HD to SD Challenge

amendegw wrote on 11/24/2010, 1:24 PM

All,

There's been much discussion recently about how to downrez HD source for SD DVD production. I propose a challenge (I'd call it a contest, but there's no prizes).

I shot a short, 10 sec clip under (what I thought) were the worst conditions for downrezzing HD to SD: a stationary figure with a a moving, high contrast, fine detailed background. The clip is 1920x1080 60i AVCHD. The Challenge is to get the best balance of sharpness while eliminating the flicker in the background.

I tried to downrez this clip and got good sharpness, but horrible flicker/moire patterns by adding the Sharpen FX = zero. Adding a Vertical Gaussian Blur of .002 minimized the flicker/moire, but caused too much loss of resolution in the central figure. I uprezzed the results to 720p and sent them to YouTube:

So, here's the rules of the Challenge:

1) Download and start with the clip in this link: AVCHD.zip
2) Render the clip to an NTSC 720x480 Widescreen format for import to DVDA.
3) All work must be done within Vegas.
4) Render to an intermediate format is okay.
5) Masking and/or keyframing are not allowed.
6) The objective is to get the best resolution while eliminating the flicker.
7) If you are tempted to post, "Can't be done" or "start with a different format" or something similar, please don't post a reply.
8) Uprez the results of your efforts to the web and post the link in this thread (YouTube is nice because it can be embedded, but any web link is okay).
9) Tell us how you accomplished the results - Workflow, FX's applied, Project Settings, Render settings, etc.

Any takers?
...Jerry

Back to post

Comments

amendegw wrote on 11/27/2010, 10:19 AM

All,

I've been taking everyone's input. Thanks to all!! Going thru trial-and-error. Not fully understanding what I was doing, trying anyway, and I may have come up with a method that, I think, produces very good results - and everything is done within Vegas.

First, check out the following render and see if you agree that the results are "pretty damn good". Lags.zip

If you don't agree that this is a good render, then there's no need to go to the next step. Which is: check out the following procedure and see if you can duplicate my results. Let me know if the procedure is unclear or incomplete.

Note: If you get to step 6-a-ii and do not find the Lagarith lossless codec - you can find it here:

1) Create a Sony Vegas project my matching the project properties to your HD clip(s) – in this case the AVCHD.zip clip.
2) Make sure the Project “Rendering Quality=Best” and the “Deinterlace Method = Interpolate”
3) Drag the HD Clip to the timeline.
4) Right-Click the Video Events(s) to set Properties->Reduce Interlace Flicker
5) Add a Sharpen FX set to 000. This is optional - it may produce excessive moire/flicker in the final render
6) File->Render-As (this will resize the project to 720x480 NTSC Widescreen)
....a. Video for Windows (*.avi)
....b.Template = “NTSC Widescreen” – but changes must be made:
........i.Click Custom
.......ii.Select “Lagarith Loseless codec” from the “Video Format” dropdown.
.......iii. Save this template as “Lagarith 720x480 NTSC Widescreen - Interlaced”
.......iv. Click “OK”
....c. Make sure “Stretch Video to Fill Output Frame (do not Letterbox)” is checked.
;;;;d. Choose a filename (with an AVI extension) to render your output and click “Save”
7) Close this project and start a new Vegas Project. Match the project properties to the AVI file you just created.
8) Drag the AVI Render you just created to the Vegas Timeline.
9) Right-Click the Video Event to set Properties->Reduce Interlace Flicker. Add Sharpen FX = 0.000
10) File->Render As (this will convert the format from AVI to MPG)
....a. Save as Type: MainConcept MPEG-2
....b. Template: DVD Architect NTSC Widescreen video stream
....c. Name your File name with an MPG extension.
....d. “Stretch Video to Fill Output Frame (do not Letterbox)” should not be checked.
....e. Click “Save”
11) File-Render As Dolby Digital AC-3 and render the auto with the same name as the video, but with an ACS extension
12) Import the newly rendered files to DVD Architect and create your Standard Definition DVD.

