OpenVDB 10.0.1
Loading...
Searching...
No Matches
Filter.h
Go to the documentation of this file.
1// Copyright Contributors to the OpenVDB Project
2// SPDX-License-Identifier: MPL-2.0
3//
4/// @author Ken Museth
5///
6/// @file tools/Filter.h
7///
8/// @brief Filtering of VDB volumes. All operations can optionally be masked
9/// with another grid that acts as an alpha-mask. By default, filtering
10/// operations do not modify the topology of the input tree and thus do
11/// not process active tiles. However Filter::setProcessTiles can be
12/// used to process active tiles, densifying them on demand when necessary.
13
14#ifndef OPENVDB_TOOLS_FILTER_HAS_BEEN_INCLUDED
15#define OPENVDB_TOOLS_FILTER_HAS_BEEN_INCLUDED
16
17#include <openvdb/Types.h>
18#include <openvdb/Grid.h>
19#include <openvdb/math/Math.h>
25#include <openvdb/util/Util.h>
26#include <openvdb/thread/Threading.h>
27#include "Interpolation.h"
28
29#include <tbb/parallel_for.h>
30#include <tbb/concurrent_vector.h>
31
32#include <algorithm> // for std::max()
33#include <functional>
34#include <type_traits>
35
36namespace openvdb {
38namespace OPENVDB_VERSION_NAME {
39namespace tools {
40
41/// @brief Volume filtering (e.g., diffusion) with optional alpha masking
42template<typename GridT,
43 typename MaskT = typename GridT::template ValueConverter<float>::Type,
44 typename InterruptT = util::NullInterrupter>
45class Filter
46{
47public:
48 using GridType = GridT;
49 using MaskType = MaskT;
50 using TreeType = typename GridType::TreeType;
51 using LeafType = typename TreeType::LeafNodeType;
52 using ValueType = typename GridType::ValueType;
53 using AlphaType = typename MaskType::ValueType;
55 using RangeType = typename LeafManagerType::LeafRange;
56 using BufferType = typename LeafManagerType::BufferType;
57 static_assert(std::is_floating_point<AlphaType>::value,
58 "openvdb::tools::Filter requires a mask grid with floating-point values");
59
60 /// Constructor
61 /// @param grid Grid to be filtered.
62 /// @param interrupt Optional interrupter.
63 Filter(GridT& grid, InterruptT* interrupt = nullptr)
64 : mGrid(&grid)
65 , mTask(nullptr)
66 , mInterrupter(interrupt)
67 , mMask(nullptr)
68 , mGrainSize(1)
69 , mMinMask(0)
70 , mMaxMask(1)
71 , mInvertMask(false)
72 , mTiles(false) {}
73
74 /// @brief Shallow copy constructor called by tbb::parallel_for()
75 /// threads during filtering.
76 /// @param other The other Filter from which to copy.
77 Filter(const Filter& other)
78 : mGrid(other.mGrid)
79 , mTask(other.mTask)
80 , mInterrupter(other.mInterrupter)
81 , mMask(other.mMask)
82 , mGrainSize(other.mGrainSize)
83 , mMinMask(other.mMinMask)
84 , mMaxMask(other.mMaxMask)
85 , mInvertMask(other.mInvertMask)
86 , mTiles(other.mTiles) {}
87
88 /// @return the grain-size used for multi-threading
89 int getGrainSize() const { return mGrainSize; }
90 /// @brief Set the grain-size used for multi-threading.
91 /// @note A grain size of 0 or less disables multi-threading!
92 void setGrainSize(int grainsize) { mGrainSize = grainsize; }
93
94 /// @return whether active tiles are being processed
95 bool getProcessTiles() const { return mTiles; }
96 /// @brief Set whether active tiles should also be processed.
97 /// @note If true, some tiles may become voxelized
98 /// @warning If using with a mask, ensure that the mask topology matches the
99 /// tile topology of the filter grid as tiles will not respect overlapping
100 /// mask values at tree levels finer than themselves e.g. a leaf level tile
101 /// will only use the corresponding tile ijk value in the mask grid
102 void setProcessTiles(bool flag) { mTiles = flag; }
103
104 /// @brief Return the minimum value of the mask to be used for the
105 /// derivation of a smooth alpha value.
106 AlphaType minMask() const { return mMinMask; }
107 /// @brief Return the maximum value of the mask to be used for the
108 /// derivation of a smooth alpha value.
109 AlphaType maxMask() const { return mMaxMask; }
110 /// @brief Define the range for the (optional) scalar mask.
