Aliasing, artifacts and moire?

Comments

Laurence wrote on 4/26/2011, 9:23 PM
Here's a really simple way of doing a nice downrez:

1/ Buy Cineform Neo (not Neo Scene because it doesn't have the fancy Lancos3 resizer)
2/ Import your HD into Cineform using the resize to SD resolution and deinterlace (if you need it) options. The Neo lancos 3 resize is really good and the Cineform codec not only looks good, but unlike DV codec, it preserves the colorspace which you can then render in it's full intensity to SD mpeg2.


farss wrote on 4/27/2011, 4:12 AM
"I Is there something about the yadif plugin that makes it a noncontender for this kind of work? I confess, I have not tested it in Vegas yet."

I didn't even realise it was available but indeed it is.

http://www.roccocaprella.com/tutorial/1095/vegas-pro-10-yadif-deinterlacer/?lang=en

It is an OFX plugin so only runs in V10. What I don't understand is why it can only be applied as a media FX.

Bob.

[edit]
I just thought of an answer to your question. Yadif cannot be used in the method we are talking about here according to the comments i.e. it will not de-interlace fields to produce frames. Just to be clear what I mean; it WILL de-interlace fields to produce a frame. This is from the "cannot be used for slow motion" bit. Hope that makes sense.
craftech wrote on 4/27/2011, 4:49 AM
In terms of the last three posts, I am still not understanding whether the recommendation in downconverting HD interlaced to SD interlaced for DVD is to deinterlace somewhere in the workflow to get the best results.

Like I said, with the Mike Crash filter I am only a little less than half way through the 50 hour render of the deinterlaced intermediate Jerry recommended doing first above. I don't have a slow computer. I find it hard to believe that many of you would be allowing 2 days of rendering just for deinterlacing.

So are you all deinterlacing in your workflow and if so, how and where?

Thanks,

John

farss wrote on 4/27/2011, 5:04 AM
"So are you all deinterlacing in your workflow and if so, where?"

I simply floated the idea and JM tried it in AVISynth and got some very good results. When I get a chance I'll try an idea I have to do it wholely in Vegas.

Conceptually the idea is to convert fields to frames e.g. 50i to 50p with the highes possible resolution frames and then scale each frame to a field back into 50i SD. To do this my thinking is to use a 50p project, drop 50i media into it etc.
Then apply the de-interlacer in the 50p project using frame differencing. Then nest that project into a 50i SD project.
How this would hopefully help is to increase the resolution of each frame in the 50p child project by blending the parts of the 'fields' where there's no motion.


Bob.
craftech wrote on 4/27/2011, 5:21 AM
my thinking is to use a 50p project, drop 50i media into it etc.
======================
That's probably similar in terms of render time to what I am doing because of the Mike Crash deinterlacer. I guess it would cut out the intermediate though.

Which script of JM's did you try?

Thanks,
John

EDIT: This is a pretty interesting read on the subject although a bit dated.

His take on the Mike Crash Filter:

* Does not always eliminate all interlaced lines.
* Sometimes eliminates the wrong video data.
* Sometimes complicated parameters that can differ from video material to video material. Click the pic below, and tell me what's best for your movie:
http://www.100fps.com/ohmygod.gif
* Movie can become unnaturally blurred (unsharp) during movements.
farss wrote on 4/27/2011, 6:17 AM
"Which script of JM's did you try?"

None. John too k my idea on board and did the tests himself.

I've read 100fps several times over the years.
keep in mind though that he's talking about a different problem. He's talking about how to de-interlace interlaced video for display on a progressive only display such as an LCD. It is kind of dated because a) if you've got deep pockets you can get an LCD monitor that does display interlaced video and b) the latest things with HTDVs is "motion flow" which raises the bar dramatically.

The point of what I'm hoping to achieve is to get a better outcome using just Vegas when going from HD interlaced to SD interlaced. The issue is this is tricky. Interlaced video is already compromised is vertical res, each field is LESS than half the maximum limit. Bumping it up by blending the fields in the static areas would increase the perceived resolution of the SD. I hope by not too much or then you hit the problem of line twitter.

