When downscaling HD interlaced to SD interlaced, Vegas splits the video into individual fields of 1920x540 progressive at double the original frame rate, then resizes each progressive field and then re-interlaces the video back into interlaced frames. Vegas does it precisely the way it should be done


Well, this is where it all seems to go off the rails - if Vegas were resizing fields instead of frames, you'd expect to see 'dogteeth' - they're still gonna be there once you view the whole frame. (And this is what I get using the Virtual dub method)

Even more interesting - if no 'dogteeth' are seen in a resized frame, then something more than simple resizing has occured to the fields of that frame. Puzzling still - if blend or interplolate show no difference - just what *is* vegas doing then?

I'm still convinced Vegas is resizing based on a frame image - I can't see any other reason that would cause Vegas to lose dogteeth in an interlaced image when downsizing. And the 'none' setting produces a result that I would expect of a interlaced 'frame' with motion 'dogteeth' - you see an aliasing effect in the thickness of the 'dogteeth' themselves that way - assuming that 'none' means "no interpolation OR blending of the fields of the frame to be resized" which is what it appears to be. The only way you could possibly avoid that (re-aliased dogteeth when using 'none' )in a resized frame of time-differing fields showing dogteeth is if the downsize, vertically, were an exact multiple of the original size. However, 1080 to 480 is greater than 2:1.

It's just not possible to see that aliasing if the fields are resized individually.

"As far as you not seeing any interlace artifacts, was your preview window set to "Best" > "Full"?"

Absolutely however when I say "artifacts" I don't mean interlaced combing I mean dogs teeth as shown in the first image on this page labelled "Resized before deinterlaced". The normal interlace mice teeth were there.

The thing I'm really grappling with is the loss of resolution.
I started with an image that had 830 lines in 1080 and scaled it to 576. That should at the simplest analysis yielded a resolution of 442 lines. Pretty good for SD PAL.
Assuming Musicvids factor of 0.83 then I should get around 360 lines but I'm getting way less than that if I select either Interpolate or blend. As you say it doesn't seem to matter which one despite many posts discussing which one :)

Bob.

"I'm still convinced Vegas is resizing based on a frame image"

It's not. Really.

I just took some 1080i material and rescaled it to widescreen 480i with the "interpolate" delinterlace method, dropped it back on a "DV Widescreen" timeline and set the preview window to 873x480 with the Preview Quality set to "Best - Full." I can EASILY see the "dogs teeth" artifacts of interlacing which is as it should be.

Re-read my message above. Vegas is scaling the video using the absolute best possible method, one a field at a time. Really.

Well I learnt something there also (even if I am an interloper).

To get noticeably better resolution, when time permits, I pre-process interlaced footage to deinterlace it based on the motion-compensated deinterlacer TDeint in AviSynth. TDeint lets you deinterlace to either 25p or 50p, in each case by motion-estimating where the pixels (of various shades) would be in the "missing" lines, to get full-resolution fields (hence progressive frames). Motion estimation is not common in NLEs etc. as it is heavier to process and would reduce the preview framerate.

I typically use AvsP to edit and check the AviSynth script then save it straight to AVI-Cineform then read that into Vegas. Then in Vegas ensure its Media Properties for that file says Progressive (sometimes it gets it wrong, a limitation of the AVI file format rather than of the NLE, which has to guess).

We're not talking about deinterlacing to progressive, we're talking about rescaling interlaced HD to interlaced SD. Vegas does this PERFECTLY, no additional third-party software or plugins are necessary.

" Vegas does this PERFECTLY, no additional third-party software or plugins are necessary. "

I wish I could be so certain of that. Surely we shouldn't be loosing so much resolution in the process?

If I treat the exact same video as progressive the outcome is double the resolution, why?
I'm NOT saying this is a uniquely Vegas problem but clearly something is wrong here. From my memory banks 480 lines is considered good looking PAL SD. 230 lines is not very good at all, period. This is putting a quantitative result to the years of complaints from people here and elsewhere who have bought HD cameras and are trying to deliver good looking SD interlaced video.

Bob.

I just took the reschart.png file, cropped it in Photoshop and resized it to 1920x1080, pulled that into a DV Widescreen project in Vegas 10.0c, added the bouncing ball and rendered to an interlaced file using the DV Widescreen template but changing the codec to the Lagarith lossless codec. I then pulled the rendered file back into Vegas and compared the two. They are absolutely identical and the trumpet shows just over 375 lines lines of vertical resolution and about 500 lines of horizontal resolution. The interlacing shows dog's teeth as expected. I'm sorry, but I'm just not seeing a problem.

