mingzailao
/
mitsuba
mirror of https://github.com/Quartenia/mitsuba.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

added left-balanced kd-tree construction as another testcase

metadata
Wenzel Jakob 2011-02-09 02:13:52 +01:00
parent 9c1c16ab87
commit 0516aacfe2
2 changed files with 39 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
include/mitsuba/core/kdtree.h
View File

@ -97,6 +97,9 @@ public:
enum EHeuristic {
/// Create a balanced tree by splitting along the median
EBalanced = 0,
/// Create a left-balanced tree
ELeftBalanced,
/**
* \brief Use the sliding midpoint tree construction rule. This
@ -365,10 +368,39 @@ protected:
case EBalanced: {
/* Split along the median */
split = rangeStart + (rangeEnd-rangeStart)/2;
axis = m_aabb.getLargestAxis();
std::nth_element(rangeStart, split, rangeEnd, CoordinateOrdering(axis));
};
break;
case ELeftBalanced: {
size_t treeSize = rangeEnd-rangeStart;
/* Layer 0 contains one node */
size_t p = 1;
/* Traverse downwards until the first incompletely
filled tree level is encountered */
while (2*p <= treeSize)
p *= 2;
/* Calculate the number of filled slots in the last level */
size_t remaining = treeSize - p + 1;
if (2*remaining < p) {
/* Case 2: The last level contains too few nodes. Remove
overestimate from the left subtree node count and add
the remaining nodes */
p = (p >> 1) + remaining;
}
axis = m_aabb.getLargestAxis();
split = rangeStart + (p - 1);
std::nth_element(rangeStart, split, rangeEnd,
CoordinateOrdering(axis));
};
break;
case ESlidingMidpoint: {
/* Sliding midpoint rule: find a split that is close to the spatial median */
axis = m_aabb.getLargestAxis();

12
src/tests/test_kd.cpp
View File

@ -130,10 +130,10 @@ public:
void test03_pointKDTree() {
typedef TKDTree< BasicKDNode<Point2, Float> > KDTree2;
size_t nPoints = 20000;
size_t nPoints = 50000, nTries = 20;
ref<Random> random = new Random();
for (int heuristic=0; heuristic<3; ++heuristic) {
for (int heuristic=0; heuristic<4; ++heuristic) {
KDTree2 kdtree(nPoints, (KDTree2::EHeuristic) heuristic);
for (size_t i=0; i<nPoints; ++i) {
@ -146,8 +146,10 @@ public:
if (heuristic == 0) {
Log(EInfo, "Testing the balanced kd-tree construction heuristic");
} else if (heuristic == 1) {
Log(EInfo, "Testing the sliding midpoint kd-tree construction heuristic");
Log(EInfo, "Testing the left-balanced kd-tree construction heuristic");
} else if (heuristic == 2) {
Log(EInfo, "Testing the sliding midpoint kd-tree construction heuristic");
} else if (heuristic == 3) {
Log(EInfo, "Testing the voxel volume kd-tree construction heuristic");
}
@ -156,7 +158,7 @@ public:
Log(EInfo, "Construction time = %i ms, depth = %i", timer->getMilliseconds(), kdtree.getDepth());
for (int k=1; k<=10; ++k) {
size_t nTraversals = 0, nTries = 10;
size_t nTraversals = 0;
for (size_t it = 0; it < nTries; ++it) {
Point2 p(random->nextFloat(), random->nextFloat());
nTraversals += kdtree.nnSearch(p, k, results);
@ -170,7 +172,7 @@ public:
}
Log(EInfo, "Average number of traversals for a %i-nn query = " SIZE_T_FMT, k, nTraversals / nTries);
}
size_t nTraversals = 0, nTries = 10;
size_t nTraversals = 0;
for (size_t it = 0; it < nTries; ++it) {
Point2 p(random->nextFloat(), random->nextFloat());
nTraversals += kdtree.search(p, 0.05, results);