actually used. However, I know someone who could be
> helped by the restoration of this kind of telecining.)
Yes, see above.
> Although it's probably true that I am a purist, I think it's important
> for anyone to remove those duplicated frames. One of the key benefits
> of pulldown removal is a decrease in file size for the resulting video.
For sure.
Don
Legend65
wrote:
>Great, Kevin. I'd hate to inconvenience anyone. :-)
>
>The new version seems to be a really big improvement.
>I captured 1200 frames off Cartoon Network and the filter
>emitted only 2 minor falsely aligned frames that I couldn't
>even notice at normal playing speed.
I just ran through a 4 minute clip that contained several dropped
frames. Nothing noticeably wrong at normal playback, haven't gone
through frame by frame. Good enough to give it a shot at a full 45
minute episode. Seems to be a bit slower than the smart deinterlacer
in the HQ mode.
>By the way, here is a nice tip for Telecide users.
>Due to the one-frame delay it introduces, the audio and video
>will become desynchronized by one frame (33 ms at 30 fps).
>But do not despair! Just use VirtualDub's Audio/Interleaving/
>Audio skew correction dialog to set a 33 ms correction. QED!
Thanks for this, I did notice it was slightly out of sync.
-Kevin
Youri Pepplinkhuizen
I tested your new filter on a short 320x240 clip at 30 fps and even though
there was a noticable less amount of erroneus frames, quite a few still
slipped through. My guess is this is because it's a single-field capture,
therefore could you explain to me exactly what the algorithm used to
determine the redundant frame in single-field captures is (that is, if there
is one)?
Thanks, keep up the good work!
-Youri
Chris
Donald wrote:
>...one dup'ed progressive frame per sec. The solution
> I am working on will remove these extra frames.
...with the option of optically marking (corner tabs) or replacing those
frames with some color or other.
Speaking of duplicated data...
There's some _free_ noise reduction available in the duplicated _fields_ of
telecined source material, since they won't contain the same noise in the same
places. Seems a waste to just throw away perfectly good data. Got to be
locations where it can be used to improve the picture without temporal
distortion. Any chance to do that shouldn't be missed. It would require a
simple discrete average of the raw duplicate fields. (which would be also
useful for full-frames in a 2 frame sliding window)
And to that end...
A filter can't change the number of frames, but --- can it read from another
AVI file and use it to crossfade? With a horizontal nudge control, it could
line up identical clips from different broadcasts to get serious noise
reduction without smearing. I do this in MSP6, but it's hard to prepare and
crop/pad the clips with the resize or null filters. Resize seems to pad the
left edge even when I don't want it to.
How about just fading to black, without reading another file?
That's my holiday wish-list. That, and "Peace On Mars". Hey, what if we get
there and it's really annoyingly loud? Then we're screwed.
Youri Pepplinkhuizen
of
> telecined source material, since they won't contain the same noise in the
same
> places. Seems a waste to just throw away perfectly good data. Got to be
> locations where it can be used to improve the picture without temporal
> distortion. Any chance to do that shouldn't be missed. It would require
a
> simple discrete average of the raw duplicate fields. (which would be also
> useful for full-frames in a 2 frame sliding window)
Interesting. You mean that if a pattern would be ABBCD, for example, it
could compare the differences between the two B frames, interpolate the
results and then discard one of both frames? (since they're both identical
then)
-Youri
Cart Wheel
>
> There's some _free_ noise reduction available in the duplicated
_fields_ of
> telecined source material, since they won't contain the same noise in
the same
> places. Seems a waste to just throw away perfectly good data. Got to
be
> locations where it can be used to improve the picture without temporal
> distortion. Any chance to do that shouldn't be missed. It would
require a
> simple discrete average of the raw duplicate fields. (which would be
also
> useful for full-frames in a 2 frame sliding window)
>
>
> And to that end...
>
> A filter can't change the number of frames, but --- can it read from
another
> AVI file and use it to crossfade? With a horizontal nudge control,
it could
> line up identical clips from different broadcasts to get serious noise
> reduction without smearing. I do this in MSP6, but it's hard to
prepare and
> crop/pad the clips with the resize or null filters. Resize seems to
pad the
> left edge even when I don't want it to.