...Jerry

Edit: Re-tested this procedure - changes are in italics.
Edit2: Added reference URL to the Lagarith Codec
Edit3 Added text to 5) - Sharpen s/b optional

David Newman wrote on 11/27/2010, 12:04 PM

Jerry,

Regarding using CineForm, it would help you a lot to use HDlink (in NeoHD) over NeoScene, as the "magic" is the scaling operation and not the codec. NeoScene doesn't include that scaler. The results solely depend on getting the highest quality de-interlacing scaler. I can get most of the way there using VirtualDub's Lanczos-3 scaler (the filter HDlink uses internally) the deinterlacer in VirtualDub is not as good as HDLink's for high frequency detail.

The scaler in HDLink has been used for downsizing HD for SD broadcast / DVD mastering at a major Hollywood post facility, it is one of the few to use high precision native YUV and get the 709 to 601 colorspace correct.

David Newman
CTO, CineForm

musicvid10 wrote on 11/27/2010, 3:28 PM

Looks like another endorsement for Lanczos 3 over bicubic.
Every bicubic downsample I have tried on this material causes extreme moire.
Although Lanczos 3 doesn't entirely eliminate moire, it reduces it to almost subliminal levels.

So what is HDLink using for deinterlacing? Something along the lines of mod yadif + Mcdeint or EED12 ?

Arthur.S wrote on 11/29/2010, 6:26 AM

"Select “Lagarith Loseless codec” from the “Video Format” dropdown."

I don't have this option in the video format dropdown. Has it to be installed separately?

Edit: Found it :-)

Arthur.S wrote on 11/29/2010, 7:02 AM

OK, I'm running through your procedure Jerry. First obstacle is on the 1st render, I end up with a 4:3 pic instead of 16:9. I've tried it with the “Stretch Video to Fill Output Frame (do not Letterbox)” checked and uncheck. Also tried using square pixels. Should add, I'm in PAL land so using the PAL DV WS template.

amendegw wrote on 11/29/2010, 12:39 PM

Arthur.S,

First, I must supply an caveat - this is the first time I've ever messed with PAL - so, my following comments may be way off base.

That said, I tried to follow my own procedure and came up with the following specs:

When I drop this in the time line of the second project, the apect ratio looks fine. I will say, when I play the AVI in WMP, it plays as 4:3 (this is also true for my NTSC project). It's a mystery to me why this should be. When I use MediaInfo to display the clips properties, it reports "Display aspect ratio : 5:4" for the PAL AVI and "Display aspect ratio : 3:2" for the NTSC AVI (I don't understand this). However, since Vegas appears to handle the aspect ratio correctly and I never need to play the AVI directly, I haven't worried about it.

Maybe someone who understands this better than I can comment.

Good Luck!
...Jerry

Arthur.S wrote on 11/29/2010, 12:47 PM

Your settings are the same as mine Jerry. I also tried the NTSC WS template and got the same 'square' results. Difference is; Mine is 4:3 on both the Vegas timeline, and in WMP.

amendegw wrote on 11/29/2010, 1:04 PM

The only other thing I can suggest is to make sure the project properties of Project B match the render properties in Project A.

Good Luck!
...Jerry

EdinKent wrote on 12/13/2010, 11:15 AM

Well, I found this not too difficult at all..

I followed the original instructions using an intermediary m2v file in which I turned the interlaced video into progessive at double the frame rate (blend fields using box rather than Gaussian, and dissable resampling - also removed ticks in project settings for adjust source media, bypassed motion blur, and removed the switch for maintain aspect ratio - this ensured no cropping or adverse pixel shapes). Then created progressive downres ntsc video for DVDA and I then went ahead to prepare the DVD folders. I have uploaded the VOB file to Vimeo (don't like YouTube), having renamed it first as an mpg file. Only used codecs that came with Vegas Pro 10, and no effects or filters (tried various eg. blur, sharpen, chroma blur, but they didn't improve already good results).

Question is: Was I supposed to make an interlaced end result? If so, why would you?

Video is here if you wanna have a look..

http://vimeo.com/17773353

amendegw wrote on 12/13/2010, 12:40 PM

"Question is: Was I supposed to make an interlaced end result? If so, why would you?"Only because the DVD specs call for Interlaced.

