FXAA in jME 3.0
FXAA in jME 3.1
without FXAA in jME 3.1
Note to readers who do not wish to fully open the images in multiple tabs and click back and forth… OP is reporting the FXAA is not doing anything in 3.1.
LOL, thanks for that Paul, I was feeling a bit like an idiot trying to compare the differences, it’s like “find 10 differences”
Hard to tell without more information.
I don’t remember any change on the FXAA filter for 3.1.
That said, it looks like the actual main difference is the lighting in your scene but it indeed look a lot smoother in 3.0…
For me there is definitely an effect in your picture in 3.1, with compared to without.
Maybe try to play with the parameters or use 2 FXAA…
Note: OP = original poster… that’s you. I was explaining to everyone else what you were reporting because you didn’t.
So, @nehon is right that there is at least a slight effect in the second image. Here is a triple expansion of the same area:
Click to expand as I increased the size 2x. They are in the same order as the original pictures.
And yeah, the way different lighting could have some effect on how well the FXAA works… but it is strange.
OP (as in Original Poster, as in person posting the issue), maybe try comparing the scenes with the same lighting setup?
Fixed in master.
The issue is that FXAA wants the input scene texture to use bilinear filtering, whereas FilterPostProcessor uses nearest filtering by default in 3.1.
you’re the man.
Hmmm… feels like this will mess with lots of post proc filters?
Wondering how this doesn’t affect bloom, light scattering, depth of field, etc…
But it makes sense,
always use the fastest as default, let users increase quality if necessary.
However this should probably be stated in the changelog.
Sure, but if it’s true that it broke all of those other filters then it would have made sense to test and fix them when changing the default.
I don’t know if it did. Many of them stopped working properly for me (like light scattering) some time ago and I’ve just never looked into it again.
All the filters are working for me?
It is quite rare that filters take advantage of bilinear filtering, and if they do then now there’s a mechanism for them to enable that.