Copyright © 2013 by the Association for Computing Machinery, Inc.
Permission to make digital or hard copies of part or all of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for commercial advantage and that copies bear this notice and the full citation on the
first page. Copyrights for components of this work owned by others than ACM must be
honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on
servers, or to redistribute to lists, requires prior specific permission and/or a fee.
Request permissions from Permissions Dept, ACM Inc., fax +1 (212) 869-0481 or e-mail
permissions@acm.org.
I3D 2013, Orlando, FL, March 21 – 23, 2013.
© 2013 ACM 978-1-4503-1956-0/13/0003 $15.00
Hybrid Transparency
Marilena Maule
∗
UFRGS
Jo
˝
ao Comba
UFRGS
Rafael Torchelsen
UFFS
Rui Bastos
NVIDIA
Figure 1: Order-independent transparency (OIT) images generated in real-time using hybrid transparency. The maximum depth complexities
are: 588 layers for the power plant, 671 layers for the hair, and 66 layers for the 11 dragons. Our hybrid approach closely approximates OIT
by storing only 16, 4, and 8 of the transparent layers (left to right), combined to an approximation of the remaining transparent layers.
Abstract
Hybrid transparency is an approach for real-time approximation of
order-independent transparency. Our hybrid approach combines an
accurate compositing, of a few core transparent layers, with a quick
approximation, for the remaining layers. Its main advantage, the
ability to operate in bounded memory without noticeable artifacts,
enables its usage with high scene complexity and image resolution,
which other approaches fail to handle. Hybrid transparency is suit-
able for highly-parallel execution, can be implemented in current
GPUs and further improved, with minimal architecture changes.
We present quality, memory, and performance analysis and com-
parisons which demonstrate that hybrid transparency is able to gen-
erate high-quality images at competitive frames rates and with the
lowest memory consumption among comparable OIT techniques.
Keywords: order-independent transparency, visibility determina-
tion, real-time rendering
1 Introduction
Order-independent transparency (OIT) rendering has received sig-
nificant research, but it is still considered an open problem. The
main difficulties are memory consumption (usually unbounded) and
processing cost (highly-varying with scene depth complexity). Cor-
rect rendering of scenes with transparent objects involves process-
∗
e-mail: mmaule@inf.ufrgs.br
ing transparency layers in depth-sorted order—i.e., either front-to-
back (FTB), which is how a viewer perceives a scene, or back-to-
front (BTF), which is how light traverses a scene. The seminal
work on alpha compositing in [
Porter and Duff 1984] defines the
algebra to combine two fragments, given a depth-order relation be-
tween them. The equations it defines are fundamental for proper
compositing of transparent layers under the sorting requirement.
When looking at a scene with transparent objects, in FTB order,
we notice that the objects behind a transparent layer get dimmed
by the opacity of the layer. When looking through multiple trans-
parent layers, the dimming of visibility accumulates successively at
every layer. This leads to the conclusion that, after a given thresh-
old, the remaining layers became so dimmed that make a very small
contribution for the final blended color. This suggests splitting the
transparent layers of a pixel in two classes: a core and a tail. The
first transparent layers, close to the viewer, tend to have high impact
on the color of a pixel and define the core of the blended fragments.
Remaining layers (away from the viewer) have lower-and-lower im-
pact, and define a tail of visible fragments.
Hybrid transparency (HT) is based on these observations, combin-
ing an accurate method for computing OIT on k core layers of each
pixel and an approximate (but very fast) method f or the remaining
layers (tail). Both methods handle order-independent rasterization,
but only the core requires sorting. HT presents high image quality
and no visible artifacts (the viewer gets the correct notion of layers
spatial relationship, without discontinuity artifacts, e.g. noise), as
shown in Figure
1, but runs in bounded memory with two geometry
passes. HT provides real-time frame rates, competitive with high-
performance techniques which use unbounded memory—i.e. prac-
tical implementation of adaptive transparency [
Salvi et al. 2011].
We summarize the main contributions of our work as follows:
• a practical method for OIT that can run in bounded memory
in today’s GPUs;
• rendering of larger transparent scenes at higher resolutions
than current methods can handle, at competitive frame rates;
• no visually-noticeable artifacts (such as noise and flickering).
103