Problem w/artifacts using Thalen DeShaker in Virtualdub

[brassplyer]

Mine looks like the source. Yours had corrupted the two fields. Did you do some other filtering?

No other filtering. It would appear the absence of the 16:9 is clearly a factor. I'd be satisfied if I could get it to look like that but in 16:9.

It's already 16:9. It's just not flagged as such. If you open it in an editor all you have to do is override the source AR.

Playing it in WMP it comes up looking 4:3. How do you override the source AR?

I just triedconverting your file to 16:9, it results in what appears to be the same artifacts as I'm getting originally.


What are you doing to "convert" it to 16:9? I think this is the source of your problems, not the Deshaker filter.

Saved as AVI using Mainconcept, with the 16:9 box checked. That changes the AR to where it plays 16:9 but also adds the artifacts.

[brassplyer]

I couldn't reproduce your issues with the same filters, but I'm using a different DV decoder and encoder, cedocida.

Tried it, apparently it doesn't do 16:9.

[poisondeathray]

Yes, jagabo already stated that. You fix the aspect ratio in your editor or encoder for DVD

What is your final format goal? Surely it's not DV? Are you putting it back on tape?

[brassplyer]

Yes, jagabo already stated that. You fix the aspect ratio in your editor or encoder for DVD

What is your final format goal? Surely it's not DV? Are you putting it back on tape?

Primarily DVD, but possibly also upload to the web - YouTube, Vimeo, etc. I can't discount the possibility of re-loading to DV - it *should* work.

[poisondeathray]

For DVD you set the AR in the encoder

For web you have to deinterlace, and resize to 16:9 and using 1:1 square pixels

Re-loading DV, you would interpret the footage in your editor and re-encode to 16:9 DV

Eitherway, you shouldn't encode to DV using vdub. DV is lossy and you incur extra quality loss by exporting DV then encoding again to MPEG2 for DVD , (or some other format for web). You should use a lossless format ..(e.g. huffyuv, lagarith) or uncompressed for your intermediate stages whether importing into another editor or cleaning up the footage, color correction etc

[jagabo]

Post a short clip of my DV AVI file after you converted it to 16:9. You may just have a playback problem.

[brassplyer]

For DVD you set the AR in the encoder

For web you have to deinterlace, and resize to 16:9 and using 1:1 square pixels

Isn't this generally done automatically by the various converters like Super, WinFF etc?

You should use a lossless format ..(e.g. huffyuv, lagarith) or uncompressed for your intermediate stages whether importing into another editor or cleaning up the footage, color correction etc

Trying Huffyuv and uncompressed, it looks odd too. I find a "waviness" in the video when played on WMP - even when saved without any filters. Is this normal?

[jagabo]

For DVD you set the AR in the encoder

For web you have to deinterlace, and resize to 16:9 and using 1:1 square pixels

Isn't this generally done automatically by the various converters like Super, WinFF etc?

Only if they correctly identify your source's DAR and that it's interlaced. If you give them a HuffYUV encoded file they'll have no idea what the correct DAR is. And they probably won't notice the video is interlaced. This is why you have to specify those things when you use them.

You should use a lossless format ..(e.g. huffyuv, lagarith) or uncompressed for your intermediate stages whether importing into another editor or cleaning up the footage, color correction etc

Trying Huffyuv and uncompressed, it looks odd too. I find a "waviness" in the video when played on WMP - even when saved without any filters. Is this normal?[/quote]
I still think you have a display issue, not a conversion/encoding issue. WMP will automatically deinterlace DV AVI files as it plays them. It will also adjust the frame to match the display aspect ratio of the source. These two features can lead to jagged, buzzing edges, etc.

[brassplyer]

I still think you have a display issue, not a conversion/encoding issue. WMP will automatically deinterlace DV AVI files as it plays them. It will also adjust the frame to match the display aspect ratio of the source. These two features can lead to jagged, buzzing edges, etc.

