I can't help you with the codec problem, but I did spend some time just now looking at the video examples on the main product page:

Respeedr Product Page

I have spent many dozens of hours evaluating, learning, and using various slow motion tools including MotionPerfect, Twixtor, and several free tools available for AVISynth. I am very intrigued by Prodad's claim that they are using a mixture of "sophisticated frame-blending, optical-flow, and keyframe technologies." Optical flow (sometimes called motion estimation) produces amazing slow motion for coherent objects (like a train moving across the camera's field of view), but breaks down horribly with opposing motion (like a marching band, with everyone's legs going backwards and forwards as the band marches by the camera). It also has problems with objects entering the field of view close to the camera, and really fine-structured vertical detail, like panning across a picket fence.

On the other hand, frame blending (which is what Vegas uses) never produces any unwanted, obnoxious artifacts, but the results are always very blurry and soft because adjacent frames are simply being added together to produce the intermediate frames needed for the slow motion.

The promise that these two technologies can be used together is a really good idea, and I've actually done this manually in Vegas. To do this, I create, outside of Vegas, an optical flow version and then also create the same slow motion inside of Vegas by changing the playback speed for the event. I line up the two versions on two different tracks with the optical flow on top. Then, everywhere I see a bad artifact, I just create a quick mask to let the Vegas version show through, but only for the small portion of the frame where the optical flow has failed. It works well, but is a lot of work.

However ...

I just went to the ProDad Respeedr page I linked to above and played all the examples.

After being very enthused at the possibility of a nifty new tool, I am sorely disappointed with the results.

In some examples (like the very first video with the family running towards the camera), they seem to let the frame blending take over, and the results don't look much better than what you can achieve inside of Vegas. Watch the little kid's left arm (he's on the right side of the frame) as it swings back and forth across his body. It is clearly frame blended, which is good. However, so is almost everything else.

I wonder what settings they provide to control when optical flow is used rather than frame blending.

I then watched the second example, which is an aerial shot of one UAV photographing another UAV in flight. Watch very closely, just after the clip goes into slow motion and the UAV crosses the field of view from right to left. Look at the landing support structure and watch how it "breaks apart" in a classic optical flow failure. This is exactly the sort of thing that I would mask out in the technique I described above, but which ReSpeedr failed to sense and correct. This is the situation I was hoping they had figured out how to switch part of the slow motion to frame blending.

Clearly the optical flow works just fine on the easy stuff. The tower shot of the freeway is the simplest possible use of optical flow, and the results are excellent.

However, the last video which is the pit crew at the race track, is an excellent test of the technology because it shows people walking across the field of view. Even though they are at some distance from the camera, which is usually easier for the optical flow technology to handle, their legs still "break" quite a bit, and the result in unsettling.

So, unless someone can show better results, I am going to save my time and not download, install, and evaluate this any further. I am still going to stick with AVISynth and MVTools2 which is still the best solution I've found, even better than the very $$$ (or €€€) Twixtor.

Thanks John. I got the ProDad invite too. But after reading your valued views I quickly put my CC back into my wallet.

I''ll stay with the inferior to AVISynth, £££ Twixtor and can only dream of the day I'll be able to wrangle the MVTools.

Hey hooo.....

Grazie

Prodad is responsible with its codec packs, to the best of my knowledge, in that they install at the application level, not the system level, so they are used only by Prodad. Registration of codecs used in Vegas should not be affected. This is different than installing a codec pack on your general system.

That being said, I've hesitated over Prodad's choice of codecs when contemplating trial versions. ffmpeg would be far easier on my peace of mind to k-lite, which has earned itself a nasty reputation.

I also got the invite and watched the samples. Thanks for pointing out those glaring discrepancies. that I didn't see. That is quite distracting to say the least.

I wonder what settings they provide to control when optical flow is used rather than frame blending.


In playing with it, they have auto, blend and optical flow options. It seems to be one or the other is used and auto probably picks one depending on how much slow-down you select. There is no indication it switches back and forth within a given work area. You can define multiple "work areas" and each can be optioned differently.

Anyway my quick test shows it similar to twixtor in my condition. GoPro mounted to handlebar on mountain bike with lots of trees and brush in the background. Plenty of problem conditions. Bouncing and moving camera and subjects moving across a busy background.

Better than nothing and $80US is worth it for me. I would not pull the trigger for twixtor price given what result I can reasonably expect in my conditions. I don't expect high speed cam results. Maybe with a rider against a plain wall or clear sky.

If someone wants to post some difficult footage that they'd like to slow down to 25% or less of normal speed, it would be interesting to see the results several people could achieve. We did this several years ago with a clip of a football (soccer) header into the goal and it was very instructive to see the differences.

Here is an update to a slow-motion example I did four years ago. This shows the original clip followed by three examples of 25% slow motion. The first is Vegas with resample disabled. This is followed by Vegas with forced resample. Finally, the same clip is slowed down using MVTools2 inside of AVISynth. Watch the blur on the legs and the smoothness of the motion. Note that the original footage is somewhat blurred because of the 1/60 second effective shutter speed of the video camera.

About a month ago Petr Schreiber posted a really sweet front end GUI to VDub, Avisynth, and MVTools2. The GUI just makes slo mo really easy and smooth and in my view is as good as anything. I've posted a simple but really tough example of the GUI in operation compared to Vegas on Vimeo at https://vimeo.com/79010380. All the programs are open source. No need to download codec packs. I'd have to say I'm a little suspicious of the timing of this Mercalli release. You can download the GUI from www.thinbasic.com/community/showthread.php?t=12274&highlight=petr.

Instructive example.

Walking/running has to be a tough thing for an algorithm tracking the movement of pixels. Optical flow. When the two legs cross each other you likely have a zero(near zero) contrast situation and easily lose tracking until the legs separate again.

I'd have to say I'm a little suspicious of the timing of this Mercalli release.


Can you elaborate?

It's just about enough time from when someone open source works out how to do something for someone else to bring out a commercial version of the same thing. Not saying they did, might be coincidence, might be quite different code and I doubt we'd ever get a look at it.

It's just about enough time from when someone open source works out how to do something for someone else to bring out a commercial version of the same thing


Optical Flow exists in many commercial products. Some for a while now. Twixtor, Apple motion, Boris. Prodad now coming out with one is just joining the club. Prodad also has a history with code that tracks pixel motion between frames with their stabilizer. To me the slo-mo interpolation is a logical extension to pixel/object tracking ability.

Rainer, in the Vimeo video you posted the link to, the Vegas example looks much better to me. The VDub, etc. slowdown has lots of artifacts in the background around the moving horse and the rider. Also the horse's legs are breaking up as they move. The movement in the Vegas slowdown isn't as smooth as the VDub, etc. slowdown, but doesn't have the artifacts. So, to me anyway, the Vegas method looks better.

I interpreted your post to say that the VDub, etc. method produces better results than Vegas, so maybe I am seeing it wrong or misinterpreted your post. Can you comment?
Thanks.