Recently in 3D Computer Graphics Category

Maxwell Render Photocomposite based on the Medieval tapestry Unicorn in Captivity

As an experiment I divided the product of two images (A and B) by one of the factor images (A) to unlock the other factor image (B)

I grabbed the images from my own web site, so they were 8-bit sRGB compressed jpegs to begin with.

You can see some round-off error in the quotient image.

I keep meaning to use this to divide a photo by a white diffuse Maxwell simulation of the photo in order to derive a good texture map that when lit will look pretty much like the original photo. The goal is to avoid re-lighting the lighting already in the photo.

You could heavily watermark your images on your web site and provide certain people with the key image that unlocks the watermark.

Collaboration with Italo of UTI Crew - Virtual Graffiti - initial test

Back when I was just starting out I worked on a project where my boss used to infer the focal length of a complicated lens snorkel system by photographing a 1 foot plastic cube. He did this so that we could match move CG and motion control robotic arm camera live action. (Nowadays there are better match moving solutions, but this was a while ago)

It occurs to me that this cube thing would be handy for matching CG to still photos, especially in a zoom lens where the focal length is not what the lens claims it is (as in the Nikkor 70 - 200 VRII)

Question:

Someone hands you a photo of a glass cube centered in frame and oriented with its face flat to camera.

The front face takes up 80% of the width of a Nikon FF image. The back face takes up 70% of the width of the image. The actual cube is 1 foot on each edge. The image is in 'landscape' aspect so the longest side is the width of the frame.

You have no more information.

Q1: How far is it from camera to cube?

q2: What is the focal length in mm of the lens?

---------------
Various answers are here. So far I have:

Distance to camera = 7 feet

The 133.35 mm comes from the fact that the front face is 0.8 of the frame width and the back face is 0.7 of the frame width so the back face is (0.7 / 0.8 = 0.875) of the front face. 0.875 times half a foot in millimeters is 133.35 mm

Front face projection is 80% of 36 mm = 28.8 mm
f = ?

Front face is 1 foot
D = 7 feet

1 : 7 :: 28.8 : f

Lens focal length f = 28.8 * 7 = 201.6 mm

So I divided this image

by its own perfectly fitting alpha channel

and I got this mess.

I was expecting this mess.

And I was expecting it to replicate the behavior of the John Knoll Unmult After Effects plugin.

Oh well, I hear The Gimp does it.

UPDATE

It turns out that in the mathematical expression you want the RGB in the numerator and the ALPHA in the denominator, but in Photoshop you want the alpha image in the upper layer.

Success. I think. :)

So if you layer mask that hot mess above by the original grayscale alpha key at the top of this thread then all the ugly parts go away and you should be left with a beautiful comp.

-- using a layer mask. (which you could 'apply' if you wish)

The side on the right has the ugly RGB element which makes the nice comp results.

(more conversation here)

UPDATE

There is an old use of the divide blend mode from 1983 that I know of in the Disney / John Lasseter cg test for a film based on Maurice Sendak's Where the Wild Things Are.

(youtube)

It involved drawing shapes with hard lines, blurring them to look like CG, and dividing the blurred shapes by their own masks so that the interiors had nice soft 'shading' but the blur didn't introduce darkness around the edge of every shape because the edges received equal blur and were given the 'divide' treatment.

The interior of the masks were white. Anything divided by white (1) is unchanged (left blurry)

The edges were normalized back to their original colors by being divided by the soft masks.

The divide blend mode is written to skip pure black pixels, as division by zero is undefined.