I was thinking of the same thing just the other day! Analog capture is
inherently noisy (especially in my case), but the noise is unique for
every capture, correct? It is seems to me that it would be possible to
capture a clip twice and then use a filter to combine the two clips to
do some serious noise reduction. I don't know if this is possible with
VirtualDub, but it seems like an excellent idea for the purists among
us.
Once you get everything aligned properly, does it work well for you in
MSP 6, Chris? What filters would I have to use to get the average of
the two clips?
Cart
http://www.geocities.com/lukesvideo/index.html
Donald Graft
"Chris" <nos...@nospam.com> wrote in message news:9KG%5.685$rn6....@bgtnsc06-news.ops.worldnet.att.net...
>
> Donald wrote:
> >...one dup'ed progressive frame per sec. The solution
> > I am working on will remove these extra frames.
> >The hacky idea is to have Telecide generate a cut list...
>
> ...with the option of optically marking (corner tabs) or replacing those
> frames with some color or other.
>
>
> Speaking of duplicated data...
OK.
> There's some _free_ noise reduction available in the duplicated _fields_ of
> telecined source material, since they won't contain the same noise in the same
> places. Seems a waste to just throw away perfectly good data. Got to be
> locations where it can be used to improve the picture without temporal
> distortion. Any chance to do that shouldn't be missed. It would require a
> simple discrete average of the raw duplicate fields. (which would be also
> useful for full-frames in a 2 frame sliding window)
While I understand your motivation, I would not do this for several reasons: 1) I do
not like the idea of having a clip intermittently sprinkled with cleaner frames.
2) The implementation is tricky. 3) I do not think the gain will be worth the effort
involved. 4) There is no *free*. The extra processing would require first determining
the duplicated frames/fields, and then averaging them. This would significantly slow
down the filter.
> A filter can't change the number of frames, but --- can it read from another
> AVI file and use it to crossfade? With a horizontal nudge control, it could
> line up identical clips from different broadcasts
The chances of getting two perfectly aligned broadcasts would seem to me to
be slight. You'd have to keep "nudging".
> How about just fading to black, without reading another file?
What is the point here? You are looking for a fade filter?
Don
Youri Pepplinkhuizen
cleaner frames.
That's what I figured when I thought about it for some more - if after
processing you'd get ABbCD ('b' being the frame where noise reduction was
applied upon - 'B' would be the dupe), one frame would have less noise than
the other 3 in the pair (not taking in account the dupe), which would mean
that after doing noise reduction on the video as a later step, it would
actually be less efficient, because the filter would first see a lot of
noise, then 1/2 of the noise (for example) and then twice the same amount of
noise as in the first frame again (in this example), and as such, the noise
could stay within a certain threshold. Of course, this only applies to
dynamic noise reduction (between frames) and not static noise, right? (Or am
I completely off the ball here?)
-Youri
Donald Graft
You're confusing me. :-) Perhaps I am not understanding your point.
First, if you are capturing only one field, there is no reason
to apply the filter at all! There will be no artifacts with single-field
capture.
Second, the filter does not have anything to do with redundancy.
It simply tries to align fields so as to produce correct progressive frames.
If the telecining added extra frames/fields, then redundancy arises, but it
has nothing to do with the operation of the filter.
The algorithm is as follows: The previous two frames are stored. Together
with the current frame, there are 3 available. At each frame time, the top
field of the middle stored frame is combined with the bottom frame of
each of the 3 stored frames in turn. The resulting frame that has the lowest
amount of combing is output. The test for combing is the same as the motion
test in the frame-differencing mode of Smart Deinterlacer. (Simon Walters first
conceived and implemented an adaptive field-aligner for PAL material.
The idea for the basic 3-frame store and compare was conceived by Aaron Bentley.
The combing test method was conceived by Gunnar Thalin. I managed to assemble
these ideas into a functioning filter.)
Don
Chris
Cart Wheel wrote:
> I was thinking of the same thing just the other day! Analog capture
> is inherently noisy (especially in my case), but the noise is unique
> for every capture, correct?
Yep. If it ain't on the master tape, it's unique to the transmission system
each time.
