FastV: From-point Visibility Culling on Complex Models

Comparison with Beam Tracing

In this section we compare PVS computed by FastV using 4Kx4K uniform frusta with a PVS computed by an efficient beam tracer [Overbeck et al. 2007]. We denote the PVS computed by FastV as PVS_4Kx4K and the PVS computed by the beam tracer as PVS_BT. We choose 10 key frames from the armadillo sequence in the supplement video (Download) and compare PVS_4Kx4K and PVS_BT for the camera position of these key frames (see Figure 5). The key frames are equally spaced starting at Frame 1 (the sequence has a total of 270 key frames). Figure 4 and Table 1 summarizes our results. Note that our PVS converges to within 10% of the exact from-point beam tracing solution.

Figure 4: Comparison of PVS sizes of our baseline PVS (PVS_4Kx4K) and a beam tracing [Overbeck et al. 2007] PVS (PVS_BT).

Table 1: Difference in the PVS sizes of our baseline PVS (PVS_4Kx4K) and a beam tracing [Overbeck et al. 2007] PVS (PVS_BT) as a percentage of PVS_BT.

Figure 5: Ten key frames used for comparing the from-point PVS computed by FastV and an efficient beam tracing implementation [Overbeck et al. 2007]. The PVS rendered in the images above was computed using the beam tracer [Overbeck et al. 2007].

Download