The only thing is, others have said they've seen issues with the converted video on their systems as well.

I'll make a test DVD and see how it looks.

[poisondeathray]

brassplyer- what do you mean by "waviness" ? Do you mean the interlaced combing? Do you see the macroblocks and pixellation along the edges like your 2nd video in the huffyuv playback?

jagabo - how can it be a display issue if the original 16:9 DV looks ok? I'm referring to the macroblocks and pixellation in the 2nd video he posted (test_deshake_problem.avi). I think it's his DV encoder

Here is your file, bob-deinterlaced, denoised, deshaken, reinterlaced and re-encoded back to 16:9 DV (interpreted in Premiere). See if this plays "correctly". Normally you wouldn't encode back to DV unless you were putting it back onto tape for some reason
http://www.mediafire.com/?z5jzkzmkbwm

Here is your file , bob-deinterlaced, denoised, and encoded using x264. Since it was bob-deinterlaced it is 60p; normally you would only upload 30p material to video sites like vimeo,youtube, so you would only single rate deinterlace (30p). I included the aspect ratio flag, so it's NOT 1:1 square pixels, it's like the DV which streches anamorphic for display upon playback. So you would resize it to 16:9 AR and 1:1 pixel aspect ratio, so something like 704x400 , or 840x480 (I forget which one youtube recommends) , because those sites don't support anamorphic display and they will get the AR wrong
http://www.mediafire.com/?ujdeadez5ez

[jagabo]

jagabo - how can it be a display issue if the original 16:9 DV looks ok?

I had forgetten about that.

I'm referring to the macroblocks and pixellation in the 2nd video he posted (test_deshake_problem.avi).

And I agree his second video was screwed up. But I suspect he has more than one problem. Especially since he's reporting strangeness in his huffyuv encoded video. But his reluctance to explain exactly what he sees isn't helping.

[brassplyer]

jagabo - how can it be a display issue if the original 16:9 DV looks ok?

I had forgetten about that.

I'm referring to the macroblocks and pixellation in the 2nd video he posted (test_deshake_problem.avi).

And I agree his second video was screwed up. But I suspect he has more than one problem. Especially since he's reporting strangeness in his huffyuv encoded video. But his reluctance to explain exactly what he sees isn't helping.

No reluctance, just timing.

By waviness I mean narrow horizontal waves. Hard to describe a somewhat subtle artifact more precisely than that. As I say I'm trying to create a DVD now out of test segments to see how it looks on a TV.

Btw, another reason to go back to DV is file size - clearly Huffyuv and to an even greater extent uncompressed take up far more space than DV, and further Pinnacle Studio which I'm using to create DVD's chokes on Huffyuv.

I'm hoping the solution is something besides the DV codec since - A) I paid for it B) it does 16:9 which I don't see on the freebies.

Just took a look at poison's vid file - thanks for taking the time. On WMP it displays the waviness I mentioned on the deshaken files saved to Huffyuv and uncompr. Will have to see what they look like on DVD.

[brassplyer]

If the original DV file is 16:9, shouldn't the file converted to uncompr or Huff retain a 16:9 AR?

[poisondeathray]

If the original DV file is 16:9, shouldn't the file converted to uncompr or Huff retain a 16:9 AR?

No. Avisynth and vdub assume square pixel. You change it to 16:9 in your DVD encoding software, or interpret it in your editor. I don't use pinnacle studio, so I don't know what the workflow for you would be, but most decent editors will have this functionality

Just took a look at poison's vid file - thanks for taking the time. On WMP it displays the waviness I mentioned on the deshaken files saved to Huffyuv and uncompr.

Are you saying there is "waviness" in both the files I uploaded?

[jagabo]

To AviSynth and VirtualDub pixels are dimensionless points. As Poisondeathray points out, you set the PAR or DAR in the output codec. Or in the editor you use later. Most editors have the ability to override what the editor thinks of the PAR or DAR of the source.

