A problem for video experts

[jagabo]

ok I tried the "Edge-directed Interpolate". the motions look better, but the quality became lower. so I have to choose between better quality to normal motion? thanks.

If by this you mean sharp, nearly horizontal, edges look jagged, vs looking blurry before, yes, those are your choices in VirtualDub.

Another easy solution in VirtualDub is to use the built in Deinterlace filter with the Discard Field 1 or Discard Field 2 options. Then resize to desired final size. But this will give even more jagged edges and less resolution overall

You can get slightly better (than Smart Deinterlace) results in AviSynth with TDeint() or Yadif(). And even better results with TempGauss_beta1(). But that last function is very slow.

Basically, the issue you're facing is that it's not possible to perfectly create one whole image from two half images.

[Yonidan]

well. I the built in Deinterlace filter in VirtualDub, the size has changed. how can I change it to normal size?

[jagabo]

That's why I said you have to resize aftrwards. Add the resize filter after the deinterlace filter.

[2Bdecided]

You can get slightly better (than Smart Deinterlace) results in AviSynth with TDeint() or Yadif(). And even better results with TempGauss_beta1(). But that last function is very slow.

Worth it though. Don't forget mcbob - similarly slow and good.

I don't know if Yonidan wants 25fps or 50fps output - I'd take the latter, as long as the PC can play it.

Cheers,
David.

[Yonidan]

is the difference between the Smart Deinterlace to the AviSynth and the others, worth the trying ?

[poisondeathray]

is the difference between the Smart Deinterlace to the AviSynth and the others, worth the trying ?

If you are not happy with your results, it's worth trying them out on a small sample in my opinion - it should only take a few minutes for each test so you are not wasting that much time.

I use yadif a lot, it's usually better than vdub's built-in deinterlace function

[Yonidan]

and the yadif comes with AviSynth?

[jagabo]

Here is one comparison of several deinterlacing filters:
https://forum.videohelp.com/topic354397-30.html#1877202

Here is the same frame from Smart Deinterlace in VirtualDub:


Here is the frame from MCBob():


This is only one frame of one video so don't make any final judgements on the quality of each. But it should give you a starting point. Keep in mind that TempGauss_beta1() and MCBob() are very slow, around 1/50th the speed of the others.

[Yonidan]

they are both look good, tho I prefer the first one, of the Smart Deinterlace in VirtualDub. I don't get the same result with the Smart Deinterlace...why?

the MCBob is a plug of TempGauss_beta1? what is the slow thing you said? the process of the de-interlacing of TempGauss_beta1?

[jagabo]

I don't get the same result with the Smart Deinterlace...why?

I don't know why. Every video is different. Did you apply any other filters before deinterlacing? Generally, you should deinterlace first. If you post a short sample of your source I'll take a look at it.

the MCBob is a plug of TempGauss_beta1? what is the slow thing you said? the process of the de-interlacing of TempGauss_beta1?

TempGauss_beta1() and MCBob() are two motion adaptive deinterlacing filters for AviSynth. Both are very slow. But the yield the best results, in my experience.

[Yonidan]

I don't get the same result with the Smart Deinterlace...why?

I don't know why. Every video is different. Did you apply any other filters before deinterlacing? Generally, you should deinterlace first. If you post a short sample of your source I'll take a look at it.

you mean, to post the original video, before the deinterlacing?

the MCBob is a plug of TempGauss_beta1? what is the slow thing you said? the process of the de-interlacing of TempGauss_beta1?

TempGauss_beta1() and MCBob() are two motion adaptive deinterlacing filters for AviSynth. Both are very slow. But the yield the best results, in my experience.

I didn't understand what speed do you mean..you mean the de-interlacing is slow?

[jagabo]

I didn't understand what speed do you mean..you mean the de-interlacing is slow?

Yes. A video that takes 1 hour to detinterlace with VirtualDub and Smart Deinterlace will take ~50 hours with TempGauss_beta1() or MCBob().

[Yonidan]

wow
so a chapter of 45mins, will take so many hours?

btw, do you want to see a frame of the same video I posted previously, but before the de-interlacing?

[jagabo]

I need a few seconds of the video (with motion) to deinterlace if you want me to analyze it. You can use a program like DgIndex to select a small section and save it as an elementary stream.

[Yonidan]

can you tell me the steps in "DgIndex" please? step by step..if you don't mind
it's a little complicated..

[jagabo]

File -> Open
Move slider to start of section, press [ button
Move slider to end of section, press ] button
File -> Save Project and Demux Video

upload M2V file.

