Comments on: The Holidays: Time for fun work! http://www.rorydriscoll.com/2009/01/03/the-holidays-time-for-fun-work/ Sat, 28 Jan 2012 01:33:40 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: rory http://www.rorydriscoll.com/2009/01/03/the-holidays-time-for-fun-work/comment-page-1/#comment-44 rory Thu, 08 Jan 2009 06:41:38 +0000 http://www.rorydriscoll.com/?p=70#comment-44 Thanks for the book recommendation. I haven't really looked into performance improvements too much yet, but when I get into tracing against triangle meshes then I'm sure that will jump up in priority. It's interesting to hear about the performance increase you got by using an octree though. Good luck with your project! Thanks for the book recommendation. I haven’t really looked into performance improvements too much yet, but when I get into tracing against triangle meshes then I’m sure that will jump up in priority. It’s interesting to hear about the performance increase you got by using an octree though.

Good luck with your project!

]]> By: tor http://www.rorydriscoll.com/2009/01/03/the-holidays-time-for-fun-work/comment-page-1/#comment-43 tor Thu, 08 Jan 2009 00:22:19 +0000 http://www.rorydriscoll.com/?p=70#comment-43 Very interesting post. We have a class project where you have to implement a ray tracer, and like you, I too got really annoyed at having to wait a few minutes to get any results. The book we used for the course (and something I highly recommend if you are looking into this: http://www.amazon.com/gp/product/1568811985 Realistic Ray Tracing, Second Edition (Hardcover) by R. Keith Morley (Author), Peter Shirley (Author) ) Anyway, they talk about using bounding volume hierarchies to go through the scene, we ended up doing some polygon meshes for the ray tracing, and well, I had a 3k poly model that would take about 20 minutes to do a render on. I even had a dual core and was multithreading it. Finally I bit the bullet and implemented a loose oct-tree structure (I didn't really dig the whole binary/ BVH. I suppose oct-trees just make more sense for 3d space for me. Anyway, cut the render down to ~3 seconds or so. As opposed to brute forcing through all the objects. You only have 3 spheres and a plane, but testing all 4 surfaces for every ray adds up, especially if you could just do one ray shoot and determine if you actually hit anything in one if statement! (However, the cost per actual hit goes up, but yea...) Very interesting post. We have a class project where you have to implement a ray tracer, and like you, I too got really annoyed at having to wait a few minutes to get any results.

The book we used for the course (and something I highly recommend if you are looking into this: http://www.amazon.com/gp/product/1568811985 Realistic Ray Tracing, Second Edition (Hardcover)
by R. Keith Morley (Author), Peter Shirley (Author) )

Anyway, they talk about using bounding volume hierarchies to go through the scene, we ended up doing some polygon meshes for the ray tracing, and well, I had a 3k poly model that would take about 20 minutes to do a render on. I even had a dual core and was multithreading it.

Finally I bit the bullet and implemented a loose oct-tree structure (I didn’t really dig the whole binary/ BVH. I suppose oct-trees just make more sense for 3d space for me.

Anyway, cut the render down to ~3 seconds or so. As opposed to brute forcing through all the objects.

You only have 3 spheres and a plane, but testing all 4 surfaces for every ray adds up, especially if you could just do one ray shoot and determine if you actually hit anything in one if statement! (However, the cost per actual hit goes up, but yea…)

]]>