> Once you get everything aligned properly, does it work well for you
> in MSP 6, Chris? What filters would I have to use to get the average
> of the two clips?
Zoom your editor timeline to see each frame, drop your clips into the video
tracks (pre-trimmed to both start on the exact frame), and drag the crossfade
effect (config'd for 50% with no attack/decay) into the effects track. Then
create a video. It looks _beautiful_ as long as you line up the
spatial/temporal origins of the two clips. It's not so tough once you get
used to the cropping quirks of VirtualDub. I usually leave audio out of one
of the source clips, 'cause you certainly don't want mixed audio or you'll get
flanging.
Generally speaking, you have to cap from the same broadcast channel using the
same cap settings for both your source clips just to be sure there's not a
horizontal scaling problem (you could rescale one clip if you have to)
producing different front/back porches (side margins) in your cap. The worst
that could happen is you get some horizontal blur. There's almost never a
vertical alignment issue.
I've been doing this for a long time - it can be tedious, but it's well worth
it. If VirtualDub could do it in-house, all the better.
Chris
Donald wrote:
> I do not like the idea of having a clip intermittently sprinkled with
cleaner frames.
What if you're going to use motion-blur or temporal cleaner on the clip? Then
it helps to have those cleaner frames. They count toward better accuracy in
the finished clip, especially if you tune TC's thresholds to the clean
frames/fields.
> > How about just fading to black, without reading another file?
>
> What is the point here? You are looking for a fade filter?
Sure. Why not?
Jim Leonard
Youri Pepplinkhuizen <nos...@myplace.com> wrote:
>> 1) I do not like the idea of having a clip intermittently sprinkled with
>cleaner frames.
>
>That's what I figured when I thought about it for some more - if after
>processing you'd get ABbCD ('b' being the frame where noise reduction was
>applied upon - 'B' would be the dupe), one frame would have less noise than
>the other 3 in the pair (not taking in account the dupe), which would mean
Why the dupe? I thought we were talking about an inverse telecine filter.
Don? Does Telecide spit out duplicate frames? It was my understanding
that 29.97fps source turned into 23.976 fps output, just like Avery's
inverse telecine function.
--
Jim Leonard (Trixter / Hornet)
http://www.mobygames.com
http://www.oldskool.org
Cart Wheel
video
> tracks (pre-trimmed to both start on the exact frame), and drag the
crossfade
> effect (config'd for 50% with no attack/decay) into the effects track.
Then
> create a video. It looks _beautiful_ as long as you line up the
> spatial/temporal origins of the two clips. It's not so tough once you
get
> used to the cropping quirks of VirtualDub. I usually leave audio out
of one
> of the source clips, 'cause you certainly don't want mixed audio or
you'll get
> flanging.
Thanks for the info, I might try this out when I get some extra free
time.
> I've been doing this for a long time - it can be tedious, but it's
well worth
> it. If VirtualDub could do it in-house, all the better.
Yup, but I don't think it's gonna happen. It seems like it would be an
incredibly difficult ordeal, and I don't think too many people would use
it. Besides, Don is already busy enough with improving Telecide. I
want him to have enough energy left to code 7 custom filters for me,
wash my truck, and paint my house. :-) Just kidding, Don. You
already do so much for everyone, and I really do appreciate it.
Cart
http://www.geocities.com/lukesvideo/index.html
Cart Wheel
after
> >processing you'd get ABbCD ('b' being the frame where noise reduction
was
> >applied upon - 'B' would be the dupe), one frame would have less
noise than
> >the other 3 in the pair (not taking in account the dupe), which would
mean
>
> Why the dupe? I thought we were talking about an inverse telecine
filter.
> Don? Does Telecide spit out duplicate frames? It was my
understanding
> that 29.97fps source turned into 23.976 fps output, just like Avery's
> inverse telecine function.
VirtualDub filters apparently can't delete frames to reduce the frame
rate to 23.976 fps. Because of this, the frame rate stays 29.97 fps,
and a duplicate frame is produced.
The whole noise issue centers around the fact that one frame is
duplicated in telecined material. (The duplicated frame is actually
spread across the fields of other frames.) In theory, you could use
this duplicate frame to reduce the amount of noise in 1 frame out of
every 4 frames. However, I have to agree with Don and the idea of not
doing any noise reduction in the Telecide filter. Our brains are very
adept at ignoring a constant background noise. If 1 out of 4 frames
that had less noise, the background noise would no longer be constant.
