Jinkai Hu, Milo K. Yip, Guillermo Elias Alonso, Shihao Gu, Xiangjun Tang and Xiaogang Jin
Images rendered with global illumination off/on using our SDFDDGI method for the Sponza scene. It costs only 1.3 milliseconds extra time for computing the indirect global illumination, comparing to the ground truth result rendered by an offline path tracer which costs more than 2 minutes. Our method is able to reconstruct multi-bounce diffuse global illumination and is noise-free.
We present SDFDDGI, a
novel approach to compute dynamic diffuse global illumination in real time
using Signed Distance Fields (SDF). For an input scene, we first construct
its compact representation using SDF. Different from traditional SDF which
are stored by discrete voxels, our approach approximates the scene by a set
of simple primitive shapes, which facilitates real-time computation and
dynamic changes. Then, we reconstruct the irradiance function in the space
domain by discrete samples (referred to as probes), which are positioned
heuristically for real time performance. The probe irradiance can be updated
and interpolated effectively supported by our compact SDF representation.
Subsequently, a screen-space refinement method is developed to enhance
rendering details and visual quality. We validate our approach by comparing
the performance and quality of our method to other state-of-the-art
real-time global illumination methods. Our approach is able to calculate
real-time diffuse global illumination for both dynamic geometry and dynamic
lighting efficiently without any precomputation, while also supporting
multi-bounced light. It is also hardware free, and can manage both large
open scenes and indoor high-detailed scenes.