He's right about the smart de-interlacer taking a bit of experimenting to get right. This is a given, it is technically impossible to de-interlace in a way that is 100% foolproof no matter how much money you throw at the problem.


Bob.
NickHope wrote on 4/27/2011, 6:23 AM
Yadif cannot be used in the method we are talking about here according to the comments i.e. it will not de-interlace fields to produce frames. Just to be clear what I mean; it WILL de-interlace fields to produce a frame. This is from the "cannot be used for slow motion" bit.

Is correct. I just tried it. Sujects in the even frames of the 60p output are in the same position as in the odd frames (although the frames are not identical).

However Yadif(mode=1) can be used in AviSynth to give double framerate (bob), with temporal and spatial interlacing check, as can TDeint(mode=1).
craftech wrote on 4/27/2011, 10:12 AM
Check out this video tutorial using AviSynth and some other 3rd party tools including the Debugmode Frameserver to do just what we are talking about here. He recommends sticking with either Progressive or Interlaced throughout the workflow.

The video is an offshoot of this article published two years ago.

John
amendegw wrote on 4/27/2011, 2:45 PM
"Like I said, with the Mike Crash filter I am only a little less than half way through the 50 hour render of the deinterlaced intermediate Geez - I hope the procedure I posted above is not perceived as a "recommendation" to perform the HD to SD downrez - it was merely a test combining some ideas by farss, musicvid & Nick_Hope to see what the rendered output looked like. I thought the render looked pretty good, but it should be subjected to peer review before it becomes a "recommendation".

...Jerry

System Model: Alienware Area-51m R2
System: Windows 11 Home
Processor: Intel(R) Core(TM) i7-10700K CPU @ 3.80GHz, 3792 Mhz, 8 Core(s), 16 Logical Processor(s)
Installed Memory: 64.0 GB
Display Adapter: NVIDIA GeForce RTX 2070 Super (8GB), Nvidia Studio Driver 527.56 Dec 2022)
Overclock Off

Display: 1920x1080 144 hertz
Storage (12TB Total):
OS Drive: PM981a NVMe SAMSUNG 2048GB
Data Drive1: Samsung SSD 970 EVO Plus 2TB
Data Drive2: Samsung SSD 870 QVO 8TB

USB: Thunderbolt 3 (USB Type-C) port Supports USB 3.2 Gen 2, DisplayPort 1.2, Thunderbolt 3

Cameras:
Canon R5
Canon R3
Sony A9

Laurence wrote on 4/27/2011, 4:56 PM
Back in the SD challenge thread below:

http://www.sonycreativesoftware.com/forums/ShowMessage.asp?MessageID=738210

The method that really impressed me for being drop dead easy and giving as good results as anything else was the David Newman entry:



Just import with HD Link from Cineform Neo HD (or Neo as it is now called) using the built in lancos 3 scaler. Call the downrezzed avi into Vegas again and render into mpeg.

Like I said a few posts back, this is my current favorite method to downrez. I'm still looking for a utility like Handbrake that will rerender an HD file directly into SD resolution mpeg2 though. I thought I might have it with one of theTmpgenc product line, but in order to do that I had to render into a specific format like uncompressed in HD before doing the conversion. On the other hand. HD Link will convert just about anything, and rerendering the SD resolution Cineform master into mpeg2 is a lot quicker than rendering an HD intermediate.

Another thing I like about working with SD resolution Cineform is that unlike DV codec it allows you to keep the colorspace and render it into SD. Not only that, but it smart-renders, so it is easy to compile a dvd out of sections.
craftech wrote on 4/27/2011, 5:05 PM
Like I said, with the Mike Crash filter I am only a little less than half way through the 50 hour render of the deinterlaced intermediate Jerry recommended doing first above. I don't have a slow computer.
=======================
"Geez - I hope the procedure I posted above is not perceived as a "recommendation" to perform the HD to SD downrez - it was merely a test combining some ideas by farss, musicvid & Nick_Hope to see what the rendered output looked like. I thought the render looked pretty good, but it should be subjected to peer review before it becomes a "recommendation".
========================
I absolutely realize that and didn't mean to suggest that it wasn't worth trying. I am always willing to try new things, and thanks for doing all the experiments you and others have done. I truly appreciate them and wanted to try this one because I liked your results.

