60
Emanuele Danovaro, Leila De Floriani, Enrico Puppo, and Hanan Samet
Table 3.2. Comparison among multiresolution approaches
approach
data
update size in RAM
El-Sana and Chiang [12]
clustering dep.
free form
atomic
clusters of C.M.
DeCoro and Pajarola [7]
clustering dep.
free form
atomic
binary forest
Danovaro et al. [6]
clustering dep.
nD
atomic
a few clusters
Hoppe [20]
partitioning
scalar field atomic
cluster hierarchy
Cignoni et al. [4]
partitioning
scalar field large
binary tree
Yoon et al. [45]
partitioning
free form
large
cluster hierarchy
Lindstrom [30]
partitioning
free form
medium
a few MB
Shaffer and Garland [43]
partitioning
free form
medium
cluster indices
Cignoni et al. [5]
partitioning
free form
large
cluster hierarchy
the cluster hierarchy is kept in the main memory, while the sequences of edge collapses are fetched from the disk. Also, vertices and triangles corresponding to the
active clusters are stored in GPU memory. This approach has been developed for 3D
meshes, but can be easily adopted to TINs by using a 2D grid built on the projection
of the data points in the plane.
3.5.3 Comparison
Table 3.2 summarizes various multiresolution techniques that we have presented, by
highlighting the approach used to organize the out-of-core data structure, the data
that can be represented, and the size of updates. Overall, models based on the clustering of nodes in the hierarchy are not only more general, but also more complex
to manage. They have the advantage that the meshes extracted from them have the
same granularity and thus the same accuracy of the meshes extracted from the corresponding in-core multiresolution models. On the other hand, dealing with large sets
of atomic updates can become a bottleneck in some visualization tasks.
On the contrary, methods that use large patches highly simplify the management
of secondary memory and result more efficient, but they trade-off this advantage by
being coarser-grained, hence less smooth in the transition between different levels
of detail. In particular, the method by Yoon et al. [45] seems to be more suitable for
large scenes than terrains, and also the dependencies among clusters are not easily
managed.
3.6 Conclusions
We have analyzed and compared out-of-core approaches for simplification of triangle meshes and for out-of-core multiresolution modeling of TINs, both for regularly
and irregularly distributed data sets. Most of the mesh simplification algorithms and
the out-of-core multiresolution representations have been developed for visualization purposes. Most of the techniques for irregular meshes have been developed for
Emanuele Danovaro, Leila De Floriani, Enrico Puppo, and Hanan Samet
Table 3.2. Comparison among multiresolution approaches
approach
data
update size in RAM
El-Sana and Chiang [12]
clustering dep.
free form
atomic
clusters of C.M.
DeCoro and Pajarola [7]
clustering dep.
free form
atomic
binary forest
Danovaro et al. [6]
clustering dep.
nD
atomic
a few clusters
Hoppe [20]
partitioning
scalar field atomic
cluster hierarchy
Cignoni et al. [4]
partitioning
scalar field large
binary tree
Yoon et al. [45]
partitioning
free form
large
cluster hierarchy
Lindstrom [30]
partitioning
free form
medium
a few MB
Shaffer and Garland [43]
partitioning
free form
medium
cluster indices
Cignoni et al. [5]
partitioning
free form
large
cluster hierarchy
the cluster hierarchy is kept in the main memory, while the sequences of edge collapses are fetched from the disk. Also, vertices and triangles corresponding to the
active clusters are stored in GPU memory. This approach has been developed for 3D
meshes, but can be easily adopted to TINs by using a 2D grid built on the projection
of the data points in the plane.
3.5.3 Comparison
Table 3.2 summarizes various multiresolution techniques that we have presented, by
highlighting the approach used to organize the out-of-core data structure, the data
that can be represented, and the size of updates. Overall, models based on the clustering of nodes in the hierarchy are not only more general, but also more complex
to manage. They have the advantage that the meshes extracted from them have the
same granularity and thus the same accuracy of the meshes extracted from the corresponding in-core multiresolution models. On the other hand, dealing with large sets
of atomic updates can become a bottleneck in some visualization tasks.
On the contrary, methods that use large patches highly simplify the management
of secondary memory and result more efficient, but they trade-off this advantage by
being coarser-grained, hence less smooth in the transition between different levels
of detail. In particular, the method by Yoon et al. [45] seems to be more suitable for
large scenes than terrains, and also the dependencies among clusters are not easily
managed.
3.6 Conclusions
We have analyzed and compared out-of-core approaches for simplification of triangle meshes and for out-of-core multiresolution modeling of TINs, both for regularly
and irregularly distributed data sets. Most of the mesh simplification algorithms and
the out-of-core multiresolution representations have been developed for visualization purposes. Most of the techniques for irregular meshes have been developed for