Optical flow works by analyzing each pixel's movement from frame to frame.

Vegas' plug-in API only allows access to one frame at a time, which limits what can be done.

nice to know John

Interesting observations. So I am trying 2 other tests. I've uploaded another youtube video. This one at 25% speed.



And I will be uploading another done a different way.

The 25% slo-mo uploaded by Perrone does slo-mo the old-fashioned way, by simply duplicating frames. I assume that this was done in Vegas by turning off resample.

This approach provides spatially perfect slo-mo, but it lacks any feeling of temporal smoothness. This is exactly what you get if you slow down a movie projector (well, not exactly because of the projector's shutter ...). However, it is not "real" slo-mo as most people usually define it, namely where the motion appears to have the same temporal cadence as the motion in the normal-speed portions of the video, except things are simply moving more slowly. Usually (although not always) that is what people are trying to achieve. Over-cranking the camera (with the appropriate shutter) is the only sure way to achieve this. The other methods -- duplicating frames, blending frames, and synthesizing frames using motion estimation -- all have differing tradeoffs compared to that ideal of over-cranking.

The goal here (in this thread) is to find the best possible solution for those starting with 24p or 25p. I still need some real-life footage to play with if I am going to provide some help. Given the significant lack of temporal information in 24p and 25p compared to 50i and 60i, I am pretty sure that the best results will come from using motion estimation. However, I still believe that some of those who have gotten results from Vegas which they characterize as horrible or "sucking" probably are running into the interlacing/de-interlacing behavior that I described a year ago in a long and (to me) very interesting thread which Bob and others contributed to. I need to first verify if this is so (which is why I need some sample footage AND a VEG file that will produce "sucking" slo-mo) and I can then see if I can get to the bottom of this.

John, you are correct. I used 60p slowed down 25%. I have have not liked my results from 24p video in the past when using Vegas re-sampling, but I am going to play with that now. Both in Vegas and in another program.

This is a link to the same video that I slowed down outside of Vegas:



I suspect it is using the same methods as Vegas with re-sampling off.

JOhn, I'll upload some footge tonight mate, Ive just juiced my bandwidth posting a file to a client in singapore so im out of bandwidth during peak time. I'll let you know once its up.

Or maybe, just maybe, it was a colloquial joke. Sort of like using "dontcha" instead of "don't you." Of course that's prolly another of your pet peeves and we all should steer away from those lest we incur your wrath.

I'm kind of taking John's side on this although maybe not in this instance. What's rattled my brain is plans to teach Text as language in schools down here. The generation gap is one thing for old farts to cope with, adding a language barrier into the mix is going to be too much to cope with in our declining years.

Bob.

Beware of the man with the wooden leg that can sing.

Now there's a peeve to pet...

Ive been frequenting online forums and boards for over a decade now and the slang bit leaks now and then... as does the lack of spell check...

the point however isnt my posting etiquette, its about the fact that there is a REAL issue with progressive footage and considering 1080p amd evem 720p is becoming a standard for delivery, Vegas is VERY much behind in supporting those formats appropriately

Prolly is accepted as a word in the English language by the OED, which is usually regarded as the ultimate authority, in the UK at least.

It goes back further than you might expect - the OED's oldest citation is from H.G.Wells in 1925.

Mark

The OED, like most dictionaries, is meant to be descriptive, not prescriptive.

In English: a word's inclusion simply reflects popular usage.

Who's to say what's right?


[Although many think it's de rigeur to use only words from Webster.]

:O)

A lot of this is going over my head. I'd like to see someone do a "how-to" that describes the different ways to achieve slow-mo with footage and then list the pros and cons to each way. Anybody? :)

A lot of this is going over my head. I'd like to see someone do a "how-to" that describes the different ways to achieve slow-mo with footage and then list the pros and cons to each way. Anybody? :)There are three main ways to do slow motion:

1. Repeat Frames. To do this in Vegas, slow down the clip by setting the event playback rate (or "Ctrl-dragging" the right edge of the event) and/or adding a velocity envelope (you can do both in order to get REALLY slow motion). Then, right click on the event, and change the resample property to “disable. This provides very clear, sharp slow motion, but the motion is very jerky.



2. Blend Frames. You do the same thing as above, but don’t change the resample property and instead leave this at the default (which is "smart resample"). This provides very smooth motion, but the image is somewhat soft because all the extra frames between each pair of real frames are created by blending those two adjacent frames. This is exactly the same thing you would get if you put the same video on the track above, offset it by one frame to the right, and then varied the opacity level of the event on the track above. If you do this while viewing a single frame, you will see that you create intermediate frames between the two frames as you vary the opacity level up and down (the opacity level for each even is controlled by a semi-invisible line that runs along the entire top of each event). This is the default of how Vegas does slow motion. The results are predictable, but soft.

3. Synthesize new frames. This is being called "optical flow" in this thread, and perhaps that is the correct term. This is a far more complex process, and uses motion estimation algorithms to try to determine where each small group of pixels is moving from one frame to the next and then, using that information, synthesize intermediate frames. It is very difficult to come up with an algorithm that works for all different situations and all types of motion, so software that does this "trick" usually requires a lot of tweaking. However, when everything is set up correctly, the results can be stunningly good. On the other hand, the results can also be totally unwatchable. By contrast, the first two approaches always produce very predictable results.

Here is a YouTube video that shows about two seconds of original footage, followed by a 1/8 speed slo-mo of that same footage using each of the three methods described above:



You will doubtless initially find the final version (using MVTools) the most appealing, but if you watch it several times and look at the tassel on the hat, or the tip of her nose, you will begin to see "morphing" distortion and other artifacts. These can sometimes be minimized by tweaking the software (which I did not do) or by putting the same footage on the track above, slowing it down in Vegas by the same amount, and then doing Bezier masking to insert small portions of the Vegas slo-mo over the portions of the "good" slow mo that are failing.