That said, as was discussed in some posts earlier, one potential solution would be to create an progressive intermediate with the final render to 720x480 60i.

Sill trying to digest your procedure. Maybe I'll try to re-create it.

Also, maybe it's my old eyes (or maybe what Vimeo is doing to your upload), but your Vimeo clip sems "softer" with less detail than some of the other entries. Can you post a downloadable link to the VOB? Edit: Whoops, forgot that Vimeo will let you download the source - I'll try that.

...Jerry

Laurence wrote on 12/13/2010, 1:30 PM

Slightly off the original topic, but here is the same clip deinterlaced and downrezzed to 700x400p resolution for online viewing with Handbrake:

http://vimeo.com/17779291

hazydave wrote on 12/13/2010, 1:57 PM

When shooting interlaced good cameras use line pair averaging to reduce vertical resolution and improve noise.

You sure about that? Comb filtering (what you're describing there) is applied on playback, not on recording. The reason is simple: the technological improvements have made this increasingly better over the year, and at least for actual broadcast NTSC (CVBS, etc), you're going to want comb filtering anyway, as a superior means of separating chroma from luma.

The better recent analog televisions (or digital TVs with analog inputs) all do 3D motion adaptive multi-line comb filtering on playback, not simply 2D (which sounds like what you're suggesting: Ln' = 1/2 Ln + 1/4 Ln-1 + 1/4 Ln+1... a couple of years ago, I wrote an NTSC decoder in software, including 2D adaptive comb filtering... 3D was too CPU intensive).

If you processed video with the same thing up front, you'd be pretty much left with mush by the time you were done. Not to mention that no DV camcorder would deliver over 240 (NTSC) horizontal lines of resolution.

amendegw wrote on 12/13/2010, 2:34 PM

@EdinKent Well, I did not try to re-create your procedure, but I did go back and re-evaluate the procedure I suggested a few posts up (i.e. render two projects, the first one to a Lagarith Intermediate, the second to the MainConcept DVD NTSC Widescreen).

I uploaded my results to Vimeo: http://www.vimeo.com/17780842 You will not be able to adequately compare the results unless you download your source clip and mine and play them side-by-side.

From what I see, your effort resulted in less definition (i.e. detail), but better moire reduction than the Lagarith procedure. Other eyes may disagree with my conclusion.

So, I think the conclusion is... Downrezzing in Vegas (like all of life) is a compromise. A balance between a sharp image and flicker/moire issues.

Thanks for your contribution!
...Jerry

EdinKent wrote on 12/13/2010, 2:48 PM

Yeah, I should have said don't view on Vimeo's web viewer.

Yes, DVD interlaced video, yes. But most TV's now de-interlace as you view them, so unless you have a 20 year old cathode ray bomb in the corner of the lounge, I'm not sure if it neccessary to adhere to current DVD specs... how long will they last anyway?

What peaked my interest about your test, was that I do not yet have a high def camera (!!!) and am all too aware and practiced in trying to surmount the problems of a) bettering standard def - your test in reverse!, and b) coping with and converting interlaced videos. Very often I will upres SD vid to HD (Christmas in France 2009 is an eg on my Vimeo page), and have realised through trial and error, that Vegas produces good results when you force it to turn each field into a whole progressive frame (hence the double-the-frame-rate part of the process). Blending fields is important because it keeps blockiness to a minimum, but bypassing any motion blur seems to help keep frames crisp.

Once you try to interlace up OR down res'd videos, having used a progressive intermediate, say hello to moire, artifacts, loss of detail etc... I don't think there is a way around it because the fields simply don't work together as they did in the original format. If you look closely at some of the other attempts posted earlier, you'll see good moire suppression but the fine CHECKERED cloth has turned into a vertically striped cloth. I tried to keep as much of what was there as I could.

PeterDuke wrote on 12/13/2010, 7:55 PM

" I'm not sure if it neccessary to adhere to current DVD specs"

You should, if you want your DVD to play on most DVD players.

You can of course use 30p (25p) flagged as 60i (50i). The downside, particularly for 25p, is the less fluid motion compared to interlaced.

