Idea: Edge Clipping
by Will Harrison · in General Discussion · 08/27/2004 (5:38 pm) · 22 replies
(EDIT - Please jump down to about my 10th post below for newer idea)
I had this idea a while back, but I thought I would present it to all of you now and get some feedback.
I call it "Smooth Mask"... here it is in action:
Basically this extends the idea of normal maps (where you are only making the shaded surface look more like the way it was originally modeled)... now the silhouette can look smooth, the way it was made, thus eliminating that "low-poly" look we see too often. And, of course, the big benefit is that you're not adding more polygons to do it.
I'm not an expert coder, so I have no idea really how to make this work in a game.... but the principle is this: take the original shape and get its silouhette/alph-channel, then apply a Gaussian Blur with a 17.0-or-whatever-looks-good radius, then run that through a threshold filter (set at 128), and finally mask/stencil buffer the render of the model.
What do you all think? Is this doable? What would be the performance hit be for this added calculation/process? Anyone want to take up the challenge of doing this? Why hasnt this been done already?
I had this idea a while back, but I thought I would present it to all of you now and get some feedback.
I call it "Smooth Mask"... here it is in action:
Basically this extends the idea of normal maps (where you are only making the shaded surface look more like the way it was originally modeled)... now the silhouette can look smooth, the way it was made, thus eliminating that "low-poly" look we see too often. And, of course, the big benefit is that you're not adding more polygons to do it.
I'm not an expert coder, so I have no idea really how to make this work in a game.... but the principle is this: take the original shape and get its silouhette/alph-channel, then apply a Gaussian Blur with a 17.0-or-whatever-looks-good radius, then run that through a threshold filter (set at 128), and finally mask/stencil buffer the render of the model.
What do you all think? Is this doable? What would be the performance hit be for this added calculation/process? Anyone want to take up the challenge of doing this? Why hasnt this been done already?
#22
10/28/2004 (8:21 pm)
...and that's how new technology solution are born... ^^
Associate Ben Garney