#SIGGRAPHAsia | #SIGGRAPHAsia2022











SIGN IN TO VIEW THIS PRESENTATION Sign In
Scalable multi-class sampling via filtered sliced optimal transport

SessionSampling and Reconstruction
DescriptionWe propose a continuous domain formulation of Wasserstein barycenters
for multi-class (-purpose) point set optimization. Our formulation is sys-
tematically derived to handle hundreds to thousands of classes for different
sampling applications. We develop a practical optimization scheme that is
closely paired with our formulation. To demonstrate the generalizability of
our framework beyond sampling, we formalize the problem of blue-noise
error distribution as a multi-class problem. This helps establish a direct
connection between sampling, reconstruction and perceptual filtering in
rendering. The resulting formulation provide error bounds on the perceptual
error which, when optimized for, gives screen space blue-noise error dis-
tribution. We demonstrate the effectiveness of our framework on different
sampling applications like stippling, object placement and rendering.
for multi-class (-purpose) point set optimization. Our formulation is sys-
tematically derived to handle hundreds to thousands of classes for different
sampling applications. We develop a practical optimization scheme that is
closely paired with our formulation. To demonstrate the generalizability of
our framework beyond sampling, we formalize the problem of blue-noise
error distribution as a multi-class problem. This helps establish a direct
connection between sampling, reconstruction and perceptual filtering in
rendering. The resulting formulation provide error bounds on the perceptual
error which, when optimized for, gives screen space blue-noise error dis-
tribution. We demonstrate the effectiveness of our framework on different
sampling applications like stippling, object placement and rendering.
Event Type
Technical Communications
Technical Papers
TimeFriday, 9 December 202210:30am - 12:00pm KST
LocationRoom 324, Level 3, West Wing




