Interlaced HD to DVD AGAIN - some test renders

NickHope wrote on 4/29/2011, 12:04 PM

Apologies for yet another thread on this subject. I could have put this here, here, or here, but eventually decided it was better to start a new one. However those threads provide important background to this one.

I set up a minute's project of various 60i HD footage and converted it to SD NTSC MPEG-2 in a variety of ways. I have put 4 of my renders into a ready-to-burn DVD project which you can download here (193 MB). Please burn the 2 folders to a DVD and let me know any reactions/preferences you have to it, and please tell us what gear you watched it on (CRT, LCD, DVD software etc.).

The project contains a Belle Nuit test card, some of my own HDV footage and parts of amendegw's hula dancer doll clip, stringer's driving clip, John Meyer's ballet dancer clip and craftech's stage show clip. I started with Musicvid's web video project and it morphed into this. Thanks to everyone who allowed me to use and upload footage.

Settings:
Sony Vegas Pro 10.0c.
Full-resolution rendering quality: "Best".
Deinterlace method: "Interpolate".
Color Curves and Levels applied to video events.
Events conformed to 16-235 (broadcast-legal) levels.
"Reduce interlace flicker" switch: OFF.
No blur or sharpen applied in Vegas Pro.
Debugmode Frameserver 2.10
AviSynth 2.5.8 single-threaded used in tests 2 and 3.
All encoded in CCE Basic at 8000 kbps 1-pass CBR, upper field first.

Test 1 - Rendering time 309 seconds (100%)
Demonstrates Vegas Pro resizing without blurring or sharpening.
HD clips in NTSC widescreen project, modified with upper field first and PAR 1.1852.
Frameserved in RGB24.

Test 2 - Rendering time 431 seconds (139%)
Demonstrates AviSynth "IResize" function developed by Gavino, IanB and others on the doom9 forum.
Script does low-pass filter "automatically" during resizing to reduce twitter/shimmer/aliasing etc..
HD clips in 1080-60i project.
Frameserved in YUY2 (slightly faster than RGB24).
AviSynth Script:

source=AviSource("d:\fs.avi").ColorYUV(levels="TV->PC").AssumeTFF  #Expands levels if frameserved in YUY2

IResize(source,720,480)

function IResize(clip Clip, int NewWidth, int NewHeight) {

  Clip

  SeparateFields() 

  Shift=(GetParity() ? -0.25 : 0.25) * (Height()/Float(NewHeight/2)-1.0)

  E  = SelectEven().Spline36resize(NewWidth, NewHeight/2, 0,    Shift)

  O  = SelectOdd( ).Spline36resize(NewWidth, NewHeight/2, 0,   -Shift)

  Ec = SelectEven().Spline36Resize(NewWidth, NewHeight/2, 0,  2*Shift)

  Oc = SelectOdd( ).Spline36Resize(NewWidth, NewHeight/2, 0, -2*shift)

  Interleave(E, O)

  IsYV12() ? MergeChroma(Interleave(Ec, Oc)) : Last

  Weave()

}

Test 3 - Rendering time 1504 secs (486% - multi-threaded would be much faster)
Demonstrates more advanced method adapted from a suggestion by Didée on the doom9 forum.
HD clips in 1080-60i project.
Frameserved in YUY2 (RGB not supported by script).
High quality Bob using TDeint.
Resizing by sequential bicubic passes.
Smoothing with QTGMC.
Sequential blur and sharpen filters.
AviSynth Script:

AviSource("d:\fs.avi")

ColorYUV(levels="TV->PC") #Expands levels if frameserved in YUY2

AssumeTFF

TDeint(mode=1)

bicubicresize(1440,960)

bicubicresize(720,960,-.8,.6)

p1 = bicubicresize(720,480,-.8,.6)

p2 = p1.QTGMC(TR0=1,TR1=1,TR2=2,InputType=1)

p2.blur(0,1).sharpen(0,.51).blur(0,1).sharpen(0,.85)

i1 = assumetff().separatefields().selectevery(4,0,3).weave()

# p1 # straight to 60p

# p2 # plus more calmed

  i1 # re-interlaced p2

return(last)

Test 4 - Total Rendering time 1822 secs (589%)
Demonstrates down-converting of HDV to DV in my Sony Z1P camera.
HD clips in 1080-60i project.
Render to HDV (quality 31)
Down-convert to DV in Sony Z1P camera.

