First page Back Continue Last page Overview Text
I’ll start experimental results with a slide showing what our compact window look like. On the left, I show in black several optimal CW that our algorithm finds, the white dot shows the pixel for which this compact window was computed. On the right, there is the ground truth. Note that windows do not cross disparity discontinuities, that is for thin leaves we have small windows. In low intensity variation areas, like this and that large leaves, the window grows large to gather enough intensity variation statistics. So we find the kind of windows we need for the local approach to succeed.