I suspect this would be noticeable and annoying to watch.
Cart
http://www.geocities.com/lukesvideo/index.html
Donald Graft
>
> Donald wrote:
> > I do not like the idea of having a clip intermittently sprinkled with
> cleaner frames.
>
> What if you're going to use motion-blur or temporal cleaner on the clip? Then
> it helps to have those cleaner frames. They count toward better accuracy in
> the finished clip, especially if you tune TC's thresholds to the clean
> frames/fields.
I'm going to call you Mr What-If. :-)
> > > How about just fading to black, without reading another file?
> >
> > What is the point here? You are looking for a fade filter?
>
> Sure. Why not?
Chris LaRosa already made one of those:
http://www.geocities.com/cplarosa/video/tweaker.htm
ciao,
Don
Donald Graft
"Jim Leonard" <jleo...@vault.oldskool.org> wrote in message news:91ou35$17e$1...@vault.oldskool.org...
> Why the dupe? I thought we were talking about an inverse telecine filter.
> Don? Does Telecide spit out duplicate frames? It was my understanding
> that 29.97fps source turned into 23.976 fps output, just like Avery's
> inverse telecine function.
Telecide is a VirtualDub plugin filter and therefore it cannot delete or drop
frames. Sorry, I wish it were not so. Therefore Telecide produces an output
frame for each input frame. For 3:2 pulldown, where there is an extra frame
(actually two extra fields), the output of the filter will have an extra frame
versus the original progressive source from which the clip was telecined.
I am working on an external post-processor that will finish the job of deleting
these extra frames. Note that for PAL progressive source this is not necessary
(although the filter is necessary if the PAL phase or telecining method changes
throughout the clip, as is often the case) but for some others, such as some
forms of cartoon pulldown, it may also be necessary. VirtualDub 2nd generation,
code-name 'Nina', will reportedly allow plugin filters to do lots of exciting new
things, such as delete and add frames.
Don
Donald Graft
"Cart Wheel" <n...@spam.com> wrote in message news:xlT%5.12062$ZC4.4...@typhoon.mw.mediaone.net...
> Yup, but I don't think it's gonna happen. It seems like it would be an
> incredibly difficult ordeal, and I don't think too many people would use
> it.
So how does Chris get two captures that line up so perfectly?
Does he capture the same broadcast simultaneously on two different
computers? If not, what if they keep throwing in different commercials,
logos, edits, etc.? I can see the idea in theory but it just seems impractical
and the temporal and spatial cleaner filters already do very good jobs.
It wouldn't be *that* difficult, actually. But I only make things I think I
might want to actually use one day. :-)
> I want him to have enough energy left to code 7 custom filters for me,
Well, what are they? Why are you holding out on me? You know, our buddy
Simon Walters is always on the lookout for good ideas for filters too. :-)
Don
Cart Wheel
> Does he capture the same broadcast simultaneously on two different
> computers? If not, what if they keep throwing in different
commercials,
> logos, edits, etc.? I can see the idea in theory but it just seems
impractical
> and the temporal and spatial cleaner filters already do very good
jobs.
> It wouldn't be *that* difficult, actually. But I only make things I
think I
> might want to actually use one day. :-)
I don't know how he does it, but I would use it for VHS transfer. The
tapes themselves don't have too much noise, but somewhere in the video
capture process a lot of noise is being introduced. By capturing from
the same tape twice, it shouldn't be too hard to line up the captures.
> > I want him to have enough energy left to code 7 custom filters for
me,
>
> Well, what are they? Why are you holding out on me? You know, our
buddy
> Simon Walters is always on the lookout for good ideas for filters too.
:-)
Whoa now, slow down there! Let's take things one step at a time. My
truck isn't getting cleaner out there on the street, and even though
it's winter over here, my grass could probably use a cutting under the
snow. I think I heard Mr. Walters say he'll shovel for you.
