They can use a different texture projection for that purpose. Real time
occlusions maps are generally much simpler and lower in resolution than
color textures, and as a result, have very little detail. It's more about
getting the essence of occlusion than pinpoint pixel accuracy. For example,
a color texture might be 2048 x 2048, but an associated occlusion map may be
512 x 512 - 1/16th the resolution of the color map. For NPR content,
occlusion data is usually modified by hand to smooth out irregularities
making physical accuracy of the sampling not terribly important.
Assuming geometry resolution is there, it would be more efficient to use a
vertex color channel for occlusion data. Would also avoid the need for
non-overlapping UVs and associated headaches of having to deal with that
restriction for workflow and tools. Addition by subtraction.
Matt
Date: Tue, 9 Jun 2015 10:58:11 +0200
From: <[email protected]>
Subject: Re: GDC presentation: Guilty Gear Xrd and Softimage
To: <[email protected]>
For the anime content he's showing on screen, there's really no reason to
have as many tiles as shown if they all consist of the same dimensions and
color. A single tan tile, for example, consisting of only a few pixels in
U and V will serve just as well and consume much less memory allowing
saved resources to be used for other purposes. UVs can overlap the same
physical
space and reuse the edge for the ink line. UV placement is independent of
image resolution.
I guess that's exactly why they do it like this - to avoid overlaps.
They mentioned using texture based occlusion I think - so they would need
non-overlapping UV's for that.
