It seems that there are ways to reduce triangles in a way to maintain shape systematically. Is there a way to give a metric of 'shape maintenance' such that rather than running this algorithm until there are 300 triangles, left, running until an additional triangle collapses would lead to too much of a loss in quality? Maybe by determining how much volume error there is?, in that there is a skull where there should be space and vice versa?
brandonlouie
The recognizability of shape reminded me of the graphical jump between Super Mario 64 and it's remake for the Nintendo DS. Here you can see a comparison between the two: https://i.ytimg.com/vi/N9cdRNd7BNA/maxresdefault.jpg
I was reminded of this because this slide reminded me that while it would be nice to have amazing looking meshes, we are ultimately limited by the power of the machines that can render our meshes and we can only strive to match a general shape. The image I linked shows a difference in hardware of about 8 years, and to me it's really interesting to see how each games' graphics are a product of their times. The model for Mario in SM64 is lower-poly due to the limited hardware power of the Nintendo 64, but looks just good enough to be recognized as Mario. The model for Mario in SM64DS, on the other hand, is noticeably smoother around the model is looks even more like the titular character
MillerHollinger
I'm wondering if there's a way to measure how much volume is lost each time the mesh is further simplified. It seems that the biggest losses in model quality occur when a part of the model goes missing, i.e. the volume rapidly decreases. If so, that could be a way to contextualize the trade-off between model quality and complexity.
aravmisra
If anyone is interested in this topic and the intersection in real-time mesh simplication + GPU utilization, I'd recommend this princeton publication (from a while back, so not sure how much has changed): https://gfx.cs.princeton.edu/pubs/DeCoro_2007_RMS/real_time_simplification.pdf
It seems that there are ways to reduce triangles in a way to maintain shape systematically. Is there a way to give a metric of 'shape maintenance' such that rather than running this algorithm until there are 300 triangles, left, running until an additional triangle collapses would lead to too much of a loss in quality? Maybe by determining how much volume error there is?, in that there is a skull where there should be space and vice versa?
The recognizability of shape reminded me of the graphical jump between Super Mario 64 and it's remake for the Nintendo DS. Here you can see a comparison between the two: https://i.ytimg.com/vi/N9cdRNd7BNA/maxresdefault.jpg
I was reminded of this because this slide reminded me that while it would be nice to have amazing looking meshes, we are ultimately limited by the power of the machines that can render our meshes and we can only strive to match a general shape. The image I linked shows a difference in hardware of about 8 years, and to me it's really interesting to see how each games' graphics are a product of their times. The model for Mario in SM64 is lower-poly due to the limited hardware power of the Nintendo 64, but looks just good enough to be recognized as Mario. The model for Mario in SM64DS, on the other hand, is noticeably smoother around the model is looks even more like the titular character
I'm wondering if there's a way to measure how much volume is lost each time the mesh is further simplified. It seems that the biggest losses in model quality occur when a part of the model goes missing, i.e. the volume rapidly decreases. If so, that could be a way to contextualize the trade-off between model quality and complexity.
If anyone is interested in this topic and the intersection in real-time mesh simplication + GPU utilization, I'd recommend this princeton publication (from a while back, so not sure how much has changed): https://gfx.cs.princeton.edu/pubs/DeCoro_2007_RMS/real_time_simplification.pdf