I recently analysed the possibility of streamed Level-of-Detail data inside ParaView. Because Paraview builds upon VTK, I used the Computer Graphics Elements Library to build up the repective data structures.
VTK itself provides 2 ways that were promising: on the one hand, it is possible to store a tree-structure of data (such as an Octree) as vtkMultiBlockData. Another way is to use vtkLODActors to – as it suggests: do Level-of-Detail.
Unfortunately, both ways have their drawbacks. For the MultiBlockData, while they are stored in a tree, ParaView typically, as all other VTK-pipelines, loads all the data at once in the memory. This is because of the VTK Pipeline, that always propagates full data containers. This somehow despises the effect I wanted to reach – just loading the buckets that are in view. So, in the end, you gain nothing with respect to „save memory consumption“ by using the structure. The Level-of-Detail unfortunately also demands that all detail levels of a particular model are permanently available in memory (traditional LoD-concept). This obviously also doesn’t work with Streaming.
I’m working on some new concepts to overcome the problem. A video about the test is appended to the post.
