Commentary 8
Due: 3/19/97
"Particle Animation and Rendering Using Data Parallel Computation" by Karl Sims
This paper is about a techique to model the movement of systems at a high level of specification. That is the user is able to give very high level representations of what type of movement is desired.
The system was implemented on a Connection Machine CM-2. The idea was to create a simulation that was more realistic looking that motion achieved by moving along spline curves. This is accomplished by making a simulation based on reality. But the foucus is not placed on modeling physics of reality but, on how good it looks or how realistic it looks. The motion of the particles is controled by Euler's method. "There are four basic movement operations; those that set the postion; those that set the veloctiy ; those that alter the postion and those that alter the velocity" The artilce goes through a series of listings of the various types of motions that might be created namely damping, sprial, bouncing.
This technique is good for animators that are not interents in perciesly how the animation moves but that it gets from point a to point b. I am wondering what where they talking about with the virtual processers and particles?
"Flow and changes in Appearance" by Julie Dorsey, Hans Kohling Pedersen and Pat Hanrahan
This paper is talking about how to use a partical based system to create wethering pattern on objects. The system is based on capturing the appearance caused by the flow of water over a surface or gemotry. The current techniques are lacking in capturing the gemometry of an object and are very time consuming. The point to be noted is that the foucus is on caturing the patterns that are left by the flow not the flow of the water itself.
The current system to do this fall into two categories: techniques for simulating fluid flow and particle system. There are some term that are introduced that are the basics of how water flows over a surface. These are the quantitiy of water, the height, inclination and geometry of surface; and absorption of water by material(s) making up the surface. The descussion of staining defines how it is that the distibution of dirt on a surface can change the appearnce of the stain. By modifiing this distribution various stains are created. There are a series of fators that end up creating the effect of the flow of water these are water particle, absorption, deposition and environment For each fatctor that determines the stainthere is model that is used to represent how it is simulated in the paper. The water model created particular charateristics to control how the water is moving. The absorption model is broken into two functions one based on the rate the water is absorbed by surface and capacity of the surface to hold water. The deposition model describes how the dirt gets picked up off the surface and the rate at which it is deposited on the surface. The Environment model is used to simulate the effects of other factors on the stain other than the flow itself.
In this paper the authors present a method for modeling changes in suface appearance due to the process of water running over a surface. They present as illustrations the effects of weathering on the sides of buildings and the effects of weathering on statues. The flow is modeled as a particle system which is goverened by properties such as gravity, surface texture, wind, etc... The interaction of the water with the surface is goverened by a set of coupled differential equations which describe the absorbtion rate of the surface and the sedimentation of particles onto a surface.
The focus of the paper is on modeling the changes in appearance due to the flow of water and not on the actual appearance of the water flow. The main contribution is stated as being the coupling of particle systems with processes similar to those used to model water color painting.
The flow is modeled as a collection of particles, where each particle corresponds to a water droplet. The surfaces that the water flows over are modeled using parametric patches. Each particle representing a drop has internal properties among which are mass, position, velocity and solubility. The effects of this model over a patch of surfaces is strongly determined by the properties of the patches. For example, if the surface is very porus, then we would expect that the water would be absorbed into the surface rather than flowing in a continuous smooth manner. Absorbtion is modeled based on a simple set of parametes which describe the rate that the surface absorbes water, the total amount of water the surface can hold and lastly, the ratio of water absorbed to capacity of the surface.
In addition to modeling the flow and the surfaces, environmental factors are modeled as well. Environmental factors include such things as sun and rain.
The method presented appeared relatively straight forward. I found this to be an interesting use of particle systems having only seen them used to represent things such as the flow of the water itself.
In this paper, Sims presents a method of using particle systems to represent "dynamic phenomena" such as wind, water, snow and fire. His method makes use of a Connection Machine CM-2 that utilizes a virtual processor mechanism to simulate more processors than the physical number present. Each particle in the system is assigned a processor, which maintains the particles state information, so the number of (virtual) processors used is application dependent.
Sims points out that the goal of the presented system is to achieve a realisitic "looking" system. That is, it is more important that the results look correct than that the results behave, in a physical sense, correct. This allows a certain amount of leeway to be taken. For example, the bouncing of a particle off of a surface: In the method proposed, a particle penetrates the surface for one iteration, where in the reality, this does not happen. Sims does mention that a more robust method for computing collision detection would benefit the system.
When the particles are rendered, they are fragmented into additional data types each of which is also assigned a virtual processor. The amount and method by which a particle is fragmented is contrained so that the number of fragments does not exceed the maximum number of virtual processors available.
Future work discussed includes issues involving particle interaction (i.e., collision avoidance). At present, particles ignore each other and only act according to global rules. Implementing particle interaction behavior could lead to more realistic looking simulations. Additionally, as mentioned above, more robust collision detection algorithms could be employed to improve realism.
It allows to render (in parallel) particles that have different attributes, providing anti-aliasing, hidden surfaces and motion blur. It basically works by assigning one virtual processor to each primitive data element. The data parallel supercomputer pr ocess the data in parallel by executing the same program on different data. It is a sort of adaptive machine that can give an efficient use of its resources (processors-memory) It uses Kinematic and physically based systems This article also describes a language for particle animation, which is used to provide higher level operations and to combine effects to produce new higher level effects.
