ADOBE AFTER EFFECTS: Forum Expressions Tutorials Creative Cloud

Re: camera projection ray-traced vs classic 3D

COW Forums : Adobe After Effects

Respond to this post   •   Return to posts index   •   Read entire thread

Andrew Somers
Re: camera projection ray-traced vs classic 3D
on Aug 16, 2018 at 10:28:18 am

To follow up Mark, I tried to recreate the issue — and couldn't till I read your post again about having the image at 1% scale. I see that the issue is identical for POINT and SPOT lights, and is NOT there with PARALLEL lights (notice also that parallel lights do NOT have a "shadow diffusion" setting).

Basically, a ray tracing engine creates depth-of-field, motion blur, and antialiasing by shooting several rays out of each pixel. As Stu Maschwitz puts it, the AE Ray Tracer is "a brute-force multi-sample renderer".

My assumption is that the "point" light is 1x1 pixel square. Considering the spot light looks the same, I imagine that the math there also starts with a 1x1 pixel light (this makes sense for computation, I'd venture that a 0x0 point would probably not be used due to computational difficulties or artifacts).

Consider that if you scale your image to 1%, that means that 100 x 100 pixels of your original image now fit in the same size area as that 1x1 point light. And when antialiasing is turned on, multiple rays are created.

So it would seem that even with the light's shadow diffusion set to 0.0, there is always *some* due to the anti-aliasing multi-ray-sample nature of the system, not to mention the "math precision" (or lack thereof) resulting from *both* scaling small and being very close to the light.

In an attempt to quantify this, I did the following experiment. I created a 100x100 checkerboard, where each square is 10x10. Then setup a scene where the white solid was 1000 from the point light. And then the checkerboard was various distances and scales from the light, with scale 100% being halfway between the light and the white solid.

The second test was just one square, set 5 from the light and zoomed in to show the detail, with only a minor scale change between 8% and 12%.

Here are the results:


Interpretation: I think this may indicate the point light sources is 1x1 pixels, because at 10%, each square is 1x1 of an unscaled pixel, and in this one each black square is a nearly solid dot, and it rapidly degrades as the scale/distance decreases. This is echoed on the right where at less that 10% (meaning the square is 1x1) it becomes a fading dot, but the square shape becomes more apparent as we go larger than that.

As to fixing the problem for your project - is there a reason you can't move the light farther back, or put the image layer closer to the white?

My suggestion would be to scale no smaller than 50%, and place the light and image as needed to accommodate that (or larger) scale. Alternately use a parallel light, though that might not work for the rest of what you are building.

Andrew Somers
VFX & Title Supervisor

Posts IndexRead Thread 

Current Message Thread:

© 2019 All Rights Reserved