Added super-sampling when generating reflection cube maps to improve quality.
Here are some screenshots comparing build 240 with 241:
And now a comparison between the screenshot of the Gorgeous Trailer and build 241:
Much better! Thanks!
Earlier today, I came up with a concept for a method to potentially filter out artifacts in the reflections caused by bright sources of light that are far away from a reflection_probe entity and also well outside the boundary set by it's "Maximum Distance" parameter. But before I get into the details, here are a couple of screenshots that show what sort of artifacts I'm talking about:
I refer these as artifacts because they are almost always misaligned. This appears to be caused by the fact that those sources of light are far beyond the imaginary sphere created by the reflection_probe and controlled by its Max Distance property. To filter these parts of the reflection out, I propose one of the following potential solutions:
While generating the cube maps, render a thick black fog that starts just after the Max Distance and quickly fades to opaque.
While generating the cube maps, generate an alpha channel for the cube maps based on the depth (distance) of each pixel.
Because reflections are blended with the environment additively, black areas appear invisible. By rendering a black fog that hides distant objects, you can filter those objects out of the reflection.
The idea that uses the alpha channel is basically the same concept as the one that uses the fog except that it has the added potential of being able to also filter out objects that are too close to the reflection_probe center, although this is likely to be less useful.
A couple of parameters could be added to the reflection_probe to allow map authors to control this system. Assuming we're using the black fog method, they'd be something like this:
Fog Begin - Distance at which the black fog is fully transparent. (Value should be somewhat greater than Max Distance for best results.)
Fog End - Distance at which black fog is fully opaque. (Set to 0 to disable fog.)
This is how I originally implemented it, but the mappers requested it work the way it is right now so I changed it.
Interesting. So, how close was my description to the actual way that it worked?
I'm guessing the mappers requested the change because the filtering could not be disabled. Perhaps, if there's time, you might consider re-adding this functionality in such a way that allows mappers to configure it or disable it for each reflection_probe. I think the added flexibility would be useful. For ease of use, the default parameters could be set so that the filtering is disabled by default and only gets enabled if the mapper customizes the filtering distance values.
I want to get involved in some mapping for NS2 soon (that is - if I can find the time, which is proving quite hard).
Comments
I'm guessing the mappers requested the change because the filtering could not be disabled. Perhaps, if there's time, you might consider re-adding this functionality in such a way that allows mappers to configure it or disable it for each reflection_probe. I think the added flexibility would be useful. For ease of use, the default parameters could be set so that the filtering is disabled by default and only gets enabled if the mapper customizes the filtering distance values.
I want to get involved in some mapping for NS2 soon (that is - if I can find the time, which is proving quite hard).