And what to you mean by waviness? Poisondeathray's file play fine here -- especially the h.264 encoded one.

[brassplyer]

Are you saying there is "waviness" in both the files I uploaded?

The first time I missed the x264 file. The waviness is there in the Adobe file, I don't see it in the x264 file, it looks clean.

[poisondeathray]

Are you saying there is "waviness" in both the files I uploaded?

The first time I missed the x264 file. The waviness is there in the Adobe file, I don't see it in the x264 file, it looks clean.

Then I suspect you are looking at deinterlacing artifacts. Your real time deinterlacer for the interlaced DV file is probably leaving jaggies, and I suspect that is what you mean by "waviness". Compare your original file to the "Adobe" DV file; is there any "waviness" ?

That is the reason I put the deinterlaced x264 file there; for comparison, and to narrow down if it is a decoder issue for you.

Whether or not you get "waviness" on a DVD will depend on how well your DVD player deinterlaces

Did the DV file play @ 16:9 DAR for you in WMP?

Can you try another player like VLC? you can activate deinterlacing in the options

[brassplyer]

Here is your file, bob-deinterlaced, denoised, deshaken, reinterlaced and re-encoded back to 16:9 DV (interpreted in Premiere). See if this plays "correctly". Normally you wouldn't encode back to DV unless you were putting it back onto tape for some reason
http://www.mediafire.com/?z5jzkzmkbwm

I made a test DVD using my Deshaken file, my original source and the short one you treated. What I find is that playing back on a TV, my deshaken file looks about comparable to the original, however your treated file looks cleaner than either. Most notably, there's some "fringing" around the orange objects that's quite noticeable on either of mine, that I don't see on yours.

What part of the treatment did you do within Vdub and what part was done in Adobe?

[poisondeathray]

I made a test DVD using my Deshaken file, my original source and the short one you treated. What I find is that playing back on a TV, my deshaken file looks about comparable to the original, however your treated file looks cleaner than either. Most notably, there's some "fringing" around the orange objects that's quite noticeable on either of mine, that I don't see on yours.

What part of the treatment did you do within Vdub and what part was done in Adobe?

Hold on. Are you saying the "test_deshake_problem.avi" looks comparable to the original "test.avi" when put on DVD ?????! That makes no sense at all. Are you saying those blocky edge artifacts are gone now? I thought that was the biggest issue and the whole reason for this thread ??

Or is it because you installed cedocida, and everything decodes OK now using that? But the errors are encoded into that "test_deshake_problem.avi" file, so the DVD should have those errors as well - or did you re-do the test with cedocida?

The file I uploaded looks different because I filtered it with denoisers; and probably not from the deinterlacing or reinterlacing. You shot it low light, so there is a lot of general noise, and chroma noise. Everything was done in avisynth , except for the deshaking , which was done in vdub.

The only reason I used "Adobe" is to export as DV-AVI in 16:9 for demonstration purposes to see if it's your DEcoder that is the problem. (i.e to see if your decoder "sees" those blocky edge artifacts). It's actually lower quality than if I would have just exported with huffyuv after deshaking and encoded straight to MPEG2 for DVD. There is absolutely no reason to export DV-AVI if you're going to DVD - it's lossy.

[brassplyer]

Hold on. Are you saying the "test_deshake_problem.avi" looks comparable to the original "test.avi" when put on DVD ?????! That makes no sense at all. Are you saying those blocky edge artifacts are gone now? I thought that was the biggest issue and the whole reason for this thread ??

I did another one with maybe slightly different parameters but similar to the test_deshake file. As seen when playing back letterboxed on a 4:3 CRT TV the two files look about the same as far as seat of the pants assessment of quality. The problems I've seen are with the avi files when played on the computer. I don't see specific evidence of that on the DVD but see other things - i.e. the orange fringing I mentioned, and not as generally clean looking as your treated segment.

