Many disciplines must handle the creation, visualization, and manipulation of huge and complex 3D environments. Examples include large structural and mechanical engineering projects dealing with entire cars, ships, buildings, and processing plants. The complexity of such models is usually far beyond the interactive rendering capabilities of todays 3D graphics hardware. Previous approaches relied on costly preprocessing for reducing the number of polygons that need to be rendered per frame but suffered from excessive precomputation times --- often several days or even weeks.

with shadows and refractions), a Conference room (5.5 fps, without shadows), reflective and refractive Spheres-RT in an office (4.5 fps), and UT2003 a scene from a current computer game (7.5 fps, precomputed illumination). Recursive ray tracing is a simple yet powerful and general approach for accurately computing global light transport and rendering high quality images. While recent algorithmic improvements and optimized parallel software implementations have increased ray tracing performance to realtime levels, no compact and programmable hardware solution has been available yet. This paper describes the architecture and a prototype implementation of a single chip, fully programmable Ray Processing Unit (RPU). It combines the flexibility of general purpose CPUs with the efficiency of current GPUs for data parallel computations. This design allows for realtime ray tracing of dynamic scenes with programmable material, geometry, and illumination shaders. Although, running at only 66 MHz the prototype FPGA implementation already renders images at up to 20 frames per second, which in many cases beats the performance of highly optimized software running on multi-GHz desktop CPUs. The performance and efficiency of the proposed architecture is analyzed using a variety of benchmark scenes.

We present a signal-processing framework for light transport. We study the frequency content of radiance and how it is altered by phenomena such as shading, occlusion, and transport. This extends previous work that considered either spatial or angular dimensions, and it offers a comprehensive treatment of both space and angle. We show that occlusion, a multiplication in the primal, amounts in the Fourier domain to a convolution by the spectrum of the blocker. Propagation corresponds to a shear in the space-angle frequency domain, while reflection on curved objects performs a different shear along the angular frequency axis. As shown by previous work, reflection is a convolution in the primal and therefore a multiplication in the Fourier domain. Our work shows how the spatial components of lighting are affected by this angular convolution. Our framework predicts the characteristics of interactions such as caustics and the disappearance of the shadows of small features. Predictions on the frequency content can then be used to control sampling rates for rendering. Other potential applications include precomputed radiance transfer and inverse rendering.

Figure 1: Dress simulation: Four different images of a 210 step sequence taken from a dynamic cloth simulation and consisting of 40K triangles. By updating in real-time instead of rebuilding the BVH of the deforming model according to our heuristic, we are able to render the animation at 13 frames per second with 512 2 screen resolution using a dual-core P4 processor at 2.8 GHz. We present an efficient approach for interactive ray tracing of deformable or animated models. Unlike many of the recent approaches for ray tracing static scenes, we use bounding volume hierarchies (BVHs) instead of kd-trees as the underlying acceleration structure. Our algorithm makes no assumptions about the simulation or the motion of objects in the scene and dynamically updates or recomputes the BVHs. We also describe a method to detect BVH quality degradation during the simulation in order to determine when the hierarchy needs to be rebuilt. Furthermore, we show that the ray coherence techniques introduced for kd-trees can be naturally extended to BVHs and yield similar improvements. Finally, we compare BVHs to spatial kd-trees, which have been used recently as a replacement for AABB hierarchies. Our algorithm has been applied to different scenarios arising in animation and simulation and consisting of tens of thousands to a million triangles. In practice, our system can ray trace these models at 3-13 frames a second on a desktop PC including secondary rays.

Lighting models used in the production of computer generated feature animation have to be flexible, easy to control, and efficient to compute. Global illumination techniques do not lend themselves easily to flexibility, ease of use, or speed, and have remained out of reach thus far for the vast majority of images generated in this context. This paper describes the implementation and integration of indirect illumination within a feature animation production renderer. For efficiency reasons, we choose to partially solve the rendering equation. We explain how this compromise allows us to speed-up final gathering calculations and reduce noise. We describe an efficient ray tracing strategy and its integration with a micro-polygon based scan line renderer supporting displacement mapping and programmable shaders. We combine a modified irradiance gradient caching technique with an approximate lighting model that enhances caching coherence and provides good scalability to render complex scenes into highresolution images suitable for film. We describe the tools that are made available to the artists to control indirect lighting in final renders. We show that our approach provides an efficient solution, easy to art direct, that allows animators to enhance considerably

We present a new algorithm to render caustics. The algorithm discretizes the specular surfaces into sample points. Each of the sample points is treated as a pinhole camera that projects an image of the incoming light onto the diffuse receiver surfaces. Anti-aliasing is performed by considering the local surface curvature at the sample points to filter the projected images. The algorithm can be implemented using programmable texture mapping hardware. It allows to render caustics in fully dynamic scenes in real-time on current PC hardware.

When rendering only directly visible objects, ray tracing a few levels of specular reflection from large, lowcurvature surfaces, and ray tracing shadows from point-like light sources, the accessed geometry is coherent and a geometry cache performs well. But in many other cases, the accessed geometry is incoherent and a standard geometry cache performs poorly: ray tracing of specular reflection from highly curved surfaces, tracing rays that are many reflection levels deep, and distribution ray tracing for wide glossy reflection, global illumination, wide soft shadows, and ambient occlusion. Fortunately, less geometric accuracy is necessary in the incoherent cases. This observation can be formalized by looking at the ray differentials for different types of scattering: coherent rays have small differentials, while incoherent rays have large differentials. We utilize this observation to obtain efficient multiresolution caching of geometry and textures (including displacement maps) for classic and distribution ray tracing in complex scenes. We use an existing multiresolution caching scheme (originally developed for scanline rendering) for textures and displacement maps, and introduce a multiresolution geometry caching scheme for tessellated surfaces. The multiresolution geometry caching scheme makes it possible to efficiently render scenes that, if fully tessellated, would use 100 times more memory than the geometry cache size. 1.

The photon map method allows efficient computation of global illumination in general scenes. Individual photon hits, generated using Monte Carlo particle tracing, are stored in the maps and form a geometry independent representation of the illumination. Two important issues with the photon map are memory requirements to store the photons and the question how many photons are needed for an accurate representation of illumination in a certain scene. In this paper we introduce a method to control the density of photon maps by storing photons selectively based on a local required density criterion. This reduces memory usage significantly since in unimportant or over-dense regions less photons are stored. Results for caustic photon maps and global photon maps representing full illumination show a decrease in number of photons of a factor of 2 to 5. The required density states how accurate the photon map should be at a certain location and determines how many photons are needed in total. We also derive such a criterion based on a novel path-importance-based first pass, taking some steps towards solving the difficult 'how many photons' question.

In global illumination computations the photon map is a powerful tool for approximating the irradiance, which is stored independent from scene geometry. By presenting a new algorithm, which uses novel importance sampling techniques, we improve the memory footprint of the photon map, simplify the caustic generation, and allow for a much faster sampling of direct illumination in complicated models as they arise in a production environment.

Photo-realistic rendering algorithms such as Monte Carlo ray tracing sample individual paths to compute images. Noise and aliasing artefacts are usually reduced by supersampling. Knowledge about the neighborhood of the path, such as an estimated footprint, can be used to reduce these artefacts without having to trace additional paths. The recently introduced ray differentials estimate such a footprint for classical ray tracing, by computing ray derivatives with respect to the image plane. The footprint proves to be useful for filtering textures locally on surfaces. In this paper, we generalize the use of these derivatives to arbitrary path sampling, including general reflection and refraction functions.