Man, that is great. Thanks for spending the time doing that. You're right, the last one looks the best. The only one I've done is the default (ctrl-drag and leave resampling). Looks like MVTools is free and open source? It doesn't look too straight-forward though.

Thanks again for your time. I'm going to add this to my list of how-to's.

amazed again at the sharing of iknowledge and donated time by some on this forum. Great info JM

Thanks John I think your results are terrific

I always look to your comments with interest cause I know I am going to benefit a lot,unfortunately a lot of the technical stuff
Is beyond me but I always read it a few times to keep up

I looked up mv tools and there was a lot of different aps called mvtools

Can you direct me to the right one

Thanks once again for your imput

Rory

Sorry if it's a lamer question...

What John showed makes me wanting to experiment with this. I got MVtools 1 & 2, put the DLL files in AviSynth plugins folder.

Using VirtualDub I can get output from a simple script containing Version() or AviSource() but I am unable to use MVtools functions. Trying MVtools v1 or v2 names, I always get an error dialog box stating that the function doesn't exist. I also tried to force plugin load with LoadPlugin() but it ended in error too.

I fail to understand what I did wrong. Can a guru provide some help ?

Many thanks in advance.

Using VirtualDub I can get output from a simple script containing Version() or AviSource() but I am unable to use MVtools functions. Trying MVtools v1 or v2 names, I always get an error dialog box stating that the function doesn't exist. I also tried to force plugin load with LoadPlugin() but it ended in error too.You may need the FFTW library (http://www.fftw.org). You put this in the C:\WINDOWS\SYSTEM32 folder (for WinXP).

Here is the script I use for 50% slo mo on interlaced SD material. It assumes that you are using the "special" version of AVISynth which permits (with some plugins) the use of multi-cores. I get about 140 fps (i.e., 4x real-time) rendering speed with this script. If you are using the "regular" AVISynth, delete the two "SetMTMode" lines.

loadplugin("C:\Program Files\AviSynth 2.5\plugins\MVTools\mvtools2.dll") 



setMTMode(5)

AVISource("E:\Disney Films\101 Unbelievable Sports Plays0002.avi")

setMTMode(2,0)



# the following is for DV, change to tff for HDV

assumebff()

converttoyv12()



#separatefields()

source=Bob()

super=MSuper(source,pel=2)



backward_vec = MAnalyse(super,blksize=16, overlap=4, isb = true,  search=3 )

forward_vec =  MAnalyse(super,blksize=16, overlap=4, isb = false, search=3 )



#ml is important parameter for tuning out artifacts. Eliminate it and look at artifacts.

#Use values >100 to reduce artifacts.

#"num" equals result of 30 fps times 2 times slo-mo factor - This is for 50% slo-mo

#So, 60000 equals normal speed; 120000 equals 50% normal speed, etc.

MFlowFps(source,super,backward_vec, forward_vec, num=120000, den=1001, ml=100)



#Use the following when using bob

assumebff().separatefields().selectevery(4,0,3).weave()

assumefps(29.97,true)

Maybe I'm missing something here, but I learned long ago that you MUST disable resampling for slo-mo to look good in Vegas. Anytime you are trying to interpolate frames or blend frames you'll get crappy blurry frames. I've been doing lots of slo-mo in Vegas and generated video that is clear and sharp. If there is software that can make clear 'tween' frames in-order to slo-mo low framerate video, let me know. If you want professional slo-motion you need to shoot video at a high framerate... That's why I recommend the Sanyo HD2000 for amateurs as it can record 1080x1920 at 60 fps in progressive. 50% slo-mo on that video looks great in Vegas (just make sure resampling is disabled!).

but I learned long ago that you MUST disable resampling for slo-mo to look good in VegasWell, that depends on what you define as "look good." If you disable resample, Vegas will repeat frames and/or fields in order to get the slower playback rate. The result will look exactly as sharp as the original, so the spatial quality of your slow motion will be really good. However, the temporal quality will be terrible because it will begin to look like an old fashioned flip-book animation where you can actually see each individual picture.

The correct way to do slow motion is to overcrank the camera. There is no debate or argument about that point. Until recently, in the video world, such cameras were very expensive and rare. It is still difficult to find relatively inexpensive (prosumer price range) video cameras that can overcrank AND do it in high def. There are lots of camera that can do overcranked slow motion, but most of them do so simply by halving the resolution. You can do this same trick in Vegas, if you know how.

In case anyone missed it, I described the various ways of doing slo-mo in a post earlier in this thread:

Slo mo examples

and showed the results on a very simple clip. For my money, the disable resample produces the worst results simply because it doesn't look like slow motion. However, there are some instances were this is the effect you want, and in that case, by all means disable resample.

"If there is software that can make clear 'tween' frames in-order to slo-mo low framerate video, let me know."

If you don't mind the cost then After Effects can do exactly this:

http://www.vimeo.com/2866915


Like anything though pushed too hard or in situations where it cannot resolve the different parts it can go wrong:



Whilst it's true high speed cameras are the way to fly the good ones are not exactly cheap, I was quoted $2,500 / day + expenses + lenses but that is not the only cost. You will need a LOT of light if shooting indoors and you have to choose the type of lights with a lot of care. Also keep in mind that the high speed camera generate a lot of data very quickly, it takes a lot of time to offload and you need backups.

Bob.

PS: How to embed Vimeo???