Back to post

Comments

NickHope wrote on 7/20/2011, 12:05 PM

Geoff, I tried feeding some 1080-50i HDV into my 7/7/2011 script and changing 480 in line 14 to 576. In each of these 3 cases, TMPGenc 4 Xpress had no problem loading the file into a PAL DVD template:

1. 720x576 without borders.
i.e. Spline36Resize(720,height)

2. 720x576 with borders at the sides.
i.e. Spline36Resize(704,height).addborders(8,0,8,0)

3. 704x576 without borders.
i.e. Spline36Resize(704,height)

I don't think padding 560 to 576 by adding borders at the top and bottom is a good idea.

If you still have trouble could you post the whole script that TMPGenc will not accept?

diverG wrote on 7/21/2011, 1:27 PM

Nick, Virtualdub was quite happy with the script but not TMPEG. Tracked the problem down to avisynth.Whilst my mod to the script worked there was no real logic in what I did. I was using v2585. 'Set' (doom9) has released V2585 MT. Basically a new avisynth.dll and Mt plugin. Dropped these in and TMPEG accepted the file. Problem resolved.

Also found a workflow under /Anton' Edious (green valley) rendering via TMPEG. The render was slow but good result.

NickHope wrote on 7/22/2011, 12:17 AM

If you're going to use SET's MT build, I would go for his very latest build of 2.6, released 3 days ago. that's the one I'll be trying first when my quadcore is back in action.

diverG wrote on 7/22/2011, 2:29 AM

I did but still had same problem. I'll give it a week or two and try again. Trying to put off moving to quad for a while. I'm retired now and the long rendering times with 2 cores help heat the office in winter.

On a lighter note shifting the cat off the computer (remove lagging) drops the core temp by 3degC.

NickHope wrote on 7/22/2011, 2:58 AM

If you have less than 4 cores then I would probably forget the MT stuff and just use the standard single-threaded version of 2.5.8 or the recent 2.6 Alpha 3. Renders are slow but they're more likely to be stable and there's always overnight if you're not in a hurry.

NickHope wrote on 12/5/2011, 5:53 AM

I was contacted by Nicolas Robidoux with an explanation of his Robidoux spline filter that we discussed above (the parameters in the bicubic resampling that AviSynth's "Bob" uses). He says this:

Indeed, it is not very different from Mitchell-Netravali.

craftech wrote on 12/5/2011, 7:19 AM

Thanks for doing that Nick.

I guess that confirms what you suspected all along.

Regards,

John

craftech wrote on 2/2/2012, 11:25 AM

I finally got around to trying Johnmeyer's most recently posted script on a current project I am working on. To avoid confusion, it is this script:



#Re-size function

#Serve out of Vegas using RGB24, and use 2.5.8.5 MT AVISynth (8/16/2009 version)



SetMTMode(5,0)

source=AVISource("e:\frameserver.avi").assumetff()

SetMTMode(2)



colorcorrected=ConvertToYUY2(source,interlaced=true, matrix="PC.601")



IResize(colorcorrected,704,480)

#IResize(source,720,480)



# Use this level shift, if needed with MainConcept MPEG-2 encoder

#Levels(16, 1, 235, 0, 255, coring=false)



function IResize(clip Clip, int NewWidth, int NewHeight) {

  Clip

  SeparateFields() 

  Shift=(GetParity() ? -0.25 : 0.25) * (Height()/Float(NewHeight/2)-1.0)

  E  = SelectEven().Spline36resize(NewWidth, NewHeight/2, 0,    Shift)

