Hi All

I have often seen comments that 1280x720 footage on the timeline resizes better to SD than 1920x1080

Does any bright spark out there know exactly why??? Is it applicable to both NTSC and PAL SD too???

Both HD formats (assuming it's progressive) have identical aspect ratios so is there any real reason why Vegas might do a better job downsizing 720 ???? or has this got something to do with only interlaced footage where the file is now 1440x1080 which has a closer aspect to SD interlaced as one is 1.333 and the other is 1.365???

Some mystery solving would be appreciated

Chris

What I remember from a couple of years ago is that some folks think full rate progressive --- 50p in PAL and 60p in NTSC --- downconverts better to SD because you are working with full frames to make each corresponding interlaced field. They mostly discussed 1280x720 (aka 720p) because we have only recently started seeing a lot of quality cameras that shoot full rate 1920x1080 progressive (1080/50p and 1080/60p). The idea makes intuitive sense since a full-rate progressive frame gives you a lot more image to work with than coming from an interlaced one. However, not everybody sees a significant difference in their work on their computers and projects. I played around with with comparisons with some footage from dance recitals, comparing some 720p from an NX5 with 1080i (some from an NX5 and some from the essentially similar HDV sibling, the FX1000). It seemed to me that the 720p footage was slightly cleaner when viewing the very-high-motion-very-high contrast footage from the cameras on a calibrated 30-inch HDTV, but I personally could not see any significant differences when everything was transcoded to either SD (for timeline playback) or DVD (max render quality, etc.) Mileage seems to vary, as they say.

I don't find that to be the case any more in V10. It was certainly true in previous versions.

The problem doesn't have anything to do with the aspect ratio or interlacing; it was most likely an issue with the scaling algorithm.

Repeating what was said above, many NLEs do a lousy job rescaling interlaced footage. A reasonable algorithm to rescale 1080i60 to 480i60 is to first use a frame-rate doubling deinterlacer to obtain 1080p60, rescale this to 480p60 with a high quality anti-aliased scaling algorithm and then weave the resulting frames back to interlaced 480i60. I don't know why the engineers who designed the NLEs chose lower quality algorithms for rescaling interlaced video. However, an amusing result of this decision is that blu-ray disks look even better than they would otherwise when compared to DVD.

But the scaling problem wasn't limited to interlaced footage; it was also apparent in progressive footage.

Pro 10 is doing a much better job at resizing than Pro 8 did.

Yes. No longer a problem.

Thanks Guys

Much appreciated...that's what I figured too but I keep seeing posts around saying that 720 is a better "fit"

Yeah, V10 certainly does a good enough job with 1080 down to SD (We are PAL of course) and I have no complaints!

The input is much appreciated

Chris

Repeating what was said above, many NLEs do a lousy job rescaling interlaced footage. A reasonable algorithm to rescale 1080i60 to 480i60 is to first use a frame-rate doubling deinterlacer to obtain 1080p60, rescale this to 480p60 with a high quality anti-aliased scaling algorithm and then weave the resulting frames back to interlaced 480i60..

Bingo!!!
We have a winner....
This needs to be a sticky for HD>SD downrezzing..I've got a homemade diagram explaining exactly what you've stated....
A picture speaks a thousand words..

Hi Peter and Eric

That sounds like perfect sense...I assume it would also work well for PAL where we are resizing to 576/50i???
Based on your assumption then guys, would it be an advantage to NOT shoot interlaced 1080 but in fact shoot (in my case) 1080 50P which should mean that the NLE already has de-interlaced double frame rate footage to resize???

If I look at project properties with 720 50P Vegas does show it at double frame rate. Sadly my cams don't shoot 1080 50P only 25P and according to the specs the 25P is over a 50i wrapper. If the NLE doesn't have the first step to worry about then this is probably why people are saying that 720 does a better resize as plenty of cameras can shoot 720 50 or 60P without any issues.

Chris

Well firstly, to be quite honest, I'm not a Vegas user, but have stumbled here because of the thread title.
It was something i spent days on end figuring out, and hoping to save other people the same headaches got.

The thing with 720 footage, is that it can be scaled down correctly..
It starts of as progressive, gets resized as progressive, and nothing gets hurt. Think of it as bicubic resizing your 2000x2000 photos...

With interlaced footage, since they are on two different fields for every frame, if you resize this, you will ruin the picture. You can't resize every 4th line...

Therefore split the fields, doubleframe, resize do SD (via interpolation), and reweave..
This is at the essence of proper resizing interlaced frames..

