Lecture 8: Mesh Representations and Geometry Processing (9)
rheask8246
Are there any current algorithms that do a sort of dynamic regularization on topologies? For example, I was thinking with flatter surfaces, it'd be a bit computationally expensive to make meshes of similar sizes rather than enlarging them. Conversely, for intricate/complicated regions, could we quantify how small we want the mesh triangles to be to keep the quality of the image?
StaffMichaelRenMR
Not sure if this is exactly what you were thinking of, but you can take a look at LOD maps. This is a technique used to improve texture clarity without too much of a performance impact. Essentially the idea is that for objects farther away from the viewer, less detail is required so larger triangles and smaller mipmaps are fine. Meanwhile for objects closer to the viewer, we require more detail, so we use smaller triangles and higher resolution mipmaps.
jonathanlu31
Is mesh regularization similar to blurring/low pass filtering for 3D? Since the downsampled version of the cow seems to have some sharp edges not present in the high resolution version but the regularized version is smoother. Also, if the intention for downsampling is to reduce computational cost, doesn't regularization add more vertices and edges to keep track of, which increases memory usage again?
aliwang2001
Also had a question about mesh regularization -- how is it improving quality? From what I can tell, the right image looks a lot less detailed than the left. For example, the ears are less detailed, and you can't really make out the nose.
chetan-khanna
Are there any situations in which we can regularize the mesh but intentionally make the triangles of different sizes? (For example, have larger triangles for the parts of the mesh that are pretty flat, and then smaller triangles for more detailed parts—but unlike the non-regularized mesh, they are all equilateral/"Delauney-like". So we get the advantages of fairly "Delauney-like" triangles, but don't sacrifice too much detail.
Are there any current algorithms that do a sort of dynamic regularization on topologies? For example, I was thinking with flatter surfaces, it'd be a bit computationally expensive to make meshes of similar sizes rather than enlarging them. Conversely, for intricate/complicated regions, could we quantify how small we want the mesh triangles to be to keep the quality of the image?
Not sure if this is exactly what you were thinking of, but you can take a look at LOD maps. This is a technique used to improve texture clarity without too much of a performance impact. Essentially the idea is that for objects farther away from the viewer, less detail is required so larger triangles and smaller mipmaps are fine. Meanwhile for objects closer to the viewer, we require more detail, so we use smaller triangles and higher resolution mipmaps.
Is mesh regularization similar to blurring/low pass filtering for 3D? Since the downsampled version of the cow seems to have some sharp edges not present in the high resolution version but the regularized version is smoother. Also, if the intention for downsampling is to reduce computational cost, doesn't regularization add more vertices and edges to keep track of, which increases memory usage again?
Also had a question about mesh regularization -- how is it improving quality? From what I can tell, the right image looks a lot less detailed than the left. For example, the ears are less detailed, and you can't really make out the nose.
Are there any situations in which we can regularize the mesh but intentionally make the triangles of different sizes? (For example, have larger triangles for the parts of the mesh that are pretty flat, and then smaller triangles for more detailed parts—but unlike the non-regularized mesh, they are all equilateral/"Delauney-like". So we get the advantages of fairly "Delauney-like" triangles, but don't sacrifice too much detail.