111 /// @param min Minimum value of the range.
112 /// @param max Maximum value of the range.
113 /// @details Mask values outside the range are clamped to zero or one, and
114 /// values inside the range map smoothly to 0->1 (unless the mask is inverted).
115 /// @throw ValueError if @a min is not smaller than @a max.
117 {
118 if (!(min < max)) OPENVDB_THROW(ValueError, "Invalid mask range (expects min < max)");
119 mMinMask = min;
120 mMaxMask = max;
121 }
122
123 /// @brief Return true if the mask is inverted, i.e. min->max in the
124 /// original mask maps to 1->0 in the inverted alpha mask.
125 bool isMaskInverted() const { return mInvertMask; }
126 /// @brief Invert the optional mask, i.e. min->max in the original
127 /// mask maps to 1->0 in the inverted alpha mask.
128 void invertMask(bool invert=true) { mInvertMask = invert; }
129
130 /// @brief One iteration of a fast separable mean-value (i.e. box) filter.
131 /// @param width The width of the mean-value filter is 2*width+1 voxels.
132 /// @param iterations Number of times the mean-value filter is applied.
133 /// @param mask Optional alpha mask.
134 void mean(int width = 1, int iterations = 1, const MaskType* mask = nullptr);
135
136 /// @brief One iteration of a fast separable Gaussian filter.
137 ///
138 /// @note This is approximated as 4 iterations of a separable mean filter
139 /// which typically leads an approximation that's better than 95%!
140 /// @param width The width of the mean-value filter is 2*width+1 voxels.
141 /// @param iterations Number of times the mean-value filter is applied.
142 /// @param mask Optional alpha mask.
143 void gaussian(int width = 1, int iterations = 1, const MaskType* mask = nullptr);
144
145 /// @brief One iteration of a median-value filter
146 ///
147 /// @note This filter is not separable and is hence relatively slow!
148 /// @param width The width of the mean-value filter is 2*width+1 voxels.
149 /// @param iterations Number of times the mean-value filter is applied.
150 /// @param mask Optional alpha mask.
151 void median(int width = 1, int iterations = 1, const MaskType* mask = nullptr);
152
153 /// Offsets (i.e. adds) a constant value to all active voxels.
154 /// @param offset Offset in the same units as the grid.
155 /// @param mask Optional alpha mask.
156 void offset(ValueType offset, const MaskType* mask = nullptr);
157
158 /// @brief Used internally by tbb::parallel_for()
159 /// @param range Range of LeafNodes over which to multi-thread.
160 ///
161 /// @warning Never call this method directly!
162 void operator()(const RangeType& range) const
163 {
164 if (mTask) mTask(const_cast<Filter*>(this), range);
165 else OPENVDB_THROW(ValueError, "task is undefined - call median(), mean(), etc.");
166 }
167
168private:
169 using LeafT = typename TreeType::LeafNodeType;
170 using VoxelIterT = typename LeafT::ValueOnIter;
171 using VoxelCIterT = typename LeafT::ValueOnCIter;
172 using BufferT = typename tree::LeafManager<TreeType>::BufferType;
173 using LeafIterT = typename RangeType::Iterator;
174 using AlphaMaskT = tools::AlphaMask<GridT, MaskT>;
175
176 void cook(LeafManagerType& leafs);
177
178 template<size_t Axis>
179 struct Avg {
180 Avg(const GridT* grid, Int32 w): acc(grid->tree()), width(w), frac(1.f/float(2*w+1)) {}
181 inline ValueType operator()(Coord xyz);
182 typename GridT::ConstAccessor acc;
183 const Int32 width;
184 const float frac;
185 };
186
187 // Private filter methods called by tbb::parallel_for threads
188 template <typename AvgT>
189 void doBox(const RangeType& r, Int32 w);
190 void doBoxX(const RangeType& r, Int32 w) { this->doBox<Avg<0> >(r,w); }
191 void doBoxY(const RangeType& r, Int32 w) { this->doBox<Avg<1> >(r,w); }