  O  = SelectOdd( ).Spline36resize(NewWidth, NewHeight/2, 0,   -Shift)

  Ec = SelectEven().Spline36Resize(NewWidth, NewHeight/2, 0,  2*Shift)

  Oc = SelectOdd( ).Spline36Resize(NewWidth, NewHeight/2, 0, -2*shift)

  Interleave(E, O)

  IsYV12() ? MergeChroma(Interleave(Ec, Oc)) : Last



  #Next line is optional

  sharpen(0,.5) # Adjust 2nd value between 0.5 and 1.0 to taste



  Weave()

}

The subject was a staged reading of a new musical in a tiny theater so there was some pretty harsh theatrical lighting. Pretty typical actually.

The color space issues it was supposed to address were actually addressed pretty well. The issue I had (which is unacceptable for me anyway) was a pronounced amount of ringing.

I frameserved the script to Procoder 3 which is how I usually do it. The only problem I ran into was that Procoder 3 didn't allow me to change the target to 704 x 480. I had to use 720 x 480. But I don't think that would have caused the "ringing".

But I did finally try it and thanks for posting it last July John.

John

johnmeyer wrote on 2/2/2012, 11:52 AM

The issue I had (which is unacceptable for me anyway) was a pronounced amount of ringing.I didn't write that script. It was copied, more or less intact, from a post in doom9.org. Many of those who post in that forum seem to like their video sharpened. You will therefore find in the script a "sharpen(0,.5)" call at the end of the IResize function.

Delete the line in the code that contains that call, and I think you'll like the results a whole lot better.

NickHope wrote on 2/2/2012, 10:53 PM

Without the sharpen, and with a 720 width, that is exactly the same script as "test 2" in the first post of this thread.

craftech wrote on 2/3/2012, 5:37 AM

It was fun trying it on a larger project anyway. I thought the Test 2 modified version was yours John.

You may be right about the sharpening in the script causing the "ringing" (or exaggerating it).

The quest for the perfect Interlaced HD to SD script goes on. Really love that the thread has generated such interest. 155 posts and counting.

John

NickHope wrote on 10/7/2013, 4:20 AM

There is useful discussion in tjos recent thread on doom9. I haven't absorbed it thoroughly yet but will do before my next HDV>DVD project.

farss wrote on 10/7/2013, 6:31 AM

I did my best to try to follow that thread and my head hurts.
I think if I had the work and hence the money I'd part with $2K and buy a Teranex box.

One pretty simple solution I have tried more by laziness than anything else that seemed to produce good results was as follows:

Shoot 50i.
Use YADIF to de-interlace in the HD project, scale and render to 50i SD.

That very likely got around all the problems with aliasing and twitter if I'd shot 1080p and tried to create a SD DVD from that.

Bob.

johnmeyer wrote on 10/7/2013, 12:22 PM

There is useful discussion in tjos recent thread on doom9. I haven't absorbed it thoroughly yet but will do before my next HDV>DVD project.I looked at that thread when it first appeared a few days ago, and just now skimmed it again. It looks to me to cover pretty much the same ground as we all covered in this thread awhile ago. It all comes down to whether you want to take the extra step of creating an estimated version of one of the fields prior to doing a re-size, or whether you use some approximation to that process. We all know that you cannot re-size interlaced footage without at least separating the fields and operating on one field at a time, and the only real question is what are the artifacts and trade-offs involved in all the various "tricks" needed to get all fields to line up, at least temporarily, both spatially and temporally, before performing the re-size operation.

Christian de Godzinsky wrote on 10/7/2013, 12:34 PM

Hi,

HD to SD conversion is an often recurring activity among many users, including me. Do we really need to go through all this acrobatics to get decent results doing downscaling? After all, we ARE using a PRO application, aren't we? Why cannot VP produce professional results directly with its built in tools? Is this a licensing problem? Don't people at SCS consider this as an very important feature?

Don't get me wrong - I'm very grateful to all of you that have gone extreme in solving this quality issue, it has helped me a lot. But getting decent results adds lots to the workload...