Cart
http://www.geocities.com/lukesvideo/index.html
Avery Lee
>You see, the reason I was testing this on single-field material was because
>Luke recently figured out RealProducer 8 and higher can correctly inverse
>telecine single-field captures by only removing duplicate frames when it can
>accurately determine which is the dupe, otherwise it leaves the pair intact.
>That is why you can get higher fps (like 24.9, like I got on my video I
>tested with) than the standard 23.976 you should be getting after inverse
>telecining with conventional methods. However, since other encoders cannot
>do this, it would be nice to have it in a VD filter, so you can process the
>resulting AVI with different encoders, e.g. WME8.
The main problem is that even if VirtualDub could detect and remove the
duplicate frame, there's no guarantee it can write that information to disk
in a way that the encoder on the other side will detect as a variable frame
rate situation, since AVI is a constant frame rate format. The frame could
be written as a drop frame, but it's entirely possible, and correct, for
WME8 to simply encode a duplicate frame. The timing on the frames would be
slightly incorrect as well, which would lead to jerky motion.
-- Avery Lee (ule...@umail.ucsb.edu)
http://www186.pair.com/vdub/
VirtualDub 1.4c: AVI/MPEG-to-AVI processing utility with integrated capture.
Open source.
Avery Lee
>
>"Cart Wheel" <n...@spam.com> wrote in message news:xlT%5.12062$ZC4.4...@typhoon.mw.mediaone.net...
>> Yup, but I don't think it's gonna happen. It seems like it would be an
>> incredibly difficult ordeal, and I don't think too many people would use
>> it.
>
>So how does Chris get two captures that line up so perfectly?
>Does he capture the same broadcast simultaneously on two different
>computers? If not, what if they keep throwing in different commercials,
>logos, edits, etc.? I can see the idea in theory but it just seems impractical
>and the temporal and spatial cleaner filters already do very good jobs.
>It wouldn't be *that* difficult, actually. But I only make things I think I
>might want to actually use one day. :-)
I find that writing video tools ensures I don't have time to use them. ;-)
An automatic method of averaging two captures would be difficult; it sounds
suspiciously like the Unix 'diff' algorithm, which matches matching unique
lines between two files and then 'detangles' them to match the non-unique
sections. Correlating between dual sliding windows might work, though;
Avisynth could probably accommodate such a filter nicely. It could also be
done in VirtualDub, but you'd have to code a truly sick solution such as
running two separate copies in parallel and using interprocess
communication to keep the two filter instances sync'ed.
Another thing that popped into my mind is that averaging two captures only
reduces the noise margin by 30%, assuming a normal distribution.
Single-source algorithms do far better than that, with far less work. Does
anyone know how TMPGEnc's noise reduction algorithm works? It works
amazingly well!
Mikko Löppönen
"Cart Wheel"
> If you don't mind, I have another question. Would the Telecide filter
> also handle telecined PAL video where within every 24 source frames
> there were 2 fields that were duplicated to produce a 25 fps video
>(Note: I only work with NTSC video, and I don't even know if
> this pattern is actually used. However, I know someone who could be
> helped by the restoration of this kind of telecining.)
No, it is not used. 24 fps films are sped up to 25 fps for pal viewing.
-ml-
Youri Pepplinkhuizen
> duplicate frame, there's no guarantee it can write that information to
disk
> in a way that the encoder on the other side will detect as a variable
frame
> rate situation, since AVI is a constant frame rate format.
*slaps forehead* Of course, how could I forget this! If you'd get ABCD
ABBCD, it would not play back properly unless the format would support
variable framerates (which RM8 supports, but I don't know about WM8 of
course, though it most likely doesn't). So that leaves us with only the
option of removing all the duplicate frames in order to create correct
detelecined AVIs, I guess.
-Youri
Youri Pepplinkhuizen
> duplicated in telecined material. (The duplicated frame is actually
> spread across the fields of other frames.) In theory, you could use
> this duplicate frame to reduce the amount of noise in 1 frame out of
> every 4 frames. However, I have to agree with Don and the idea of not
> doing any noise reduction in the Telecide filter. Our brains are very
> adept at ignoring a constant background noise. If 1 out of 4 frames
> that had less noise, the background noise would no longer be constant.
> I suspect this would be noticeable and annoying to watch.