192 void doBoxZ(const RangeType& r, Int32 w) { this->doBox<Avg<2> >(r,w); }
193 void doMedian(const RangeType&, int);
194 void doOffset(const RangeType&, ValueType);
195 /// @return true if the process was interrupted
196 bool wasInterrupted();
197
198 GridType* mGrid;
199 typename std::function<void (Filter*, const RangeType&)> mTask;
200 InterruptT* mInterrupter;
201 const MaskType* mMask;
202 int mGrainSize;
203 AlphaType mMinMask, mMaxMask;
204 bool mInvertMask;
205 bool mTiles;
206}; // end of Filter class
207
208
209////////////////////////////////////////
210
211/// @cond OPENVDB_DOCS_INTERNAL
212
213namespace filter_internal {
214
215template<typename TreeT>
216struct Voxelizer
217{
218 // NodeManager for processing internal/root node values
219 // @note Should not cache leaf nodes
220 using NodeManagerT = tree::NodeManager<TreeT, TreeT::RootNodeType::LEVEL-1>;
221 using MaskT = typename TreeT::template ValueConverter<ValueMask>::Type;
222
223 Voxelizer(TreeT& tree, const bool allNeighbors, const size_t grainSize)
224 : mVoxelTopology()
225 , mManager(nullptr)
226 , mGrainSize(grainSize)
227 , mOp(tree, mVoxelTopology, allNeighbors ? 26 : 6) {}
228
229 /// @brief Convert tiles to leaf nodes that exist at a particular
230 /// voxel distance away
231 /// @param width distance in voxels to seach for tiles from each leaf
232 /// @return Returns how many search iterations were performed, which
233 /// also represents how many leaf node neighbors may have been
234 /// created. Returns 0 if the tree is already entirely voxelized
235 int run(const int width)
236 {
237 if (!mOp.tree().hasActiveTiles()) return 0;
238 this->init();
239 int count = 0;
240 for (int i = 0; i < width; i += int(TreeT::LeafNodeType::DIM), ++count) {
241 if (i > 0) mManager->rebuild();
242 mManager->foreachBottomUp(mOp, mGrainSize > 0, mGrainSize);
243 mOp.tree().topologyUnion(mVoxelTopology);
244 }
245 return count;
246 }
247
248private:
249 void init()
250 {
251 if (mManager) {
252 mManager->rebuild();
253 }
254 else {
255 // @note We don't actually need the leaf topology here, just the
256 // internal node structure so that we can generate leaf nodes in parallel
257 mVoxelTopology.topologyUnion(mOp.tree());
258 mManager.reset(new NodeManagerT(mOp.tree()));
259 }
260 }
261
262 struct CreateVoxelMask
263 {
264 using LeafT = typename TreeT::LeafNodeType;
265 using RootT = typename TreeT::RootNodeType;
266
267 CreateVoxelMask(TreeT& tree, MaskT& mask, const size_t NN)
268 : mTree(tree), mVoxelTopology(mask), mNeighbors(NN) {}
269
270 TreeT& tree() { return mTree; }
271
272 // do nothing for leaf nodes. They shouldn't even be cached as
273 // part of the NodeManager used with this method.
274 void operator()(const LeafT&) const { assert(false); }
275
276 void operator()(const RootT& node) const
277 {
278 using ChildT = typename RootT::ChildNodeType;
279 static constexpr Int32 CHILDDIM = Int32(ChildT::DIM);
280 static constexpr Int32 LEAFDIM = Int32(LeafT::DIM);
281 const Tester op(mTree, mNeighbors);
282
283 auto step =
284 [&](const Coord& ijk,
285 const size_t axis1,
286 const size_t axis2,
287 const auto& val)
288 {
289 Coord offset(0);
290 Int32& a = offset[axis1];
291 Int32& b = offset[axis2];
292 for (a = 0; a < CHILDDIM; a+=LEAFDIM) {
293 for (b = 0; b < CHILDDIM; b+=LEAFDIM) {
294 const Coord childijk = ijk + offset;
295 if (op.test(childijk, val)) {
296 mVoxelTopology.touchLeaf(childijk);
297 }
298 }
299 }
300
301 offset.reset(CHILDDIM-1);
302 for (a = 0; a < CHILDDIM; a+=LEAFDIM) {
303 for (b = 0; b < CHILDDIM; b+=LEAFDIM) {
304 const Coord childijk = ijk + offset;
305 if (op.test(childijk, val)) {
306 mVoxelTopology.touchLeaf(childijk);
307 }
308 }
309 }
310 };
311
312 for (auto iter = node.cbeginValueOn(); iter; ++iter) {
