Ben's Comp Newsletter: Issue 044
|
|
Hey,
Hope you had a productive week --
enjoy this week's
newsletter!
|
|
Field-of-view Calculator.
At
least a couple times a year I find
myself needing the equation to
calculate a camera's FOV (to
have a card perfectly fit &
travel within a camera's
frustum, for example), and
I can never remember how off the top
of my head.
Embarrassingly, Frank
Reuter made a
calculator way back in 2011 that
helps to figure this out, and I only
discovered it in the most recent
instance of needing the
aforementioned equation.
Click
here for another
resource which shows
the underlying
math.
|
|
A hack to filter DeepMerge
operations.
Unless
you're Compositing at Weta,
Nuke's Deep tools are quite
barebones and don't allow much
flexibility. An issue I have
commonly encountered is a lack of
image filtering when DeepMerging
two Deep images together, which
causes crunchy edges. You could
correctly argue that this comes from
a lack of deep samples in rendered
images, but this is often necessary
to reduce render times out of
lighting.
When trying to
DeepMerge certain
furry characters into a
scene with all sorts of things
flying around, I encountered
situations where fur was doing
things fur shouldn't be doing. I had
to figure out a solution to
DeepMerge patches back
into said fur to fix the issues, all
while retaining nice soft edges.
What I came up with was essentially
a hack that utilizes the
DeepExpression node to blur
the intersecting pixels from the A
and B pipes by 1 pixel (as seen
in the gif above). As a
nice bonus, blurring further
actually helped to get a sense of
overlapping fur!
I have wrapped this solution up into
a WIP gizmo for you to check out,
should you be interested. It's a
gizmo I intend on developing into a
more robust solution down the line,
so please reach
out should you have any
input, or another tool that solves
this issue in a better way!
|
|
Quick Tip: Disable geo when you're not
using it.
A
common complaint I hear among
compositors, especially those
working on large shots is: "My
script is so slow". There
are many things to consider
when trying to speed up a cumbersome
script, but the sneaky culprit
usually reveals itself in the form
of a ReadGeo node importing geometry
over the network.
Even when you're importing basic
LIDAR information to perhaps project
cleanup onto, or help create
occlusion shadows, I've found that
the performance of reading geo into
Nuke can be slow. Most of the time,
this is a necessary evil, but can be
easily dealt with by keeping a few
things in mind:
-
If
you're not using the
geo, delete it from your
script! Even if
it's hanging out off to the
side and not being used,
Nuke will still read in the
data, and it will continue
to slow down the speed
you're moving around your
Node Graph.
-
Utilize
localization.
When you're not having to
pull massive amounts of data
through the network, things
naturally speed
up!
-
If
you're using heavy,
sub-divided
geometry,
perhaps when using a
StickyProject, do
your work and then take
advantage of $gui to
only enable the geometry
during render
time.
-
ImagePlane
gizmos come
with the same, albeit less-severe
warning, as they
will slow down your script
if you're using too many of
them.
Hopefully
with this knowledge in your pocket,
you'll have another useful idea to
call upon next time your Nuke script
grinds to a halt, so you can
get back to working speedily &
efficiently in no
time.
|
|
An Interactive Introduction to Fourier
Transforms.
Fourier
Transforms allow us to break down
certain datasets into their sine
wave components. This has many
implications & applications in
many industries involving sound and
imagery, including audio &
image-related compression,
reconstruction & filtering. The
following article does an
amazing job of describing this
fundamental concept in
a non-mathematical, easily-digestible
way.
You can see a visual representation
of your images' frequencies in Nuke
using the
FFT
(Fast Fourier
Transform) Node. This is hidden by
default, and is most easily accessed
by hitting
x with
your cursor over your node graph,
and typing
FFT.
Should we ever be working with
plates shot on old film stock
(i.e. lots of damage),
converting the image into this space
could allow us an easy solution for
removing dust & scratches, by
finding obvious flaws in the
frequency's pattern and
clone-painting them out, before
converting back to our regular RGBA
image with the
InvFFT node.
If you would like to discover more
about Fourier Transforms, I have two
additional resources to share with
you:
-
This
video by
Smarter Every
Day on
YouTube does a similarly
great job of explaining the
basic concepts behind
Fourier
Transforms.
-
This
article breaks down
the math behind what's
actually going on with
Fourier Transforms, as
they relate to
images.
|
|
Did you find this
newsletter informative?
Have
you created, or do
you know of any
outstanding
Gizmos, Python
Scripts or Tutorials
that you would like
to share with the
global Compositing
community?
Please reply
to
this email, and I
will do my best to
include it in a
future issue of this
newsletter.
|
|
|
Support on Patreon
Thankyou
to the following
supporters
|
|
|
Adam Kelway
Adrian Winter
Hugo's Desk
Julien Laperdrix
Lee Watson
Kim
|
Michael Loithaler
Tiscar Coig
Vincent Desgrippes
William Towle
+ 2
others...
|
|
|
|
|