It uses Kinematic and physically based systems . The motion results achieved using this systems should look more realistic than the animation achieved by moving objects along curves or keyframes, mainly because it is a physically based simulation. But it is important to mention that it is difficult to obtain a desired motion by specifying only forces and acceleration. Using this animation system it is possible to obtain some basic operations using kinematic and physic based techniques.
In general particle animation is done by controlling particle state variables, they mention some of them, like age, mass, spiral-axis, etc. These states are controlled and updated by operators that are applied during the simulation, they can be processed in subgroups or individually. A particle animation-rendering process is discussed in the paper. The system also allows to mix the particles with other data types like polygons.
Some results - animations were explained in which the system seemed to give some good results. I think that particle systems for animation goes beyond the traditional objects made of surfaces, which is a nice addition to animation. These systems can simul ate only some sort of phenomena, (those that produce a sort of dynamic and fuzzy effect). Particle evolution is determined by applying certain rules to them. Non-procedural motion is mainly used, very complex particle behavior can be computed. It can foll ow for example deterministic or sthocastic motion. But I also think that it can be too expensive computationally.
The model uses a particle system, a drop of water is represented by each particle, and the motion of them is controlled by physical parameters and other constraints. The also consider the absorption of water by the surface and sedimentation deposits.
They describe the main variables that influences the flow: quantities of incident water, height, inclination and geometry of the surface and the absorption of water by the material. They modeled and take into account different phenomena that are characte ristic of the patterns that are generated by flow of water (differential flow, splashback, rainfall, dirt,...).
Their flow model is based on surface geometry, materials for the structures and loose deposits, and the environment. Water is modeled as a collection of particles, they have attributes like mass, position, velocity, etc. It can simulate absorption of wate r, transport of deposits (by dissolving and carrying surface material).
It is in general a very robust approach, which considers a lot of variables,models such as the deposition, absorption, it also models the environments (sunlight, rain, ...) but at the same time it keeps its simplicity at a understandable level. Although I think that most of the models are a very simple simplification, the results seemed to be realistic.
This approach is very helpful, and more efficient and realistic than painting textures to achieve the same effect. It basically demonstrated that the effect can be captured with a simple model.
The results seemed good (intuitivelly), although sometimes it is hard to say what the result should look like very accurately. A lot of effects can be modeled with this approach, at least in a semi-automatic way, which is practical, but it also need to ha ve a good 3D definition of the model.
This paper describes a method for modeling changes in the appearance of a surface (say, the front of a cathedral) caused by the flow of water over it. The main idea behind this work is to provide an effective and realistic model of appearance changes caused by the weathering of an object exposed to the forces of nature (more specifically, water flow). The best way of approaching the problem of modeling fluids is the use of particle systems and their interactions. The use of particle systems to model fluids is obviously easier and gives more realistic results than any other method. However, particle systems need the modeling of large numbers of particles and the interactions between and this is clearly computationally a highly expensive task which would not be justifiable for modeling rigid body mechanics. But they are very useful for realistic models of fluid bodies (such as rain, fire, etc).
The authors discuss alternative ways of showing weathering which involve painting a large number of different textures to surfaces and composing them within a programmable shader such as was done for Toy Story. However, this process is very time consuming and relies a lot on the intuition and intelligence of the programmer and a method that incorporates a certain degree of automation would be useful. It has been the authors' goal to propose such a mechanism in this paper. The authors have attempted to model the changes in appearance of a buildaing (say) due to the runoff of dirt caused by water flow over the surface of the building. They make the assumption that before the flow of water takes place, dirt is deposited uniformly over the surface. This may not be an entirely reasonable assumption when modeling the real world, but seems to achieve the desired results. Besides, it is unclear how one can effectively model the texture dirt, considering that the term dirt itself is a very subjective and vague term. Certain other phenomena such as the washing off of concrete on a building due to water flow and the change in color due to waterlogging are also addressed in the paper, as also the splashback of dirt around the bottom of the structure.
I found this paper very interesting. The simulation results shown are very impressive. Though the model proposed by the authors is conceptually very simple and easy to understand, I find it hard to believe their claim that "it requires only modest computation". Particle systems, by their very nature would necessitate large amounts of computation of interactions between particles.
This paper mainly addresses the computational issue that I have mentioned as being the key problem with particle systems. The author presents an implementation of particle systems on a parallel machine: the Connection Machine. By their very nature, particle systems are very amenable to parallel implementation and the Connection Machine seems to be an ideal platform for the implementation of such a system.
The thing that stuck me most in this paper was the simplicity of the computation proposed in it. The equations are very intuitive and easy to understand and I found it amazing that so much could be achieved from such simple equations. In fact, the idea that one could model such complex phenomena as fire and waterfalls from such a basic starting point is mind boggling.
Stan Sclaroff
Created: March 12, 1997
Last Modified: March 12, 1997