Slo-mo revisited: questions for Sony engineers and others in the know

vitamin_D wrote on 3/31/2004, 4:42 PM
Hi,

I'm rendering out slo-mo comparisons pitting Vegas against Dynapel's SloMo. Vegas wins, hands down, but...

The render process has raised a few questions in my mind about how I could make the Vegas render look even better. For reference, here's the Vegas output in 1mbps QT:

http://ideaspora.net/slomo_vegas.mov

There's minor, noticeable "suttering" at the high-contrast edges in the render, and I'm wondering how I can remove them. Here's where I'm confused and seeking the help of the coders who know Vegas's rendering engine:

What generates a better slomo -- velocity envelope or ctrl+drag? What separates the two in terms of rendering and how Vegas interpolates the data?

Is there a difference between ctrl+dragging an event and right-click/properties/playback?

Does Vegas's slomo rendering improve relative to the percentage setting based on even or odd divisors? In other words, does a 27% velocity envelope setting look better or worse than a 25%? If so, why, and what figures should be aimed at when rendering slomo? I'm thinking about how photos look in Photoshop -- terribly blocky at, say, 53%, but at 50% they look wonderful.

Should motion blur be combined with supersampling when rendering out to slomo? If so, what setting would best preserve the integrity of the original footage while smoothing out the stuttering edges? I'm aware this question may be best answered generally, but it would be nice to have a general idea of where to go instead of pure trial-n-error.

Are there "tricks" to be had by rendering out footage to a higher or lower framerate and then rerendering to a different framerate? In other words -- BJ_M suggests rendering everything out to a higher framerate with supersampling set to max, and then rerendering to a lower framerate when mixing source media -- I'm wondering if this would improve things for changes in motion as well?

Thanks!

- jim

A note about the media and rendering seen above: the original clip was NTSC DV media captured with Vegas. I rendered the footage out with a velocity envelope set to 33% and the supersampling envelope set to max (8). To generate a progressive, final file, I rerendered the same media to Upper fields first, then re-inserted it in the Vegas timeline below the original slomo file (set to lower fields), set the opacity on the upper track to 50%, and then rendered it out using Vegas's NTSC DV template set to "field order: none." From there I rendered it out to QT at 1mbps.

Comments

johnmeyer wrote on 3/31/2004, 5:06 PM
I attempted to create better slow motion and came up with some ideas which I shared in this thread:

Slow Motion Settings

In partial answer to some of your questions:

I don't think there is any difference at all between Ctrl-drag, setting the number in the properties, and using a velocity envelope, at least not in terms of how the fields are created, blended, etc.

My understanding of slo-mo is that Vegas does this by blending adjacent frames. This blending is done if you have resample set to Auto or Force for the event properties. If you turn off resample, then no adjacent frames are blended, and frames are simply repeated instead.

Supersampling, I think, up-samples the frame to a higher resolution and then performs the operation (in this case adjacent frame blending), and then downsamples back down to 720x480.

The Best/Good settings in the Render As dialog change the algorithm used on frames that are interpolated. Most scuttlebutt on these boards seems to indicate that this only makes a difference on video that is generated within Vegas, such as keyframed titles.

The motion blur settings do pretty much what the title suggests. Using it trades off the "stutter" you are experiencing with a slightly softer, blurrier effect.

As I have said in several other posts, I believe that there must be a better product out there. The Dynapel Motionperfect product is the only one I know of (but there must be others) that uses the concept of motion estimation, which is at the heart of MPEG compression, to predict where each pixel in the synthesized intermediate frames should be, and then creates that new frame, rather than simply blending adjacent frames. I have had wonderful results on certain clips with Motionperfect, but only on progressive footage (actually, it was from film captured using the "Workprinter"). Motionperfect does not seem to work well with interlaced source material.

We've had discussions about slow motion many times on this forum. Perhaps this time someone will tell us about a new discovery of a product that does really great slow motion.

I know how slow motion is done in film (of course) but, if I had all the money in the world, how would I do really great slow motion? Do the pros use special high-frame rate cameras (I assume that is what is used for the ultra slo-mo I see on major league sport replays), or do they have some secret hardware/software?
vitamin_D wrote on 3/31/2004, 5:10 PM
Thanks for that lengthy and informative response, man. Incidentally, I had selected "force resample" in my renders.

As for a superior slomo product out there, I know of ReTimer, which will set you back a huge amount of cash but the results look fantastic.

I'm just interested in generating great media with Vegas because I've little money to spend on other things at the moment :/

- jim
vitamin_D wrote on 3/31/2004, 5:15 PM
Since Vegas interpolates frames, would it be wisest to render out the source media to a 30fps progressive .avi BEFORE attempting any slomo renders? I'm gonna try and see...
rs170a wrote on 3/31/2004, 5:27 PM
"Do the pros use special high-frame rate cameras...?"