313 const Coord& ijk = iter.getCoord();
314 // @todo step only needs to search if a given direction
315 // depending on the face
316 step(ijk, 0, 1, *iter);
317 step(ijk, 0, 2, *iter);
318 step(ijk, 1, 2, *iter);
319 }
320 }
321
322 template<typename NodeT>
323 void operator()(const NodeT& node) const
324 {
325 using ChildT = typename NodeT::ChildNodeType;
326 static constexpr Int32 CHILDDIM = Int32(ChildT::DIM);
327 static constexpr Int32 LEAFDIM = Int32(LeafT::DIM);
328
329 static auto step =
330 [](const Tester& op,
331 const Coord& ijk,
332 const size_t axis1,
333 const size_t axis2,
334 const auto& val,
335 std::vector<Coord>& coords)
336 {
337 Coord offset(0);
338 Int32& a = offset[axis1];
339 Int32& b = offset[axis2];
340 for (a = 0; a < CHILDDIM; a+=LEAFDIM) {
341 for (b = 0; b < CHILDDIM; b+=LEAFDIM) {
342 const Coord childijk = ijk + offset;
343 if (op.test(childijk, val)) {
344 coords.emplace_back(childijk);
345 }
346 }
347 }
348
349 offset.reset(CHILDDIM-1);
350 for (a = 0; a < CHILDDIM; a+=LEAFDIM) {
351 for (b = 0; b < CHILDDIM; b+=LEAFDIM) {
352 const Coord childijk = ijk + offset;
353 if (op.test(childijk, val)) {
354 coords.emplace_back(childijk);
355 }
356 }
357 }
358 };
359
360 /// Two types of algorithms here
361 /// 1) For the case where this node is the direct parent of leaf nodes
362 /// 2) For all other node types
363 ///
364 /// In general, given a tile's ijk, search its faces/edges/corners for
365 /// values which differ from its own or leaf level topology. When a
366 /// difference is detected, mask topology is generated which can be used
367 /// with topologyUnion to ensure valid voxel values exist in the source
368 /// grid.
369 ///
370 /// This operator handles all internal node types. For example, for the
371 /// lowest level internal node (which contains leaf nodes as children)
372 /// each tile is at the leaf level (a single tile represents an 8x8x8
373 /// node). CHILDDIM is this case will match the valid of LEAFDIM, as we
374 /// only need to check each tiles immediate neighbors. For higher level
375 /// internal nodes (and the root node) each child tile will have a
376 /// significantly larger CHILDDIM than the grid's LEAFDIM. We
377 /// consistently probe values along the LEAFDIM stride to ensure no
378 /// changes are missed.
379
380 if (CHILDDIM == LEAFDIM) {
381 // If the current node is the parent of leaf nodes, search each
382 // neighbor directly and use a flag buffer to test offsets in
383 // this node which need converting to leaf level topology.
384 // This is faster than the more general method which steps across
385 // faces (unecessary due to CHILDDIM == LEAFDIM) and provides
386 // a simpler way of tracking new topology
387
388 std::vector<char> flags(NodeT::NUM_VALUES, char(0));
389 tbb::parallel_for(tbb::blocked_range<size_t>(0, NodeT::NUM_VALUES),
390 [&](const tbb::blocked_range<size_t>& range) {
391 const Tester op(mTree, mNeighbors);
392 for (size_t n = range.begin(), N = range.end(); n < N; ++n) {
393 if (node.isValueMaskOn(Index(n))) {
394 // if index is a tile, search its neighbors
395 const Coord ijk = node.offsetToGlobalCoord(Index(n));
396 flags[n] = op.test(ijk, node.getValue(ijk));
397 }
398 }
399 });
400
401 // create leaf level topology in this internal node
402 Index idx = 0;
403 for (auto iter = flags.begin(); iter != flags.end(); ++iter, ++idx) {
404 if (*iter) mVoxelTopology.touchLeaf(node.offsetToGlobalCoord(idx));
405 }
406 }
407 else {
408 // If this is a higher level internal node, we only need to search its
409 // face/edge/vertex neighbors for values which differ or leaf level
410 // topology. When a difference is detected, store the coordinate which
411 // needs to be voxelized.