Good point. What do you mean with that the dupe is spread across frame
fields though? I assume that you're talking about double-field captures, cos
otherwise there wouldn't be any fields present anymore.
-Youri
Youri Pepplinkhuizen
> works amazingly well!
No idea, but why don't you email him about it? He seems like a nice guy, I'm
sure he would share his source with a fellow video editing perfectionist,
hehe. Furthermore, we could use a better noise reducer.
-Youri
Youri Pepplinkhuizen
> Simon Walters is always on the lookout for good ideas for filters too.
:-)
For me, about all filters I would ever need have been created already. What
I would like to see are improved versions of the various noise reduction
filters available. 2D Cleaner works pretty well, but it seems Jim Casaburi
quit working on it or something, apparantly. However, noise reducers can
always be improved upon, and I find it very odd that no-one tried to merge
the efforts of 2D Cleaner, Smart Smoother and fxVHS yet.
Anyway, I'm doing a little _static_ noise reduction filter comparison, so
we'll see from those results which filter works best and whether they still
need improvement.
-Youri
Youri Pepplinkhuizen
I never heard of this before. I know NTSC material that has undergone 3:2
pulldown is slowed down by 0.1%, but afaik, PAL material includes some extra
frame (created from different fields) somewhere in the pair.
I could very well be wrong though.
-Youri
Jim Leonard
Cart Wheel <n...@spam.com> wrote:
>> Why the dupe? I thought we were talking about an inverse telecine
>filter.
>> Don? Does Telecide spit out duplicate frames? It was my
>understanding
>> that 29.97fps source turned into 23.976 fps output, just like Avery's
>> inverse telecine function.
>
>
>VirtualDub filters apparently can't delete frames to reduce the frame
>rate to 23.976 fps. Because of this, the frame rate stays 29.97 fps,
>and a duplicate frame is produced.
Forgive me for sounding harsh, but what's the point of using it then?
That's the same output as if I just captured a single field at 320x240,
which has horrible motion quality. Every 5th frame is duplicated, and
things look jerky. Is the idea to code the filter now and add filter
removal support later? Otherwise, I must be missing something, as I
can't see a practical use of the filter.
>The whole noise issue centers around the fact that one frame is
>duplicated in telecined material. (The duplicated frame is actually
>spread across the fields of other frames.) In theory, you could use
>this duplicate frame to reduce the amount of noise in 1 frame out of
>every 4 frames. However, I have to agree with Don and the idea of not
>doing any noise reduction in the Telecide filter. Our brains are very
>adept at ignoring a constant background noise. If 1 out of 4 frames
>that had less noise, the background noise would no longer be constant.
>I suspect this would be noticeable and annoying to watch.
Yes, it would, but if you had a cleaned up frame 1 out of every 4,
you could use that cleaned up frame as a reference against the other
frames if you had a static scene for additional noise reduction of
the entire scene, as long as most of it was static. Food for thought.
(For the record, I'm against the noise reduction in the inverse telecine
filter as well -- I was just throwing the idea out there.)
Jim Leonard
Thanks very much for the clarification. I, along with thousands of
others, await Nina. No pressure, AL ;-)
Jim Leonard
You are. :-) For film-to-PAL transfers, the film is sped up from 24fps
to 25fps, and the audio sped up by 4% to match. Look at most PAL DVD
statistics on www.imdb.com and you'll notice that the european versions
almost always have a 4% shorter running time.
Youri Pepplinkhuizen
> you could use that cleaned up frame as a reference against the other
> frames if you had a static scene for additional noise reduction of
> the entire scene, as long as most of it was static. Food for thought.
Indeed. But how many times would the distribution system add the exact same
static noise to a pair of frames? Usually, it's dynamic over all frames.
-Youri
Chris
Donald wrote:
> So how does Chris get two captures that line up so perfectly?
> Does he capture the same broadcast simultaneously on two different
> computers?
No, he doesn't. The point is to get the same sequence transmitted at 2
different times, hopefully via the same channel. If there is any variation,
it's probably in horizontal position, and rarely in the pixel clock(s) (H
width) of the station's framesync.
