jbikker wrote:Trying to implement 'Irregular grids'. Almost there.
Looking good! I might give this a try myself . Are you looking at: http://www.kalojanov.com/data/irregular_grid.pdf ?
ultimatemau wrote:Looking good! I might give this a try myself . Are you looking at: http://www.kalojanov.com/data/irregular_grid.pdf ?
ziu wrote:So this paper lacks essential references. Plus it has too many pointless BVH and kd-tree references. This part of the paper, the related work and the references, could use some improvement.
ziu wrote:Their grid construction algorithm could also be faster. Looking at their 'emit_new_refs', for example, it uses a per primitive loop to generate the prim and cell ids, which will have poor workload distribution in scenes with dissimilar sized primitives. We solve that in our paper:
https://www.academia.edu/13928983/Effic ... hitectures
javor wrote:The Nemoto & Omachi and Taranta et al. citations are certainly an oversight on our side, thanks for pointing this out. I was not aware of the first paper, and have no excuse for forgetting to cite the Macro/Micro regions papers. When writing the paper, we considered positioning the paper only as a grid-based ray tracing method, but decided against that. Personally, I think that pointing the reader towards the state-of-the-art ray tracing approaches on current hardware is more important than differentiating our method from algorithms tested so far back in time. If it was up to me, we would also leave out the comparison with macro-regions, which was forced on us by the primary reviewer. We are aware that a multitude of similar approaches exist and have been tried before. The important point is that the particular combination of techniques we implemented delivers surprisingly good performance from a type of acceleration structure, which, as you said, does not see a lot of attention.
javor wrote:The construction speed is indeed the part where we expect to see improvements in the future, however, I would look elsewhere for optimization opportunities. Overall the initial two-level grid construction is the fastest part of the build process, and most of the time is spent merging and expanding cells. Also, the first level grid we create is too sparse for the suggested technique to work, which is why Arsene did not include it in the implementation. Of course, we are happy to be proven wrong!
Users browsing this forum: No registered users and 3 guests