High-quality curve rendering using line sampled visibility

Abstract

Figure 1: Our novel thin curve rendering algorithm used on a test production model to compute accurate visibility. The model has 32,000 unique hair strands, which consists of over one million Bézier curves with varying thickness. As can be seen, our algorithm works at all different scales, from cases where there are hundreds of hair strands per pixel to zooming in on the hair strands. All images were rendered at 1024 × 1024 pixels with our GPU implementation. The leftmost image took 109 ms to render, while the close-up on the face took 468 ms. The rightmost image showcases our ability to handle thick curves. Hair model courtesy of Weta Digital. AbstractComputing accurate visibility for thin primitives, such as hair strands, fur, grass, at all scales remains difficult or expensive. To that end, we present an efficient visibility algorithm based on spatial line sampling, and a novel intersection algorithm between line sample planes and Bézier splines with varying thickness. Our algorithm produces accurate visibility both when the projected width of the curve is a tiny fraction of a pixel, and when the projected width is tens of pixels. In addition, we present a rapid resolve procedure that computes final visibility. Using an optimized implementation running on graphics processors, we can render tens of thousands long hair strands with noise-free visibility at near-interactive rates.

A 4

“…We also want to explore whether the sampling rate can be increased adaptively when a small triangle is missed inside a pixel. Furthermore, we also want to incorporate analytical [Catmull 1978;Auzinger et al 2013] and semi-analytical methods [Barringer et al 2012] into our renderer. It may, e.g., be possible to develop an analytical method that handles 2-4 triangles per pixel, and then revert to MSAA for more triangles.…”

Analysis of Sample Correlations for Monte Carlo Rendering

“…While several point sampling constructions and transformations are proposed to improve stratification, MC estimation need not always be a point sampling process. Over the past couple of decades researchers have started employing MC-like estimators using line (segment) samples for rendering problems ranging from anti-aliasing [JP00], distribution effects [GDAM10, TPD * 12], hair rendering [BGAM12] to path sampling [NNDJ12, KGH * 14] or density estimation [JNSJ11]. Billen and Dutré [BD16] demonstrated improvements with line samples for direct illumination.…”

Line segment sampling with blue-noise properties

“…In recent years, many efficient rendering algorithms have benefited from line segments for sampling both lighting and viewing rays [Sun et al 2010;Jarosz et al 2011a;Jarosz et al 2011b;Novák et al 2012b;Novák et al 2012a]. Researchers have also begun to use line samples for hair rendering because of their efficiency in the computationally intensive and highly detailed visibility test [Barringer et al 2012]. In these algorithms, the computational cost of a single line segment sample is usually much greater than that for a point sample.…”