I do not no of any program, except for AVISynth (or within VirtualDub), that can do this.

As far as using 25FPS footage inside a 50i wrapper, the important thing is to have your NLE interperate the footage correctly.
But rule of thumb is, if it's progressive, don't deinterlace..As far as DVD and BluRays go, never deinterlace unless absolutely necessary. These are mediums which were intended for interlaced output..

Update:
Hey Chris...
Just read your post a bit closer.
Advantage of shooting 1080P (wrapper or not), your NLE should be able to rescale better. But i don't understand the whole deinterlace, doubleframe withing Vegas. I'm assuming that you think that Vegas is doing some extra work in the background?
Another advantage is for web export (the obvious)..
Unless shooting sporting events, or broadcast, or fast action cars, stick with progressive. Especially if you want to avoid headaches at the rescaling end..

Good luck..

But rule of thumb is, if it's progressive, don't deinterlace..

Uh, sorry? If it’s progressive, you can’t de-interlace.

Uh, sorry? If it’s progressive, you can’t de-interlace.

I believe if you have for example a 1080p30 clip (which is in a 60i wrapper) and then run a deinterlace on it using interpolate, then you will loose half your lines of resolution. De-interlacing progressive footage will make it worse.

If you have 30p wrapped in 60i, then it still is progressive, so what is there to de-interlace? De-interlacing only makes sense with true interlaced footage.

Vegas will automatically deinterlace 60i footage in a 30p project. This results in quality degradation of 30p progressive video wrapped as 60i unless you either set deinterlacing mode to none or reset the properties of the 60i wrapped clip to progressive. This is most important when editing for blu-ray.

Note that when scaling from 1920x1080 to 720x480 it is important to low-pass filter the image to avoid aliasing. The sharper the original image the more important the filtering. In particular, a blurry HD image may look better than a sharp HD image after both have been scaled to standard definition using some editing systems. An amusing comparison of anti-aliasing filters is given at

http://bvdwolf.home.xs4all.nl/main/foto/down_sample/down_sample.htm

In seems possible that the image blur introduced by accidentally applying blend fields deinterlacing to 30p progressive video wrapped as 60i could actually improve the quality of subsequent downscaling by the editing system.

Hi Adam

Quite correct.. the wrapper doesn't make it an interlaced file at all..it's progressive all the way so importing it into Vegas, one would leave Project Properties de-interlace box at none..which I tend to leave at default!

If I do shoot at 1080 50i I usually use VAAST's Upshift to transcode down to HDV (MPEG2) and then also tell Upshift to make the output file progressive so I always have progressive files in my project media. The end result from Upshift certainly seems to give a pretty sharp and pristine image but I have no idea what method NewBlue use (they make the software) to de-interlace. I wonder if V10 would de-interlace better than Upshift but I'm always in a quandry whether to use interpolate or blend...it's mainly weddings so there isn't really any high motion involved ... the help says use "blend" for high motion...do they consider high motion someone walking or more a Nascar race????

Yes, I know I should be shooting progressive but in PAL we only have 25P not 30P so you tend to get motion problems on frames unless you drop to 720 and use a double frame rate...with weddings I don't take the chance so they get done interlaced!!

Chris

If you have 30p wrapped in 60i, then it still is progressive, so what is there to de-interlace? De-interlacing only makes sense with true interlaced footage.

Yes Adam...
That's the problem. Some people actually DO deinterlace PsF footage.
Some softwares interperate PsF footage as interlaced, and process as such..

I see what you mean now. And I agree, deinterlacing PsF is a no-no.

the help says use "blend" for high motion...do they consider high motion someone walking or more a Nascar race????

Typically blend fields is only used when high motion is combined with a fast shutter. Even in these situations many people would prefer the stop-action strobing-effect caused by interpolate fields to the blurry double-exposures caused by blend fields. To deinterlace a wedding recorded at 50i with a shutter speed of 1/50 second, I would use interpolate fields.

And I agree, deinterlacing PsF is a no-no.

You basically would have to de-interlace before rescaling because in interlaced material the fields are incomplete images, they're missing every other line and regular rescaling algorithms can not cope with that (they assume an image with a homogeneous pattern of pixels).

From what I gather from the help, Blend fields would combine the lines of the two fields while interpolate drops one field then interpolates the missing lines in the remaining field. For 'PSF' material blend should be used. For rescaling PsF you'd de-interlace using blending.

Frans...

As far as the frame is concerned, it is considered progressive, regardless of what the sensors or internal mechanisms of the camera output...Deinterlacing the PsF frame should be avoided, since it is, in essence, a fully displayed frame, and not a frame consisting of two separate fields..

