I am a bit confused on how this differs from what we have been doing so far in this class. Don't we normally have an object description (maybe some math or specifications for a mesh) that we then turn into a more efficient representation (like mesh itself with a BVH or something) that we then do computation in? How is this model different?
Staffjamesfong1
@AlexSchedel Good question. I agree that a lot of what we have done in class fits into this model. This slide serves to emphasize that there is a similar theme in physics simulation.
There is one important distinction compared to BVH though. The BVH lets us accelerate a slow algorithm without affecting the final output. e.g. you get pixel-for-pixel identical outputs with or without BVH. In the case of model analysis, you are introducing a rich trade-off between speed, accuracy, and stability. e.g. if you only keep a handful of modes, your simulation will be very fast and stable but it will be less accurate.
I am a bit confused on how this differs from what we have been doing so far in this class. Don't we normally have an object description (maybe some math or specifications for a mesh) that we then turn into a more efficient representation (like mesh itself with a BVH or something) that we then do computation in? How is this model different?
@AlexSchedel Good question. I agree that a lot of what we have done in class fits into this model. This slide serves to emphasize that there is a similar theme in physics simulation.
There is one important distinction compared to BVH though. The BVH lets us accelerate a slow algorithm without affecting the final output. e.g. you get pixel-for-pixel identical outputs with or without BVH. In the case of model analysis, you are introducing a rich trade-off between speed, accuracy, and stability. e.g. if you only keep a handful of modes, your simulation will be very fast and stable but it will be less accurate.