I'm puzzled to see the pristine quality coming out of Edius conversions (HD2SD) - without any extra acrobatics!!! What have these guys done that SCS has not been able to solve?

Christian

TeetimeNC wrote on 10/9/2013, 11:47 AM

>I'm puzzled to see the pristine quality coming out of Edius conversions (HD2SD) - without any extra acrobatics!!!

I'm the same. Also, Adobe's Premiere now uses Lanczos for downscaling when using GPU renders. I hope SCS will give higher quality downscaling some priority.

/jerry

Arthur.S wrote on 10/9/2013, 12:57 PM

Having just tried to cut corners, and downscaled from HDV using a Vegas 2 pass variable setting, I was astonished how poor it was compared to my recent workflow of frame serving to TMPGEnc VMW. My "corner cut" has meant I've had to do it all again. :-(
It does seem pathetic that a so called "Pro" application is so far behind a cheapo programme in doing what to most of us here would appear to be a "bread and butter" operation.

MikeLV wrote on 3/8/2014, 6:24 PM

I would be willing to pay someone here who understands all that was researched and concluded to make a step by step Youtube tutorial for achieving the best quality DVD video when the source footage is 1080i. It would take me more $$ worth of time to read, comprehend and implement what was discussed here rather than just pay someone a fee to create a video tutorial of the process. Or maybe it's time for me to take another look at Premiere!

videoITguy wrote on 3/8/2014, 6:39 PM

How much $? Do you know what my going NLE rate is working an independent business of an NLE suite and render farm keeping 7 render stations working 24hrs a day? What is your idea of time worth?

MikeLV wrote on 3/8/2014, 6:46 PM

No, I don't know what your going NLE rate is and I'm not asking for you or anyone else to do any editing, or work on any projects for me. I was just putting the idea out there that if someone wants to create a tutorial video like musicvid did for Handbrake, I'd be more than willing to contribute $ for the time spent to make that tutorial video. So when another novice like myself comes on the forum and asks the gurus around here the best way to go from AVCHD to DVD, all you have to do is direct them to that video tutorial.

MikeLV wrote on 3/8/2014, 6:51 PM

To add to all this confusion, I just came across this:

http://www.dvxuser.com/V6/archive/index.php/t-310268.html

which states that the solution to get better quality when resizing is to simply set the project properties to "Best." The OP then asked, "So all this stuff about needing some sort of crazy workflow to use Lanczos instead of bicubic resizing, how all NLEs have poor resizing, etc, doesn't apply to Vegas? Is Vegas really better than all the others??? Because this seems like it used to be a pretty well discussed topic."

By "pretty well discussed topic," it seems obvious to me, he was referring to this very thread......

And another thread reply by John Rofrano:

http://forums.creativecow.net/thread/24/937947

"Make sure that you render using BEST mode. BEST will use a resizing algorithm that will get better results than GOOD. Most render templates use the settings from your project so just change the template to use BEST (press Customize Template and look on the Project tab under Video render quality)"

John_Cline wrote on 3/8/2014, 11:45 PM

I have gotten perfectly acceptable results using Vegas with the "best" render setting and "interpolate" as the deinterlace method. Sometimes I'll get into Virtual Dub and use Lanczos3 resizing, but I don't see where it is appreciably better than Vegas' Precise Bicubic used in the "Best" mode, both methods can produce ringing artifacts on sharp edges.

musicvid10 wrote on 3/9/2014, 10:09 AM

One thing I have tried a couple of times, is making a lossless render in Handbrake, taking advantage of Lanczos and Decomb, and rendering that for Architect in Vegas. It's time consuming, but the results from a couple of short projects look promising.

FYI, the lossless setting in Handbrake is RF=0, and I use RF=1 to save a little space. There is not going to be a packaged tutorial on this, because the time collateral is too great; but for that occasional "money" project, is an alternative a few may want to consider.

1 … 4 5 6 7

of 7