Afaik PsF is in interlaced format, each frame is split in two fields each with half of the lines missing. At least that is what I always believed interlace means.

From Wikipedia, the article Progressive segmented frame states
With PsF, a progressive frame is divided into two segments, with the odd lines in one segment and the even lines in the other segment. Technically, the segments are equivalent to interlaced fields, but unlike native interlaced video, there is no motion between the two fields that make up the video frame

For an optimal resizing of such a frame, the software (or other equipment) will have to combine the two segments or fields (whichever word you prefer), into one frame before resizing. The recombination is the same operation as de-interlacing, although given the progressive origin of the segmented (or interlaced) frame that operation is pretty straightforward.

Actually, "blend" is what you should NOT use if you have a lot of motion, because it'll be blending two fields, and in those two fields, moving objects will be far apart. This will require a lot of blurring, and you're almost certainly going to be seeing quite a bit of ghosting.

"Interpolate" is better because it works with single fields. It may end up having slightly lower resolution, but the other kinds of artifacts will be far less.

Many thanks guys

That's a neat summary and I can really see the sense in de-interlacing with interpolate as a wedding is fairly slow moving.

I was transcoding with Upshift as mentioned and looking at the final SD DVD you can pick up ghosting here and there occasionally so I guess Upshift uses the blend method. What seems strange is you have 3 de-interlace options... "copy from input", "progressive", and "interpolate"

To me the 2nd and 3rd options both deinterlace!! but it seems one uses blend. The next batch I'll skip using Upshift and just transcode the clips to CanopusHQ and use Vegas to do the interlacing!!

Chris

Hi

Hi Peter and Eric

That sounds like perfect sense...I assume it would also work well for PAL where we are resizing to 576/50i???
Based on your assumption then guys, would it be an advantage to NOT shoot interlaced 1080 but in fact shoot (in my case) 1080 50P which should mean that the NLE already has de-interlaced double frame rate footage to resize???

If I look at project properties with 720 50P Vegas does show it at double frame rate. Sadly my cams don't shoot 1080 50P only 25P and according to the specs the 25P is over a 50i wrapper. If the NLE doesn't have the first step to worry about then this is probably why people are saying that 720 does a better resize as plenty of cameras can shoot 720 50 or 60P without any issues.

Chris

I found 1080/50p looks even worse resized to DVD than starting with 1080i, it was a shock.

The main problem seems too be there is too much detail, when you start with 1080i and de-interlace and bob up to 50p, in the process you are approximating the image and essentially this removes detail. When it resizes down to 576i, the interlaced lines smudge up against each other and jaggies are less of a problem, and on de-interlacing the interlaced lines merge and knit together better. With 1080/50p straight down 576i those interlace lines sharply meet each other and you get jagged lines and it looks rough.

I think this is why 1080->720->576 is consider better looking, the extra resizing is helping remove detail.

I found better methods using AVISynth and a script, this works great on 1080i without having to de-interlace it, and does a good job with 1080p.

If you can, 1080p to 576p looks fantastic, like a good commercial film transfer if your footage can withstand being 25p.

I have tried resizing from 1080/50p to 576p at 50fps and it looks great, so you think just take odd lines from one frame and even lines from the other to get 50i and it should look equally great but interlaced, wrong, it looks horrendous, just too much detail in each interlaced frame.

Regards

Phil

so you think just take odd lines from one frame and even lines from the other to get 50i and it should look equally great but interlaced, wrong, it looks horrendous

Taking the "odd lines from one frame and even lines from the other" is not a good way to weave 50p into 50i. The usual method is

1. average line 1 and line 2 of frame 1 to obtain line 1 of field 1,
2. average line 2 and line 3 of frame 2 to obtain line 1 of field 2,
3. average line 3 and line 4 of frame 1 to obtain line 2 of field 1,
4. average line 4 and line 5 of frame 2 to obtain line 2 of field 2,
5. average line 5 and line 6 of frame 1 to obtain line 3 of field 1,
6. average line 6 and line 7 of frame 2 to obtain line 3 of field 2,
7. and so on

which has the advantage of being simple without introducing unacceptable aliasing or interline twitter.

What either Vegas or DVD Architect actually do to weave 50p to 50i is beyond me; however, there are reports that DVD Architect is better at such conversions than Vegas.

http://www.dvinfo.net/forum/what-happens-vegas/499039-anyone-tried-dvd-architect-5-a.html