Spherical Q2-tree for Sampling Dynamic Environment Sequences

Liang Wan, Tien-Tsin Wong and Chi-Sing Leung,
in Proc. of Eurographics Symposium on Rendering 2005 (EGSR 2005),

Konstanz, Germany, June 2005, pp. 21-30.

　

Abstract

Previous methods in environment map sampling seldom consider a sequence of dynamic environment maps. The generated sampling patterns of the sequence may not maintain the temporal illumination consistency and result in choppy animation. In this paper, we propose a novel approach, spherical q2-tree, to address this consistency problem. The local adaptive nature of the proposed method suppresses the abrupt change in the generated sampling patterns over time, hence ensures a smooth and consistent illumination. By partitioning the spherical surface with simple curvilinear equations, we construct a quadrilateral-based quadtree over the sphere. This q2-tree allows us to adaptively sample the environment based on an importance metric and generates low-discrepancy sampling patterns. No time-consuming relaxation is required. The sampling patterns of a dynamic sequence are rapidly generated by making use of the summed area table and exploiting the coherence of consecutive frames. From our experiments, the rendering quality of our sampling pattern for a static environment map is comparable to previous methods. However, our method produces smooth and consistent animation for a sequence of dynamic environment maps, even the number of samples is kept constant over time.

Download

  • Paper: q2tree.pdf (13.2 MB) (Revised on 13 July 2005 with figure correction)
  • Companion video: Divx (20 MB)

Visual Comparison

The table below compares static rendering results of several sampling methods.

 

Teapot

Zoom in

PSNR

Control

 

Q2-tree

33.9dB

Structured

33.9dB

Penrose

33.3dB

LightGen

31.9dB

 

Dynamic Results

Here shows some rendering results extracted from the animation sequence. The lighting video is a synthetic data, which is also available to download [HDR panoramic fire sequence].

 

Scene

Frame 89

Frame 90

Frame 91

 

copyright © Liang Wan / HOME