NickHope wrote on 2/16/2011, 12:42 AM

Jerry, your .mpg and .mts download links don't seem to be working. Are they dead? I would be interested to try CCE and TMPGEnc MPEG-2 renders some time as a comparison to your MC render.

(p.s. I'm back into the YouTube/Vimeo project and will add to that thread soon)

amendegw wrote on 2/16/2011, 2:49 AM

"Jerry, your .mpg and .mts download links don't seem to be working. Are they dead?"Okay, the links should be fixed now. Seems that the only way I can get the links to work properly is to zip up the source and link to the zip file.

...Jerry

NickHope wrote on 2/16/2011, 4:02 AM

Thanks Jerry. Will have a go at it some time.

EDIT: This discussion continues

amendegw wrote on 12/14/2011, 8:19 PM

Okay, throw things at me, call me names, do what you will, but I need to make one more post to this thread.

Over the past year, I've keep the Hula-dancer in my mind, trying various experiments and never coming up with a good render.

Recently, everything seemed to come together. What about using the 24p 2-3 pulldown DVD standard? So, here's what I did:

1) New Vegas Project with 1920x1080 29.97 Progressive Project Properties.
2) Drop the 1080i Hula Dance clip on the timeline
3) Apply the yohng yadif Deinterlace FX as a Media FX.
4) Render as the builtin MainConcept "DVD Architect 24p NTSC Widescreen video stream"
5) That's it!! Very simple. No Sharpen FX, no Reduce Interlace Flicker, no Gaussean Blur, no nuttin'

Best thing is - look at these results: 24p-HG21.zip

To my eyes, this image looks great!

As I see it, the only downfall is the 24p could introduce some stutter from the original, but I can live with that versus the quality/simplicity of this method.

Comments are invited.

...Jerry

Edit: As I scroll thru this thread, 24 p was mentioned by Bob("farss") & musicvid earlier, but I don't think anyone ever made a test render.

musicvid10 wrote on 12/14/2011, 8:48 PM

That template flags the actual 24p for 59.94i DVD playback with pulldown . It looks really smooth on playback with VLC, but I'm searching for the reason(s).

Is it the pulldown itself that smooths the moire and twitter, was resampling "on" during the conversion to 24p, which deinterlace method was used, and how does it play rendered in pure 24p, without pulldown, resampling, or neither?

amendegw wrote on 12/14/2011, 8:59 PM

1) The default, "smart resample" was not changed.
2) Deinterlace was set to "None" - after all, this is now a Progressive Project, with progressive footage on the timeline.
3) I rendered a different clip using the same procedure, except 24p w/o pulldown and it looked great and even played on my old Toshiba SD-1800 DVD player with Sony Triniton TV.

I suspect, but cannot be sure, that the reason for the improvement is that yadif is now used for deinterlacing & Vegas now only has to resize progressive footage.

...Jerry

btw: this is the thread from the dvinfo forum that put me on this track: http://www.dvinfo.net/forum/what-happens-vegas/503403-how-well-does-vegas-resize-1080-a.html

Edit: Hmmm... seems to play much better in WMP than VLC. Wonder why?

musicvid10 wrote on 12/14/2011, 9:41 PM

Being as this is one hole in my workflow I have not yet filled, I suppose I will be compelled to run more tests over the holidays again this year . . .

Thanks as always, Jerry!
;?)

NickHope wrote on 12/15/2011, 5:25 AM

Yeah thanks Jerry... Just when my brain had nearly recovered from the previous fryings ;)

Interesting discovery. In a similar vein I noticed a "improvement" in the output when I tried resampling 1080-50i to SD NTSC using QTGMC in this thread, and speculated that it might be due to some extra degree of "low pass filtering" as a by-product of the retiming. I think there may well be something in it, but the science behind it is beyond me.

amendegw wrote on 12/15/2011, 5:34 AM

"but the science behind it is beyond me."Nick, I can't believe you said that! IMO, you are the "go to" authority on this subject.

btw, has the flooding subsided?

...Jerry

1 … 2 3 4 5

of 5