The file I uploaded looks different because I filtered it with denoisers; and probably not from the deinterlacing or reinterlacing. You shot it low light, so there is a lot of general noise, and chroma noise. Everything was done in avisynth , except for the deshaking , which was done in vdub.

The only reason I used "Adobe" is to export as DV-AVI in 16:9 for demonstration purposes to see if it's your DEcoder that is the problem. (i.e to see if your decoder "sees" those blocky edge artifacts). It's actually lower quality than if I would have just exported with huffyuv after deshaking and encoded straight to MPEG2 for DVD. There is absolutely no reason to export DV-AVI if you're going to DVD - it's lossy.

The question is whether the "lossiness" is actually going to be discernible.

There's a very good reason for me to go back to DV, as previously mentioned - file size. I've got around a TB, maybe more in DV files much of which I need to treat and when I've got it all done decide how I want to organize them on DVD.

It's not HD - the file you did is well within "good enough" parameters. I just need to be able to replicate that and I'll be happy. My main editing app is now Vegas Pro 8, which I gather beats the pants off Pinnacle - when I saw "VST plugins" it was clear Vegas is in a different universe tan PS9. I'll only use Pinnacle for creating DVD's - it does have pretty decent menu creation options.

Which did you do first, the deshaking or the other treatment in avisynth? Can you outline specifically what you did?

[poisondeathray]

Hold on. Are you saying the "test_deshake_problem.avi" looks comparable to the original "test.avi" when put on DVD ?????! That makes no sense at all. Are you saying those blocky edge artifacts are gone now? I thought that was the biggest issue and the whole reason for this thread ??

I did another one with maybe slightly different parameters but similar to the test_deshake file. As seen when playing back letterboxed on a 4:3 CRT TV the two files look about the same as far as seat of the pants assessment of quality. The problems I've seen are with the avi files when played on the computer. I don't see specific evidence of that on the DVD but see other things - i.e. the orange fringing I mentioned, and not as generally clean looking as your treated segment.

Are you saying you don't see those horrible edge artifacts anymore? So was it the decoder or encoder? or something else? Are you using cedocida for both now? What was causing those blocky edge artifacts?

There's a very good reason for me to go back to DV, as previously mentioned - file size. I've got around a TB, maybe more in DV files much of which I need to treat and when I've got it all done decide how I want to organize them on DVD.

Ok. Fair enough, I guess that's valid I guess I should have qualified my statement "if quality is top priority"

It's not HD - the file you did is well within "good enough" parameters. I just need to be able to replicate that and I'll be happy. My main editing app is now Vegas Pro 8, which I gather beats the pants off Pinnacle - when I saw "VST plugins" it was clear Vegas is in a different universe tan PS9. I'll only use Pinnacle for creating DVD's - it does have pretty decent menu creation options.

I can tell you the script in avisynth, but not sure if it will help you if you're not willing to learn some basics about avisynth. As an alternative, you will be able to get similar denoising (or even better) with neat video ; it has a plugin for vdub and vegas and more user friendly with a GUI than avisynth filters; unfortunately, it's not free.

Most filters work better on progressive content, but deshaker, neat video, MCTD and FFT3Dfilter in avisynth have an interlaced mode (none of the mentioned filters are as effective as when used on progresive) .

You do have sharp ringing (haloing) around the orange areas (and other objects) , perhaps your camera was set too high for artifical sharpening or detail. I did use a dehaloing script to help with that, and that probably helped with the "orange fringing" too.

So I imported this .avs script into vdub to use deshaker

AVISource("test.avi")
ConvertToYV12(interlaced=true)
TempGaussMC_Beta1(sharp=0)
Dehalo_Alpha(darkstr=0.9,brightstr=0.9)
MCTemporalDenoise(settings="high")

Then I used deshaker , exported a lossless lagarith 60p file, and re-interlaced it

AVISource("deshaken.avi")
AssumeBFF() 
SeparateFields() 
SelectEvery(4,0,3) 
Weave()

[brassplyer]

Are you saying you don't see those horrible edge artifacts anymore? So was it the decoder or encoder? or something else? Are you using cedocida for both now? What was causing those blocky edge artifacts?