"I just took the reschart.png file, cropped it in Photoshop and resized it to 1920x1080, pulled that into a DV Widescreen project in Vegas 10.0c, added the bouncing ball and rendered to an interlaced file using the DV Widescreen template but changing the codec to the Lagarith lossless codec. I then pulled the rendered file back into Vegas and compared the two. They are absolutely identical and the trumpet shows just over 375 lines lines of vertical resolution and about 500 lines of horizontal resolution. The interlacing shows dog's teeth as expected. I'm sorry, but I'm just not seeing a problem"

Doing that I wouldn't expect you to see a problem either. Doing something similar I don't see a problem. What you've done is get Vegas to rescale a 1920x1080 image as progressive to SD and rendered that as interlaced. Result is very good. From that I concluded contrary to what many have said that Vegas's precise bicubic rescaling is pretty darn good.

The problem I am addressing and concerned about is rescaling interlaced HD to SD and the results are dramatically different compared to rescaling progressive HD to SD. If progressive can be rescaled and produce such a good result and the rescaling algorithm is good and the tests show it is good what is the issue with rescaling interlace, that is the question.

Bob.

You're right, I wasn't using an interlaced HD file. I just took the reschart, rendered it as 1080i using Lagarith, brought it back in and rendered to 480i, I'm still getting the same results. ~375 vertical, ~500 horizontal. Maybe I should just shoot the reschart with an HD camera at 1080i to simulate a more "real-world" test. Could be the difference between NTSC and PAL scaling. I'm really trying to replicate what you're seeing, I'll play with it more later, I have to get some work done.

" I have to get some work done"

Same here :)

Here's images of my results:

Progressive

interlaced

I have a suspicion the issue I'm seeing has something to do with the difference between scaling the image as one frame and scaling the image split into two fields. The former means it is scaled once only the latter might mean there's two lots of scaling going on. Just a hunch though. Or as you say it could be something to do with a difference in what we're scaling too.

If I can find the time I'll repeat the tests to NTSC SD

Bob.

Bob, on that link you gave - the second image, resized before deinterlaced 2 - *that's* the one!

I seriously misundserstood you on 'dogsteeth' though - it's combing I meant though.

Image two show exactly the effect of resizing a frame of interlaced video with no blending or interplolating - 'none'. There's the tradmark aliased combs that vary in width. This is not possible with field-based resize.

Loss of resolution in the other two Vegas resize/deinterlace modes... three guesses....

Bob & John,

Highly appreciate you efforts in this matter - trying to find the truth in the higher-loss-of details-than-aticipated in the Vegas HDi to SDi conversion. I am also in PAL-land and it will be easy for me to duplicate your test. I have never been satisfied with the Vegas downscaling and would also eagerly want to find at least an explanation for this, and a viable solution. A good PAL resolution should at least be 360 horizontal lines.

Will report my findings after I find time to do the test.

Cheers,

Christian

Could someone please tell me briefly how to read those ISO res charts? What I should be looking at to see how much resolution I've got?

Nick,

The slanted Horizontal Line is used to measure the vertical sharpness and the slanted Vertical Line is used to measure the horizontal sharpness.

The line widths per picture height are analyzed using Imatest software.

I think?

AFAIK when they test a camera's video resolution they now use a DSC Labs MultiBurst Maxi Focus Test Chart

John

I think there has been in excess of (perhaps) 600 messages over the past few weeks searching for "the truth" to downscaling video. It is unbelievable to me that no one from SCS contributes to these types of discussions.

/jerry

I think there has been in excess of (perhaps) 600 messages over the past few weeks searching for "the truth" to downscaling video. It is unbelievable to me that no one from SCS contributes to these types of discussions.
=====================
Excellent point Jerry. It's the number one complaint from those of us that own HD cameras, and the most often posted topic here regarding working with HD footage. It also is on other sights like DV Info. Most of us are still producing SD DVDs.

A few weeks ago I mentioned that I wish SCS would work on this aspect of making Vegas better instead of the bells and whistles they usually develop with each new release. Some argued that it is fine the way it is. However, it would be discussed this often if it were.

John

Deinterlacing to double framerate (50p if you're using PAL), as TDeint can, allows Vegas to scale-down each target interlaced field (odd & even) from a full-definition progressive source frame, each derived from an odd or even field of the source. The third-party software just gives Vegas something better to start from.

@farss et al: Great thread!

@TeeTime: Here's the question I submitted to the recent DVD Webinar: "What is the recommended method for downrezing Vegas HD projects to SD DVDs - recognizing that resizing often results in flicker and moire patterns as well as loss of sharpness. And are we getting any help in future versions of Vegas / DVDA to assist us in this process?" SCS promised to post all questions with answers once NAB is over - I'm waiting with "worm on tongue" (i.e. bated breath).

...Jerry

farss wrote: Looked pretty good, can still read the same resolution from the trumpets... ...the res was only 230 lines!
John_Cline wrote: ...the trumpet shows just over 375 lines lines of vertical resolution and about 500 lines of horizontal resolution
craftech wrote: The slanted Horizontal Line is used to measure the vertical sharpness and the slanted Vertical Line is used to measure the horizontal sharpness.


