What's the best way to decide what shape to use as a bounding volume? In this case it seems the first rectangle might be the best.
CarneAsadaFry
@faywerliu There are definitely some design tradeoffs to consider here. The first rectangle is definitely the tightest. but as was mentioned in lecture, it is also harder to compute and has less simple normal vectors in comparison to the second bounding box, and in this case the volume is not too much different. If you had some object that looked more like a diagonal rectangle, then you could imagine that the second bounding box would be much, much larger than the first, and then maybe the first would be better. So, it seems like it could depend on the types of objects you have the scene.
Dezhang1999
@faywerliu I can imagine there might be some ML Algorithm to do it. It reminds me of object recognition where it will always shows a container that is just right around the object in the camera.
Zenuriken
In lecture, the professor mentions how the leftmost rectangle provides the most accurate/tightest of bounds for the teapot, but it's also the most computationally intensive. I was wondering what sort of algorithm could yield such a bound? The only method I could think of would be to repeatedly create a box perimeter around the teapot using tangent vectors until you find a combination that yields the minimum area while still maintaining the entirety of the teapot within it, but this method seems very inefficient.
gowenong
Bounding volumes is also very relevant in FPS Video Games such as CS:GO, where a player model is bounded by 3D boxes to determine if a player is hit during gunfight. Because of the frame precise calculation needed for fast-paced gameplay, the tradeoff between calculating hitbox position and accuracy in binding the object is even more high stakes. In 2015, CS:GO decided to abandon rectangular prism hitboxes in favor of cylindrical ones - more info can be found on this Wiki here
madisonakemi
This also makes me wonder how objects that are obscured from view in the render frame are considered in the ray tracing process? Does it still take the same amount of calculation and run time for those objects vs ones that are in full view? Or is there a way to optimize rendering to cut down the time in these cases?
What's the best way to decide what shape to use as a bounding volume? In this case it seems the first rectangle might be the best.
@faywerliu There are definitely some design tradeoffs to consider here. The first rectangle is definitely the tightest. but as was mentioned in lecture, it is also harder to compute and has less simple normal vectors in comparison to the second bounding box, and in this case the volume is not too much different. If you had some object that looked more like a diagonal rectangle, then you could imagine that the second bounding box would be much, much larger than the first, and then maybe the first would be better. So, it seems like it could depend on the types of objects you have the scene.
@faywerliu I can imagine there might be some ML Algorithm to do it. It reminds me of object recognition where it will always shows a container that is just right around the object in the camera.
In lecture, the professor mentions how the leftmost rectangle provides the most accurate/tightest of bounds for the teapot, but it's also the most computationally intensive. I was wondering what sort of algorithm could yield such a bound? The only method I could think of would be to repeatedly create a box perimeter around the teapot using tangent vectors until you find a combination that yields the minimum area while still maintaining the entirety of the teapot within it, but this method seems very inefficient.
Bounding volumes is also very relevant in FPS Video Games such as CS:GO, where a player model is bounded by 3D boxes to determine if a player is hit during gunfight. Because of the frame precise calculation needed for fast-paced gameplay, the tradeoff between calculating hitbox position and accuracy in binding the object is even more high stakes. In 2015, CS:GO decided to abandon rectangular prism hitboxes in favor of cylindrical ones - more info can be found on this Wiki here
This also makes me wonder how objects that are obscured from view in the render frame are considered in the ray tracing process? Does it still take the same amount of calculation and run time for those objects vs ones that are in full view? Or is there a way to optimize rendering to cut down the time in these cases?