Not sure about the artifacts. Going through the DVD conversion may simply be masking it somehow, or ....not sure. I tried cedocida, but went back to Mainconcept.

Ok. Fair enough, I guess that's valid I guess I should have qualified my statement "if quality is top priority"

Certainly I'm all for quality but under the circumstances it's likely to be viewed, I don't think the difference is going to be visible.

I can tell you the script in avisynth, but not sure if it will help you if you're not willing to learn some basics about avisynth.

Absolutely willing.

So I imported this .avs script into vdub to use deshaker

AVISource("test.avi")
ConvertToYV12(interlaced=true)
TempGaussMC_Beta1(sharp=0)
Dehalo_Alpha(darkstr=0.9,brightstr=0.9)
MCTemporalDenoise(settings="high")

Then I used deshaker , exported a lossless lagarith 60p file, and re-interlaced it

AVISource("deshaken.avi")
AssumeBFF() 
SeparateFields() 
SelectEvery(4,0,3) 
Weave()

So let the learning begin - are you saying Avisynth is used within Vdub?

[poisondeathray]

Yes, you can use avisynth within vdub. Avisynth is a frameserver

http://avisynth.org/mediawiki/Main_Page
http://avisynth.org/mediawiki/First_script

You will have difficulties using DV-AVI as an intermediate with the workflow I used, because I bob-deinterlaced to a 60p intermediate ; DV-AVI doesn't support that. So if you have limited HDD space, you might want to explore alternate methods or filters that work with interlaced content

Also it's sometimes better to use lossless intermediates for stages, because for some 2pass encodes and filters (like deshaker) , instead of using the heavy filters twice, you only have to do it once. It is much faster. I actually used that 1st script to encode a lossless lagarith intermediate first (with the denoise filters), then deshaker export to lagarith , then reinterlaced that deshaken .avi with the 2nd .avs script. You could frameserve out of vdub directly to your DVD encoder without using intermediates at all, there are guides on how to do this if you search

[jagabo]

TempGausMC_beta1() is a very slow filter, a few frames per second. If you have a lot of footage you will want to run TempGauss and save as an intermediate file, then do the two deshaker passes in VirtualDub. Or you can deshake in AviSynth with DePan() or maybe Stab().

I noticed I left out the deshaking line in the script I posted a few days ago. It should have looked more like this:

import("C:\Program Files\AviSynth 2.5\plugins\TempGaussMC_beta1.avs")

AVISource("test.avi")
ConvertToYV12(interlaced=true)
TempGaussMC_beta1()
#
#deshake goes here
#
SeparateFields()
SelectEvery(4,0,3)
Weave()

[Soopafresh]

Get yourself a Hague MMC Stabilizer (steadicam) device for $120 and you'll have more stable shots to begin with. You'll definitely reduce the times you'll need to use Deshaker.

http://www.vimeo.com/8800412

http://vimeo.com/804381

[brassplyer]

You do have sharp ringing (haloing) around the orange areas (and other objects) , perhaps your camera was set too high for artifical sharpening or detail. I did use a dehaloing script to help with that, and that probably helped with the "orange fringing" too.

So I imported this .avs script into vdub to use deshaker

AVISource("test.avi")
ConvertToYV12(interlaced=true)
TempGaussMC_Beta1(sharp=0)
Dehalo_Alpha(darkstr=0.9,brightstr=0.9)
MCTemporalDenoise(settings="high")

Okay, picking this apart -

ConvertToYV12(interlaced=true)

I read some on YV12, can't say it's crystal clear to me - if it's being converting to YV12, what was it before and why are you converting to it?