However, while it worked well on the short doll video clip, I wanted to see if it would be worthwhile on something longer. I planned to post back the results.

The problem is that the intermediate isn't done rendering yet. My estimate of 50 hours total render time for the hour and twenty minute video was estimated after 24 hours of rendering had passed. It is now two days (48 hours later) and it is only 55% done. So my revised estimate for completion of the rendering of the intermediate is more like FOUR DAYS or more like 90 hours.

I think I am going to stop the render in a few minutes and can safely report back that the Mike Crash deinterlacing filter is only practical for the shortest of videos.

At this point, even if it turned out great it wouldn't be worth that kind of render time. Even back when I was rendering with Vegas 3.0 on a Windows 98 computer there was absolutely nothing that I did that took anywhere near that long. I think I remember enduring a 30 hour render for something or other back then, but nothing resembling a four day render (of an intermediate to boot).

It was definitely worth a try Jerry and thanks again posting it.

John
craftech wrote on 4/27/2011, 5:11 PM
Just import with HD Link from Cineform Neo HD (or Neo as it is now called) using the built in lancos 3 scaler. Call the downrezzed avi into Vegas again and render into mpeg.
===============================
Laurence,

Neo HD is $400. Did you spend $400 just for the Lanczos 3 resizer? And isn't that resizer also a part of Virtual Dub which is free?

John
ushere wrote on 4/27/2011, 5:13 PM
hey john,

with a render time like that perhaps it's time to upgrade from that 286 ;-)
Laurence wrote on 4/27/2011, 5:26 PM
Actually Neo (as it is now called) is $299, but yes I paid more. Buying it makes sense to me for several reasons:

1/ The lancos 3 resizer makes a noticable quality difference as you ingest video of different resolutions. The lancos 3 resize at this stage means that everything will look as good as it can in your most important hd master render.

2/ First Light does a better job of color correcting gradients than Vegas does. Actually Vegas comes close in 32 bit modes, but it is so cumbersome in that mode that I never use it. In practice, I do a combination of interviews from my HDV camera that are nicely white balanced and exposed, and b-roll from a variety of sources: Go Pro, Canon SX-1IS, etc. Often the b-roll is shot in auto mode and needs some serious correction. Neo uprezzes 720p to 1080p very nicely and gives you very low overhead yet extremely smooth color, white balance, and exposure correction with First Light.

3/ HD link will convert cRGB to sRGB and fix unstandard frame rates like 30p on ingest. You can slow 60p to 30p, color correct cRGB to sRGB, and uprez from 720p to 1080p all at once and have the resulting intermediate work smoothly on a Vegas timeline. This works very much the way I do in practice.

3/ HD link will convert anything you throw at it. Unlike Virtualdub which is picky about input formats. HD link will convert a .mxf master for instance. HD link will uprez 1440 to 1920 for mixes of HDV and full 1920 resolution HD source material.

4/ HD link does an outstanding job of downrezzing to SD. You can do this with your source material for SD projects that will pop with higher resolution colorspace, or you can do it to downrez an HD master to SD for DVD delivery.

5/ The $299 version of Cineform Neo now does up to 4k resolution, really high colorspace formats, uprezzes the colorspace on conversion (which for some reason that I don't understand looks really good to me), and it now gives us an alpha layer format with no artifact problems (that I can see) and 10 bit precision. Black halos around the edges of the alpha layer anyone?

Yeah it's worth it!
craftech wrote on 4/27/2011, 5:39 PM
OK Laurence,

It seems like it has a lot of useful features. I'll look into it. Sounds like a nice program.