If you want an easy way to try this, just look for
commercials/promos/title-sequences that air several times an hour or at the
same times every day. Also, cable channels like TLC or Discovery will air a
show in EST primetime, then again 2 to 4 hours later. You could look for the
same show on different feeds (east/west, 1,2,3), but then the settings might
not be as close.
> If not, what if they keep throwing in different commercials, logos, edits,
etc.?
Commercials get edited out (unless that's what I'm capping), so the clip only
has to match until the next commercial. That's the great thing about digital
video - you can slice and dice your caps at will.
Logos get mixed, else I remove them with LogoAway or cropping. If a logo
appears in one clip but not another, I can do a soft edge diagonal wipe or
overlay mask with the corresponding area of the no-logo clip (using msp6).
Finished broadcast programs rarely have different edits, so I've never run
into that.
Avery wrote:
> An automatic method of averaging two captures would be difficult [...]
Oh no, I'm not looking for that. Just a quick manual way to preview the
averaged-together starting frames of the two clips, and a nudge control that
bumps either of the frames left or right a pixel at a time. The result would
be no change in frame size, just the relative positions - it either eats or
pads pixels if needed. Maybe a little horizontal scaling factor could be
applied to either clip. It would be handy also to choose any frame as a start
frame, to save editing time and disk space. Then the filter provides the
position-adjusted averaged frames to the rest of the filter chain. Audio
would come from one of the files (by radio button).
It could even be a pre-processor - part of the file-open dialog - and would
provide all manner of fading/crossfading services. Let's say you want a fade
to black - don't specify a second file, and you'd be able to pick a start/end
frame. How 'bout a fade to/from grayscale? Or, get this, you specify the
SAME file twice, choosing start frames that differ by one.... Yes! Discrete
windowed temporal averaging with less motion blur!
(I also do this in MSP6 by crossfading the same clip to itself, bumping the
second instance 1 frame on the timeline)
What about specifying a STILL picture as the second file to crossfade to/from?
Any appended video segments could use these services or just be appended
"dry". Perhaps an appended file dialog could specify (with negative start
frame numbers) frames of the preceding clip(s) to be used in crossfading to
the appended clip. This provides much utility for any number of projects!
A pre-processor probably wouldn't have to be operating in RGB24 like the
filters, so you could do it in YUV for speed before it gets to the filters,
and nothing in the filter architecture has to be altered.(!)
> Another thing that popped into my mind is that averaging two captures
> only reduces the noise margin by 30%, assuming a normal distribution.
> Single-source algorithms do far better than that, with far less work.
> Does anyone know how TMPGEnc's noise reduction algorithm works?
> It works amazingly well!
I just tried it, and while it is very good, any tweak took seconds to preview
(K7 650). What was effective on background noise made people look wax-like,
so it has to use low settings for natural images. I'm not real clear on the
spatial/temporal/range sliders, though.
The 2-clip method reduces noise obviously and naturally with no artifacts on
detailed and fast moving video. What better to filter with than a darn-near
pixel-precise map of each frame? It won't completely squash the noise, but it
puts you in a better position to use the more destructive filters at lower
settings.
Youri Pepplinkhuizen
> to 25fps, and the audio sped up by 4% to match. Look at most PAL DVD
> statistics on www.imdb.com and you'll notice that the european versions
> almost always have a 4% shorter running time.
You are correct. When film is converted to PAL, the audio just undergoes a
pitch correction, apparantly. This makes me wonder why there is an option to
deinterlace PAL in the Smart Deinterlace filter, though. If I wanted to
convert PAL back to film, I could just compress the audio and change the fps
to 24 in VirtualDub, right? If this is all true, exactly what is the reason
for PAL deinterlacing in the SD filter anyway?
However, I also read somewhere that certain film material that is telecined
for PAL contains an extra field every 24 fields "to align the 24-Hz film to
a 50-Hz field rate". Ha, seems I was (sorta) right after all! :)
-Youri
Cart Wheel
viewing.
> >
> >I never heard of this before. I know NTSC material that has undergone
3:2
> >pulldown is slowed down by 0.1%, but afaik, PAL material includes
some extra
> >frame (created from different fields) somewhere in the pair.
> >
> >I could very well be wrong though.