TempGaussMC_Beta1(sharp=0)
Dehalo_Alpha(darkstr=0.9,brightstr=0.9)
MCTemporalDenoise(settings="high")

These are 3 specific filters that need to be found/downloaded?

As of this point the file is deinterlaced? Where was it deinterlaced?

Then I used deshaker , exported a lossless lagarith 60p file, and re-interlaced it

60P meaning 60fps because of how it was deinterlaced?

And this below is to reinterlace?

AVISource("deshaken.avi")
AssumeBFF() 
SeparateFields() 
SelectEvery(4,0,3) 
Weave()

[jagabo]

ConvertToYV12(interlaced=true)

I read some on YV12, can't say it's crystal clear to me - if it's being converting to YV12, what was it before and why are you converting to it?

DV decoders may output YV12, YUY2, or RGB. Use Info() to find out what comes out of your decoder. Some of the other filters require YV12.

TempGaussMC_Beta1(sharp=0)
Dehalo_Alpha(darkstr=0.9,brightstr=0.9)
MCTemporalDenoise(settings="high")

These are 3 specific filters that need to be found/downloaded?

Yes. But some are actually scripts and rely on several other filters. You'll have to read the docs to find out what filters you need to download and install.

As of this point the file is deinterlaced? Where was it deinterlaced?

TempGaussMC_beta1() converts 30 interlaced frames into 60 progressive frames.

And this below is to reinterlace?

AVISource("deshaken.avi")
AssumeBFF() 
SeparateFields() 
SelectEvery(4,0,3) 
Weave()

Yes.

[poisondeathray]

I read some on YV12, can't say it's crystal clear to me - if it's being converting to YV12, what was it before and why are you converting to it?

It was YUY2 before, with 4:1:1 chroma sampling. Some filters only work in YV12 or RGB. Deshaker works in RGB. When it comes out of deshaker and vdub, it will be RGB. Colorspace conversions from RGB=>YV12 are lossy , but from YV12 or YUY2 to RGB are not. Eitherway, it's a good practice to minimize colorspace conversions if possible.

These are 3 specific filters that need to be found/downloaded?

Yes, just search google

As of this point the file is deinterlaced? Where was it deinterlaced?

Right after TempGaussMC_Beta1() it was bob-deinterlaced to 60p (double rate, not single rate 30p)

60P meaning 60fps because of how it was deinterlaced?

And this below is to reinterlace?

Yes. It will be RGB now, exported from vdub. Most NLE's will use RGB anyway. If your editor doesn't accept .avs scripts or frameserving, you still have to encode that last step to something, like huffyuv for importing. Or you could use that script to encode directly to your MPEG2 encoder for DVD, like HCenc

This is just an example of one way of doing it. There are many many other methods, filters, settings you could choose. Please don't think this is the only way you can do it. There are many more simpler workflows, and some even more convoluted

EDIT: jagabo beat me to the punch

[jagabo]

Colorspace conversions from RGB=>YV12 are lossy , but from YV12 or YUY2 to RGB are not.

YV12/YUY2 to RGB is lossy too. You don't lose resolution but you lose accuracy. Not every integer YUV value has a unique integer RGB value , and vice versa. Though the losses usually aren't really visible.

[poisondeathray]

Colorspace conversions from RGB=>YV12 are lossy , but from YV12 or YUY2 to RGB are not.

YV12/YUY2 to RGB is lossy too. You don't lose resolution but you lose accuracy. Not every integer YUV value has a unique integer RGB value , and vice versa. Though the losses usually aren't really visible.

That make sense; I guess there are different equations you could use and rounding errors too. e.g. different renders have better or worse conversions to RGB when they display the YV12 source, and some have poor chroma upsampling , like flash.

Would you agree RGB=>YV12 is definitely worse than YV12=>RGB when you "pixel peep"

There are other methods of lossless colorspace conversions, like sheervideo. I'm not sure how they do it though, they have a bunch of equations on their website that I don't understand.