With reference to one of farss' images linked above, I'm still not sure what exactly I should be looking for. I could also include that test chart in my comparative renders but not much use if I don't know how to read the result.

Guys, seriously, Vegas rescales HD interlaced to SD interlaced just fine. I'm amazed at how often this comes up mainly because there is no problem that I can see. SCS can't make Vegas scale any better because it's already the best it can be.

Some people have mentioned that the downscaled version seems a little soft to them. There's a very simple explanation for that; HD cameras don't over-sharpen like SD cameras. When you downscale HD video and you want it to look as artificially sharp as an SD camera, then just add the sharpen filter.

I've been shooting HD exclusively for probably seven or eight years now and I deliver in both HD and SD. My clients are picky and I'm simply not going to take any shortcuts. I've tried many different methods and programs to downscale HD to SD and nothing does it better than Vegas. Just select either "Blend" or "Interpolate" and "Best" render quality and render. Maybe add a little sharpening, if necessary. Done deal.

That said, there may very well be a problem with downscaling to PAL SD because I trust what Bob (farss) is seeing. I just can't see it myself.

Nick said,
craftech wrote: The slanted Horizontal Line is used to measure the vertical sharpness and the slanted Vertical Line is used to measure the horizontal sharpness.
=====================
I also said that to read those two edges you need Imatest software AFAIK. The calculations of the Spatial Frequency Response are done like this.

I am not sure you can accurately read that ISO 12233 chart by eye. So if Bob and John know a way to approximate what the charts say maybe they can explain it.
I think the point where the converging lines cannot be distinguished anymore (both the black and white lines become half gray), is the approximate number of "lines of resolution"

John

EDIT: I found a detailed explanation of how to read it by eye in dpreview. It would seem to me that the entire results could be completely thrown off by a even a slightly different white level (contrast) change anywhere in the workflow. The chart is designed to test a camera. Maybe that's why Bob and John are getting different results.

"I am not sure you can accurately read that ISO 12233 chart by eye. So if Bob and John know a way to approximate what the charts say maybe they can explain it.

Yes, by my understanding the point at which the converging lines first cannot be distinguished is the resolution limit. Beyond that point you may see lines appear again and this is caused by aliasing and you will notice you see less lines. You are I believe seeing what happens when the Nyquist limit is exceeded. Reading these charts incorrectly and reporting optimistic results is not uncommon when testing cameras.

There is another problem though. The alignment between the camera and the chart is also critical. It is recommened that the camera's ouput be monitored and the camera moved a tiny amount up/down and left/right to get the best outcome. Failure to do this explains why two different people doing the same test and reading the result correctly can get different results.

Both of these issues are unique to digital imaging. With a purely optical / analog system such as film the trumpets just merge into grey and film doesn't use discrete photosites that should be aligned to the chart.


One thing I'd like to make very clear was my inital purpose in doing these tests. I like John Cline did not believe there was anything wrong with the way Vegas does things. My subjective evaluation of how my SD DVDs looks was they looked pretty darn good. On the other hand others were not happy and this was not even limited to Vegas users, I read the same complaints from Adobe users and Apple users. No one had ever put any numbers to the outcome though, it was all subjective and heck, I'm not expert in judging picture quality so maybe I was missing something or maybe not. I just wanted to get some hard facts from tests we could all reproduce and then we could all agree if there was or wasn't a problem. Then from that we could further see where the problem arose. Were the people who say "If you want great SD, shoot SD" right or was the problem in SCS's and Adobe's and Apple's software.

At this stage I cannot find any justification to say Vegas is at fault. John Cline has gotten a significantly different result to me. This is good because it is peer group review at work so now I have to work out why the different results. Personally I'm inclined to side with John's results simply because if the SD I'm producing is only 230 lines res even I would be screaming long and loud, trust me it is not THAT bad!

My gut feeling at the monent is there's a flaw in my process. Simply putting a res chart in front of Vegas's "camera" removes much of what happens in a real world camera. I can clearly see major issues with aliasing in the 1080 source that I would not see if I'd used a camera to shoot the chart. That's hardly Vegas's fault, it is impossible to emulate well an optical low pass filter in software.

So going forward my plan is to find our printed 4K resolution chart (we moved) and park that in front of the best HD camera I can borrow (PMW 350) hook that up to a Nanoflash, monitor that on a full HD monitor and record the results at 25p, 30p, 50i and 60i. Sharing that around will be a tad more difficult than a Veg file and a png of a res chart but we only need 1 second of each format so hopefully Dropbox to the rescue.

I'd like to thank those who put forward some ideas on how to get better results using tools outside Vegas. I'm not ignoring the suggestions and indeed there may be ways to get even better results. Until we all agree on how good Vegas is doing it's job in a quantitive way I just feel we'd be going down too many rabbit holes trying other techniques.

Bob.