| 1 |
Photo-realistic vs. Physically-based Rendering |
| 2 |
|
| 3 |
Photo-realistic rendering places emphasis on the appearance of its |
| 4 |
output rather than the techniques used to derive it. Anything goes, |
| 5 |
basically, as long as the final image looks nice. There is no |
| 6 |
attempt to use physically realistic values for the light sources |
| 7 |
or the surface reflectances. In fact, the light sources themselves |
| 8 |
often have physically impossible characteristics like 1/r falloff (as |
| 9 |
opposed to 1/r^2) or there is a lot of ambient lighting that comes from |
| 10 |
nowhere but somehow manages to illuminate the room. (You are probably |
| 11 |
saying, "Hey! Doesn't Radiance use an ambient term?" The answer is |
| 12 |
yes, but only as a final approximation to the interreflected component. |
| 13 |
The renderers I'm talking about use the ambient level as a main source |
| 14 |
of illumination!) Also, surfaces typically have color but there is no |
| 15 |
reflectance given, so all the surfaces appear to have roughly the same |
| 16 |
brightness. |
| 17 |
|
| 18 |
Such numerical shortcuts are often just conveniences provided so the |
| 19 |
user can get results easily and quickly without having to worry about |
| 20 |
fussy details, like where to put the light sources and what to use |
| 21 |
for reflectances. As you might expect, there is a penalty paid besides |
| 22 |
meaningless values, and that is fake-looking images. Have you noticed |
| 23 |
how these renderings always look pastel and glowing? You're seeing |
| 24 |
the visual equivalent of AM radio. |
| 25 |
|
| 26 |
Physically-based rendering, on the other hand, follows the physical |
| 27 |
behavior of light as closely as possible in an effort to *predict* |
| 28 |
what the final appearance of a design will be. This is not an |
| 29 |
artist's conception anymore, it is a numerical simulation. The |
| 30 |
light sources start in the calculation by emitting with a |
| 31 |
specific distribution, and the simulation computes the reflections |
| 32 |
between surfaces until the solution converges. The most popular |
| 33 |
technique for this computation is usually referred to as "radiosity", |
| 34 |
or flux transfer, and it does this by dividing all the surfaces into |
| 35 |
patches that exchange light energy within a closed system. This type |
| 36 |
of calculation is limited for the most part to simple scenes with |
| 37 |
diffuse surfaces where the visibility calculation and the solution |
| 38 |
matrix are manageable. |
| 39 |
|
| 40 |
Radiance, in contrast to most flux transfer methods, uses ray tracing |
| 41 |
to follow light in the reverse direction and does not require the same |
| 42 |
discretization as radiosity techniques. This has significant |
| 43 |
advantages when the scene geometry is complex, and permits the modeling |
| 44 |
of some specular interactions between surfaces. In general, Radiance |
| 45 |
is faster than radiosity if the scene contains more than a few thousand |
| 46 |
surfaces or has significant specularity. |