[Yonidan]

ok I did it, but got a "d2v" file, without sound.
hope it's ok.

channel%20name_20081023172900.demuxed.m2v

[jagabo]

Smart Deinterlace (Edge directed interpolate) did just fine with that video:

smart.avi

I didn't resize, crop, or change the aspect ratio so you can see the results pixel-for-pixel.

What exactly are you unhappy with? And why did you like Smart Deinterlace better than MCBob in the images I posted earlier? Notice how the line between the fender and hood of the card is much smoother with MCBob.

[poisondeathray]

What exactly are you unhappy with? And why did you like Smart Deinterlace better than MCBob in the images I posted earlier? Notice how the line between the fender and hood of the card is much smoother with MCBob.

Exactly. Also notice the horizontal artifacts near the bottom and side of the door, and the headlight. MCBob did much better for that sample IMO, at least in that frame

Nice comparisons in that link jagabo.

[Yonidan]

the smart deinterlace with Edge directed interpolate yielded good results of motion, but the picture was filthy.
maybe there are more settings to be changed? I attached a pitcure of my smart deinterlace settings.
just check it..if everything is ok there.

secondly, I'd like to try Avisynth and mcbob, but you say that the process is very slow.
I want to record a tv series, every chapter of that series lasts about 45 mins.
how much time will it take to de-interlace such video with mcbob?

[jagabo]

Other than edge-directed interpolote I used the default settings in Smart Deinterlace -- the same settings you have. Are you saying the sample I uploaded was filthy?

[Yonidan]

well it's not noticeable in the part of the video I gave you, but when I watch the whole video (after edge-directed interpolote deinterlacing), there's a section that looks bad.

[Yonidan]

secondly, I'd like to try Avisynth and mcbob, but you say that the process is very slow.
I want to record a tv series, every chapter of that series lasts about 45 mins.
how much time will it take to de-interlace such video with mcbob?


jagabo, what do you say about this? (mcbob)
thanks again.

[jagabo]

well it's not noticeable in the part of the video I gave you, but when I watch the whole video (after edge-directed interpolote deinterlacing), there's a section that looks bad.

You should post a bit of the bad part.

secondly, I'd like to try Avisynth and mcbob, but you say that the process is very slow.
I want to record a tv series, every chapter of that series lasts about 45 mins.
how much time will it take to de-interlace such video with mcbob?

I get about 1 frame per second on my computer. So a 45 minute PAL video would take 18 or 19 hours.

To try MCBob() you will need to install AviSynth. You have to track down a bunch of other AviSynth filters and install them:

MVTools, preferably v1.4.13 (or newer)
MaskTools v2.0
EEDI2
RemoveGrain/Repair package
ReduceFlicker (if temp-NR for ME is used)
MedianBlur by tsp

Here is a good place to start:

http://avisynth.org/warpenterprises/

You'll need the MCBob.AVS script from here:

http://forum.doom9.org/showthread.php?p=988224#post988224

Then you need to create a .D2V file with DgIndex and an AVS script like:

import("path\to\MCBob.avs")
MPEG2Source("filename.d2v")
MCBob()

Finally open that in VirtaulDub and encode.

[Yonidan]

when you said 18-19 hours, you talked about mcbob?
well, if it will take so many hours, then I'll give up.

[jagabo]

Of course I was talking about MCBob(). That's what you were asking about.

[Yonidan]

well, after all I decided to use the smart deinterlace with blend, but I've noticed something strange: an mpeg-2 video of 28mb, becames a 20.8mb video, after the procedure in vitual dub (deinterlacing and compressing to xvid).
a video of 27.7mb (the video which I sent you to analyze) becames a 16.6mb.
I noticed that in the first video (the 20.8 mb one) the motions look ok.
the 16.6mb one, has the problem of motions, as you see. why does the difference between the both videos exist?

excuse me for my english,

[Yonidan]

jagabo or somebody?

[poisondeathray]

Can you clarify your question? Is it in regards to the size or artifacts?

The size difference maybe from settings used from the encoder (if you are using 1-pass quality vs. 2pass for example), type of movie (complex scenes will use more bitrate), deinterlace used (e.g. edge will "use" more bitrate than "blend" mode because of more detail/sharpness). Also some of the artifacting maybe from using XviD with too low a bitrate (too high a quantizer)

So did you use different settings? Perhaps there was a field order change?

TempGauss_beta1() will give a clean image with fewer artifacts (therefore "require" less bitrate), but is very slow (~10-15x slower)

[vhelp]

This source looks like it was run through a temporal filtered by the OP.

Looking at the original .m2v source (and playing it in media player) I could see major leftover temporal pixels, obviously from the previous/forward frames that were processed through some temporal noise reduction filter.

Pershaps that is the issue with some of those bad frames you mentioned.

-vhelp 4917