>
> You are. :-) For film-to-PAL transfers, the film is sped up from
24fps
> to 25fps, and the audio sped up by 4% to match. Look at most PAL DVD
> statistics on www.imdb.com and you'll notice that the european
versions
> almost always have a 4% shorter running time.
Although most films are just sped up to convert from 24 to 25 fps, I've
seen a few websites that describe a "24&1" method for PAL telecining.
This apparently duplicates fields in the same fashion as NTSC
telecining, except that it is spread out over 25 frames instead of just
5. I have a hunch that this may be used for some types of animated
video. Here are a couple URLs that describe it:
http://www.aaton.com/about/preferences/24fps.html
http://www.filmlogic.com/ug/ChapterFour.html
Cart
http://www.geocities.com/lukesvideo/index.html
Si Walters
snip
> This makes me wonder why there is an option to
> deinterlace PAL in the Smart Deinterlace filter, though. If I wanted to
> convert PAL back to film, I could just compress the audio and change the
fps
> to 24 in VirtualDub, right? If this is all true, exactly what is the
reason
> for PAL deinterlacing in the SD filter anyway?
Donald's Smart Deinterlacer help file explains the need for "PAL
Deinterlacing" quite well.
regards
Simon
snip
Donald Graft
"Jim Leonard" <jleo...@vault.oldskool.org> wrote in message news:91qr69$2oi$1...@vault.oldskool.org...
> Forgive me for sounding harsh, but what's the point of using [Telecide] then?
For PAL, when the film is sped up, there are no extra frames, so the
filter outputs nice, original, progressive, full-resolution frames, irrespective
of the field phase and/or telecining method, and these latter can change
dynamically within the clip. The field phase shift and 'perverse' telecining
were the original motivation for "PAL deinterlacers" (bad name), and the
fact that the shift and telecining could change dynamically motivated the
attempts at adaptive "PAL deinterlacers". Telecide is a very
effective and reliable adaptive "PAL deinterlacer". That is what it was
written for. Once the frame deletion is implemented, however, it becomes
a *general-purpose* inverse telecine, which can adaptively operate on any
telecining method, including random ones! That is its promise and why
some of us are excited by its algorithm. All IVTCs I have seen to date used
modal algorithms that attempt to keep a fixed telecine pattern in phase with
the incoming frames. Telecide is modeless and has no fixed patterns to
overlay. It appears to be a new approach to IVTC.
For PAL with an extra frame per 25, the result
is the same but the extra frame is included, but it's unlikely to be noticed.
For 3:2 pulldown, yes, the extra frames are a problem. The filter is
immediately useful for deriving nice stills, however. And when I complete
the extra frame deletion functionality, the whole will constitute a very
effective and reliable general-purpose IVTC, IMHO.
> That's the same output as if I just captured a single field at 320x240,
> which has horrible motion quality.
Perhaps temporally for 3:2 pulldown. But spatially, it is retaining full resolution.
> Every 5th frame is duplicated, and
> things look jerky. Is the idea to code the filter now and add filter
> removal support later? Otherwise, I must be missing something, as I
> can't see a practical use of the filter.
Obviously you are not a PAL user. :-) But yes, the extra frame deletion
is in-progress. See above.
Don
Alexander Ibrahim
broadcast. You capture using analog input to a PC then reduce the noise then
output to VHS or SVHS.
Why don't you use a recording format with lower signal to noise ratio. I
record from air using a JVC SR-VS10U MiniDV recorder. I record my programs
initially using SP mode, then I edit on my PC and output to tape at LP
speed. Sometimes I capture directly from air using the IEEE ports.
The native S/N ratio of DV is around 54dB. The blacks are very black crisp
and clear and there is very little noise on the final tape. (Just what was
there on the airing, which is actually very little.)
SVHS has a maximum S/N ratio of 45dB, VHS has a S/N of 35dB with a good
recorder. LP mode on MiniDV allows you to store a 1 hour show after edits on
a 33minute tape. SNR is 54dB.
Quality is identical to SP recording, unless the tape is damaged. (If the
tape is damaged an LP recording will have a longer playback time affected.)
Just an idea...
Alexander
http://www.zenera.com
"Chris" <nos...@nospam.com> wrote in message
news:v7706.8$Gh1...@bgtnsc05-news.ops.worldnet.att.net...