Glossary of Terms
We assembled this small FAQ for all users that would like an introduction to various terms frequently used in rendering.
Q. What are the shading normals?
A. Due to the fact that the majority of render engines use polygonal meshes for scene description, rendering such a model on the image needs to have a high number of faces (dense mesh) in order to have the geometry and shading displayed smooth. Keeping the shading smooth can be accomplished though with a workaround, without having to tessellate finely the mesh; since normal vectors play a key role on the shading, we can assign and use pseudo-normal vectors on mesh faces that are used for shading only.
Q. I have seen the term "physically-plausible", what does this mean?
A. When we say that an engine is physically-plausible, we mean that the geometry, materials and emitters are based on models that have certain physical properties. The most significant property is energy conservation, i.e. a material can not reflect more light than it receives. Other physical constraints refer to using finite quantities and reciprocity in light scattering.
Q. I see some engines described as "physically-based", does it differ from being physically-plausible?
A. When an engine is physically-based, it is certainly physically-plausible, i.e. it does not violate physical laws. But even more, the models used are built on top of a physical theory. The latter makes these models even more accurate than physically-plausible ones that usually depend on observation.
Q. Is Thea Render a physically-based engine?
A. Thea Render uses state-of-the-art materials based on a very solid theory and that makes it a very accurate physically-based engine. The material primitives and layer system itself are based on very recent research, making Thea one of the most accurate renderers. At the same time, Thea Render uses certain features (shading normals and point lights) that are there for easiness of scene description whenever user would like to do so - these features should be thought as extensions/enhancements of the strictly physically-based engine.
Q. What does the term "bias" mean?
A. The global illumination problem is governed by an integral equation (rendering equation) that is very difficult to solve due to its high dimensionality. Classical deterministic arithmetical techniques are difficult to apply and usually provide a coarse solution to a subproblem (for example, using finite elements for the radiosity problem). Monte carlo techniques, on the other hand, are perfect for this kind of problems; they are capable to solve for all dimensions and "converge" with time. When we speak about the solution for deterministic techniques, we talk also about the error or residue, in the Monte Carlo (stochastic sampling) framework, the error is called "bias"; that is the difference of the algorithmic limit to the actual solution.
Q. So, an unbiased render engine will render with zero error and all unbiased engines will have the same solution?
A. When talking about an unbiased render engine, we refer to the unbiasedness of the algorithm used. Practically, there can never be a truly unbiased implementation since there are always arithmetical errors imposed by the limitation of floating point processor units. A typical example is the sun-pool caustics lighting transfer, where some unbiased engines fail to resolve due to arithmetical errors introduced (when a high number of samples needs to be taken). In this respect, two or more unbiased engines - even with the same input and models used - may have different output; although theoretically they will arrive to the same solution, practically the different implementations will solve more accurately different lighting transfers.
Q. I see the images being very noisy and refined over time, is this a sign of being unbiased?
A. No, being a progressive rendering technique where noise is reduced over time, does not mean that the engine is unbiased. Some lighting transfer may be left out, and final result may look completely incorrect.
Q. What is "path tracing"?
A. Path tracing is the very first unbiased technique proposed at the end of 80's, relatively simple to implement, that is still in use by rendering systems when simplicity is wanted. It performs nicely on direct lighting transfer but converges rather slow (with respect to a more sophisticated technique) when indirect lighting is involved.
Q. But I see unbiased renders being so realistic, does an unbiased engine give the most photorealistic results?
A. Not necessarily! An unbiased engine refers to the way of solving a certain problem, but not whether the problem is correctly posed and contains all the "photorealistic elements" in the correct way. An unbiased engine can solve very accurately a problem with where simplistic materials and saturated colors are used in a scene with low geometric complexity; the result will be far from photorealistic. Of particular importance to the photorealism, is that the material models used can deliver all the important visual cues and there is high enough complexity in the scene description; then, on second level, is that we can solve the problem with low error (this is where unbiased techniques can help).
Q. What is the "Warm Up" phase?
A. Warm up is a phrase used to describe a phase where initial samples are taken in order to improve stochastic sampling. These samples are not used in the estimation itself (image) but as a guide to direct next samples in a more effective way.
Q. What is Photon Mapping?
A. Photon mapping is a two-pass technique solving the global illumination problem. First, particles are emitted from light sources and reflected, transmitted or absorbed on surfaces. Their positions are recorded and are used as a coarse solution of the irradiance in the scene. At second step, a normal ray tracing pass uses the stored particles to calculate the irradiance. The accuracy of photon mapping solution can be increased by increasing the number of emitted particles but it is limited by the memory resources available for recording them. The solution is usually enhanced by a technique called final gathering.
Q. What about Photon Map Caustics?
A. Photon mapping presents also a solution for computing caustics on diffuse surfaces - some very accurate renders have been generated with this technique. Because caustics are high frequency patterns, they usually need quite a lot of particles to be emitted and stored. To decrease the memory demands, photon map caustics are recorded in different - more dense - structure. In Thea Render, photon map caustics have their own settings, independent of the normal (global) photon map which is used for the smooth low-frequency irradiance solution.
Q. What is Final Gathering?
A. Final gathering is a technique that is usually used in conjunction with photon mapping, in order to provide higher quality irradiance solution. It involves one extra bounce (or more) in the ray tracing pass, before querying the photon map and computing irradiance - this extra bounce smooths the discontinuity artifacts seen in the photon map. Final gathering is usually coupled with irradiance cache, a mechanism that amortizes the effort taken for the extra bounces in the ray tracing pass by interpolating the results between the computation positions.
Q. Can we use Final Gathering alone without Photon Mapping? When should we prefer this?
A. Yes, final gathering can be used all alone without being necessary to enable photon mapping. In this case, we usually need to increase the bounces in final gathering (more than one) so that we can reach easier the light sources. This approach should be preferred when light sources are (much) further away from the scene than the viewer. In this case, photon mapping is less effective when shooting particles in the scene, since fewer of them will be recorded and the photon map will be less accurate. Typical scenes, where using final gathering alone should be preferred, are interiors with sun+sky illumination.