Thanks,

John
NickHope wrote on 4/28/2011, 6:21 AM
With reference to Jerry's liking for having "Reduce interlace flicker" on, I wasn't so keen when I did some tests on a variety of footage. The video looked more film-like, like it has been interlaced from progressive, and I didn't really like it. I did some further testing and I have discovered that "Reduce interlace flicker" simply does a blend deinterlace. To verify this:

1. Set preview window to Best (Full).

1. Put an interlaced clip with lots of movement onto the timeline in an interlaced project with "Reduce interlace flicker" unchecked. You should see lots combing as expected.

2. Check "Reduce interlace flicker". All combing disappears and the appearance is "blended".

3. Change project properties to progressive with "Blend" deinterlace method. No change to the preview.

4. Uncheck "Reduce interlace flicker". Still no change to the preview.
amendegw wrote on 4/28/2011, 8:30 AM
Nick, I don't have the time to do testing on this right now - maybe next week. Here's what I've found out by trial-and-error.

1) When putting an interlaced Media Event on the timeline and rendering it (w/ Deinterlace=Interpolate or blend), the resultant media file appears too soft with lack of definition for my eyes.
2) So, I put a Sharpen=0 FX on the Video Event and now it looks good - except that it introduces mucho flickering & moire.
3) If I use the conventional wisdom of putting some Gaussian blur on the event, I must apply so much to remove the flickering/moire that the end result is less sharp than the original - I've posted a screen capture to illustrate this near the top of this thread.
4) However, if I click the "reduce interlace flicker" option, the flicker/moire is substantially reduced while retaining most of the sharpness (in the render).

As always, I invite others disprove this. I'm more than happy to say, "I'm wrong" if we can find a better way.

...Jerry

System Model: Alienware Area-51m R2
System: Windows 11 Home
Processor: Intel(R) Core(TM) i7-10700K CPU @ 3.80GHz, 3792 Mhz, 8 Core(s), 16 Logical Processor(s)
Installed Memory: 64.0 GB
Display Adapter: NVIDIA GeForce RTX 2070 Super (8GB), Nvidia Studio Driver 527.56 Dec 2022)
Overclock Off

Display: 1920x1080 144 hertz
Storage (12TB Total):
OS Drive: PM981a NVMe SAMSUNG 2048GB
Data Drive1: Samsung SSD 970 EVO Plus 2TB
Data Drive2: Samsung SSD 870 QVO 8TB

USB: Thunderbolt 3 (USB Type-C) port Supports USB 3.2 Gen 2, DisplayPort 1.2, Thunderbolt 3

Cameras:
Canon R5
Canon R3
Sony A9

PeterDuke wrote on 4/29/2011, 7:16 AM
I thought I saw a bit of moire on Kate's veil after she got out of the car. Tsk tsk!

It's not appropriate to comment here on the sync problem at the start, so I won't.
amendegw wrote on 5/13/2011, 11:58 AM
"I did some further testing and I have discovered that "Reduce interlace flicker" simply does a blend deinterlace. "Nick,

I finally had a chance to replicate your test and see what you mean, however, I think something more is happening here.

The best HD->SD results I have been able to get is to set up two Vegas projects. Proj-A resizes the HD footage to 720x480 and Proj-B renders to MPEG2. Since Proj-B does no resizing and no deinterlacing, I set the DeInterlace to "none". However, for this test, I set the DeInterlace=Blend, and compared "Reduce Interlace Flicker" ="off" versus "on" The "off" looks bad, the "on" has less flicker, but the resolution has gone to hell.

For comparison purposes, I also included a the test with DeInterlace=None. The results appear better. RIF=Off has flicker, RIF=On looks better.

Test cases are here

Just when I think I understand what's going on, I take two steps back. Sigh!

