To wet your appetite for motion estimation, here's a short
clip that was produced with Howler 7.2, starting from a short
rendering done in Carrara:
https://youtu.be/mZrGchWmHl4
- best viewed in large or full screen mode. The big
thing to keep in mind while watching this video: It took a few
hours to render the original short clip of 2-3 seconds in
Carrara, on CPU, a 2nd generation i7 (yeah, that's old now),
but it only took a few more minutes (!) for Howler to produce
the super-slow-motion version from it, using the Motion
Prediction Module.
Howler can run over 64 threads now, so if you have one of
those coveted new AMD Threadripper CPUs with 32 cores, 64
threads - wow! you're in for a treat!
But even without such high-end gear... Motion Estimation can
sometimes help by creating interpolated frames of better
quality than what you'd get wth the traditional frame
blending, where a strobing effect can easily be a side effect
of strong contrast between nearby pixels or groups of pixels.
It might even create an angle of a still frame that you may
end up preferring over the two original neighboring frames.
Here we'll have a look at a few practical examples.
Why do we care about motion estimated interpolation? The
need for Speed
This one is relatable if you're on a slow, old system, where
the GPU has just a few dozen or hundred cores, and you are
creating an animation with Puppy Ray that, while going fast,
still isn't fast enough. Sometimes, you can skip rendering
every frame, and later create fill-in frames with motion
estimation. It just may be good enough, and very likely will
be faster.
It's also a fact that you don't always have the ability to
re-render the scene. So let's say you have a short video clip
that last 3 seconds, at 30 fps, i.e. is made of 90 frames, but
your project requires it to be 6 or 9 seconds long, a slower,
longer lasting clip. Perhaps the clip shows a landscape in a
panoramic sweep from left to right, and it goes a little too
fast?
You can use traditional time-stretch techniques, and sometimes
that will be good enough. In many cases though, you may
notice a side-effect with frame blending: some areas of the
image may appear to be strobing, flickring a few pixels, when
they are of high contrast next to eachother.
Here is an example of what we might want to try the motion
prediction module for: a panoramic sweep on a landscape. This
one was rendered in Puppy Ray GPU, but it might as well come
from another 3D program, or even be recorded video footage.
What is traditional Time Stretching or Frame Blending?
coming soon
What is a Motion Tracker?
coming soon
What if I place many motion trackers all over the frames?