Yup. They use specially modified Sony cameras & tape machines (probably DDRs now) that run at 90 fps. Rentals are a "mere" $2k/day from http://tinyurl.com/2opkq
vitamin_D wrote on 3/31/2004, 7:19 PM
Just answered my own question: render out progressive after speed changes and not before.

BTW, just to be fair I re-rendered through Dynapel's SloMo with a progressive source file and it still looked awful.

Oh well,

- jim
johnmeyer wrote on 3/31/2004, 7:53 PM
As for a superior slomo product out there, I know of ReTimer

Thanks. I'll take a look.

They use specially modified Sony cameras & tape machines

I'll check out the rental site (not that I'm going to spend that kind of money, but I'd love to know how the stuff works).

I re-rendered through Dynapel's SloMo with a progressive source file and it still looked awful.

vitamin_D, have you used their Motionperfect? It has quite a few additional features. If you click on this link:

Dynapel

and then click on Products -> Video Editing on the left side of the page, you can then click on Motionperfect. On the page that comes up, you'll find a link that compares "Slow Motion" with "Motionperfect."
johnmeyer wrote on 3/31/2004, 8:06 PM
Vitamin_D:

Check out the tutorial on this site:

SLO-MO TUTORIAL/TIP

You said you were going to do some tests by rendering as progressive. This is in essence what this tutorial suggests.

However, the interesting idea is that they talk about using the deinterlacing in Premiere's film plug in. The Vegas film look plugin doesn't have any deinterlacing filters, but it does make me wonder if there might be more than one way to deinterlace within Vegas, and if so, whether one or the other might be better for this application.

I will be very interested to hear what results you get when you deinterlace, slow things down, and then render.

P.S. Vitamin_D: You said: Just answered my own question: render out progressive after speed changes and not before.

Does this mean that your final project is going to be progressive footage? If not, when do you render back to standard NTSC interlaced?
vitamin_D wrote on 3/31/2004, 8:33 PM
it does make me wonder if there might be more than one way to deinterlace within Vegas, and if so, whether one or the other might be better for this application.

I do all my de-interlacing like this.

I will be very interested to hear what results you get when you deinterlace, slow things down, and then render.

It doesn't work any better in Vegas this way, and in fact looks worse. The de-interlace method linked to above provides the best image in terms of removing interlace artifacts while preserving resolution -- I haven't tested Vegas's slo-mo using blend or interpolate fields otherwise, but as they lose resolution I'm not about to :/

I own MotionPerfect but had assumed that its render engine would produce results identical to SloMo. Who knows?

- jim
johnmeyer wrote on 3/31/2004, 10:05 PM
I do all my de-interlacing like this.

Wow! I see how it works, but is it any better than simply setting the Field Order to "progressive" when doing a "Render As?" That would seem to be the simplest. However, maybe this loses resolution.

Another method was suggested almost two years ago. Read about it in this thread:

A better slow motion

It involves rendering to a 60p uncompressed file. I just tried it and it worked. I'm still experimenting with the results to see if I can get better slo-mo than using the techniques I discussed in the link I gave to you earlier in this thread.
farss wrote on 4/1/2004, 4:25 AM
Trying to get decent slo-mo in video is difficult to impossible. You are trying to create information that isn't there to start with, sorry to sound like the devils advocate here but I think you need to start out with that in mind. The pros spend vast sums of money just to get this stuff to look as good as it does and they sure don't like to spend more than they have to. To give you an example, one of the complelling reasons for shooting 24 Hours on 35mm over video was because they wanted to use speed ramps and there is no video camera system at any price that'll do it. Piece of cake on film camera, forget about it on video. Now that must have made a huge difference to the shows budget, particularly given the tight shooting shedule.

So OK, short of a big budget what could you do to get it to look as good as possible on video. Now I haven't tried any of this so it's just a few thought I'm making up as I go.

Firstly I suspect it's got to start with how you shoot the material, even with high end slo mo software that uses motion vectors the cleaner the image you can give it to start with the better it's going to work, fuzzy images full of noise are going to spin out this process. Even without motion estimation the interpolated frames are going to be crappy. So here's my suggestions for shooting for slo mo:

1) Fast shutter, this reduces motion blur during shooting but also causes the camera to wind up the gain and hence the noise and reduces depth of field so the more light the better.
2) Reduce all unnecessary motion. If possible keep the camera locked off. This way the only thing really taking a hit from the interpolation is the thing you're trying to slo mo, even if it gets a bit crappy looking in the process the rest of the frame should be OK. The eye is then more forgiving, if most of the images look OK but the moving bit doesn't well we kind of expect that anyway.

The aim is to end up with a sequence of fields that look like perfect stills, this way the interpolation process can produce the best possible results. Then when you add motion blur afterwards your not adding blur to blur which has got to end up just looking like mess.

Hope I've given you guys some food for thought here.
vitamin_D wrote on 4/1/2004, 8:15 AM
farss --

Glad you vocalized those points -- I've noticed that slomo packages like ReTimer use hi-res footage in most of their examples which offers a bit of a skewed view of the program's abilities.

- jim