...Jerry

System Model: Alienware Area-51m R2
System: Windows 11 Home
Processor: Intel(R) Core(TM) i7-10700K CPU @ 3.80GHz, 3792 Mhz, 8 Core(s), 16 Logical Processor(s)
Installed Memory: 64.0 GB
Display Adapter: NVIDIA GeForce RTX 2070 Super (8GB), Nvidia Studio Driver 527.56 Dec 2022)
Overclock Off

Display: 1920x1080 144 hertz
Storage (12TB Total):
OS Drive: PM981a NVMe SAMSUNG 2048GB
Data Drive1: Samsung SSD 970 EVO Plus 2TB
Data Drive2: Samsung SSD 870 QVO 8TB

USB: Thunderbolt 3 (USB Type-C) port Supports USB 3.2 Gen 2, DisplayPort 1.2, Thunderbolt 3

Cameras:
Canon R5
Canon R3
Sony A9

amendegw wrote on 5/14/2011, 10:26 AM
SCS has now posted the Q&A from the DVD Webinar. Here's my question & SCS's answer. Does this tell us anything we don't already know?

Q: What is the recommended method for converting Vegas HD projects to standard-definition DVDs—recognizing that resizing often results in flicker and moiré patterns as well as loss of sharpness? And are we getting any help in future versions of Vegas Pro and DVD Architect Pro to assist us in this process?

...Jerry

System Model: Alienware Area-51m R2
System: Windows 11 Home
Processor: Intel(R) Core(TM) i7-10700K CPU @ 3.80GHz, 3792 Mhz, 8 Core(s), 16 Logical Processor(s)
Installed Memory: 64.0 GB
Display Adapter: NVIDIA GeForce RTX 2070 Super (8GB), Nvidia Studio Driver 527.56 Dec 2022)
Overclock Off

Display: 1920x1080 144 hertz
Storage (12TB Total):
OS Drive: PM981a NVMe SAMSUNG 2048GB
Data Drive1: Samsung SSD 970 EVO Plus 2TB
Data Drive2: Samsung SSD 870 QVO 8TB

USB: Thunderbolt 3 (USB Type-C) port Supports USB 3.2 Gen 2, DisplayPort 1.2, Thunderbolt 3

Cameras:
Canon R5
Canon R3
Sony A9

SuperG wrote on 5/14/2011, 3:38 PM
It tells us that interlaced video does not get resized on a field basis but on a frame-basis, and all the blend or interpolate gobbledygoo that goes with it.


Laurence wrote on 9/29/2011, 8:02 AM
I want to post this first without saying how I did it:

http://vimeo.com/29778274

I will tell, but just not yet. I want to see what everyone thinks first.
amendegw wrote on 9/29/2011, 9:38 AM
"I will tell, but just not yet. I want to see what everyone thinks first."Laurence, It looks good, however, by the time it gets to video, the clip gets converted to Progressive.

Can you post a (Dropbox?) link to the Interlaced .mpg ?

...Jerry

btw: The best I've been able to do (using the procedure listed in the 8th post from the top is here: Proj-B-LagsV1322.zip

btw2: I downloaded your mp4 and play it locally, it looks a lot better than the Vimeo embed.

System Model: Alienware Area-51m R2
System: Windows 11 Home
Processor: Intel(R) Core(TM) i7-10700K CPU @ 3.80GHz, 3792 Mhz, 8 Core(s), 16 Logical Processor(s)
Installed Memory: 64.0 GB
Display Adapter: NVIDIA GeForce RTX 2070 Super (8GB), Nvidia Studio Driver 527.56 Dec 2022)
Overclock Off

Display: 1920x1080 144 hertz
Storage (12TB Total):
OS Drive: PM981a NVMe SAMSUNG 2048GB
Data Drive1: Samsung SSD 970 EVO Plus 2TB
Data Drive2: Samsung SSD 870 QVO 8TB

USB: Thunderbolt 3 (USB Type-C) port Supports USB 3.2 Gen 2, DisplayPort 1.2, Thunderbolt 3

Cameras:
Canon R5
Canon R3
Sony A9

PeterDuke wrote on 9/29/2011, 9:42 AM
"It tells us that interlaced video does not get resized on a field basis but on a frame-basis, and all the blend or interpolate gobbledygoo that goes with it."

On the contrary it says that it deinterlaces, implying that each (former) field gets resized independently.