412 // @todo investigate better threaded impl
413
414 tbb::concurrent_vector<Coord> nodes;
415 tbb::parallel_for(tbb::blocked_range<size_t>(0, NodeT::NUM_VALUES),
416 [&](const tbb::blocked_range<size_t>& range)
417 {
418 const Tester op(mTree, mNeighbors);
419 std::vector<Coord> coords;
420
421 for (size_t n = range.begin(), N = range.end(); n < N; ++n) {
422 if (!node.isValueMaskOn(Index(n))) continue;
423
424 const Coord ijk = node.offsetToGlobalCoord(Index(n));
425 const auto& val = node.getValue(ijk);
426 // @todo step only needs to search if a given direction
427 // depending on the face
428 step(op, ijk, 0, 1, val, coords);
429 step(op, ijk, 0, 2, val, coords);
430 step(op, ijk, 1, 2, val, coords);
431 }
432
433 if (!coords.empty()) {
434 std::copy(coords.begin(), coords.end(),
435 nodes.grow_by(coords.size()));
436 }
437 });
438
439 // create leaf level topology in this internal node
440 // @note nodes may contain duplicate coords
441 for (const auto& coord : nodes) {
442 mVoxelTopology.touchLeaf(coord);
443 }
444 }
445 }
446
447 private:
448 struct Tester
449 {
450 Tester(const TreeT& tree, const size_t NN)
451 : mAcc(tree), mNeighbors(NN) {}
452
453 inline bool test(const Coord& ijk,
454 const typename TreeT::ValueType& val) const
455 {
456 static constexpr Int32 LEAFDIM = Int32(LeafT::DIM);
457 const Coord* NN = util::COORD_OFFSETS;
458 for (size_t i = 0; i < mNeighbors; ++i, ++NN) {
459 Coord neighbor(*NN);
460 neighbor.x() *= LEAFDIM;
461 neighbor.y() *= LEAFDIM;
462 neighbor.z() *= LEAFDIM;
463 neighbor += ijk;
464 // if a leaf exists, assume its buffer is not constant
465 if (mAcc.getValue(neighbor) != val ||
466 mAcc.probeConstLeaf(neighbor)) {
467 return true;
468 }
469 }
470 return false;
471 }
472 private:
473 const tree::ValueAccessor<const TreeT> mAcc;
474 const size_t mNeighbors;
475 };
476
477 private:
478 TreeT& mTree;
479 MaskT& mVoxelTopology;
480 const size_t mNeighbors;
481 };// CreateVoxelMask
482
483private:
484 MaskT mVoxelTopology;
485 std::unique_ptr<NodeManagerT> mManager;
486 const size_t mGrainSize;
487 CreateVoxelMask mOp;
488};
489
490// Helper function for Filter::Avg::operator()
491template<typename T> static inline void accum(T& sum, T addend) { sum += addend; }
492// Overload for bool ValueType
493inline void accum(bool& sum, bool addend) { sum = sum || addend; }
494
495} // namespace filter_internal
496
497/// @endcond
498
499////////////////////////////////////////
500
501
502template<typename GridT, typename MaskT, typename InterruptT>
503template<size_t Axis>
504inline typename GridT::ValueType
505Filter<GridT, MaskT, InterruptT>::Avg<Axis>::operator()(Coord xyz)
506{
507 ValueType sum = zeroVal<ValueType>();
508 Int32 &i = xyz[Axis], j = i + width;
509 for (i -= width; i <= j; ++i) filter_internal::accum(sum, acc.getValue(xyz));
511 ValueType value = static_cast<ValueType>(sum * frac);
513 return value;
514}
515
516
517////////////////////////////////////////
518
519
520template<typename GridT, typename MaskT, typename InterruptT>
521void
522Filter<GridT, MaskT, InterruptT>::mean(int width, int iterations, const MaskType* mask)
523{
524 if (iterations <= 0) return;
525 mMask = mask;
526 const int w = std::max(1, width);
527 const bool serial = mGrainSize == 0;
528
529 if (mInterrupter) mInterrupter->start("Applying mean filter");
530
531 std::unique_ptr<filter_internal::Voxelizer<TreeType>> voxelizer;
532 if (this->getProcessTiles()) {
533 // if performing multiple iterations, also search edge/vertex
534 // neighbors for difference topology.
535 const bool allNeighbors = iterations > 1;
536 // If processing tiles, create a voxelizer and run a single
537 // width based search for tiles that need to be voxelized
538 voxelizer.reset(new filter_internal::Voxelizer<TreeType>
539 (mGrid->tree(), allNeighbors, mGrainSize));
540 if (!voxelizer->run(w)) voxelizer.reset();
541 }
542
543 LeafManagerType leafs(mGrid->tree(), 1, serial);
544
545 int iter = 1; // num of leaf level neighbor based searches performed
546 int dist = w; // kernel distance of the current iteration
547 for (int i=0; i<iterations && !this->wasInterrupted(); ++i, dist+=w) {
548 if (i > 0 && voxelizer) {
549 // the total influence distance in voxels of this iteration
550 // minus how far we've already accounted for
551 const int remain = dist - iter * int(TreeType::LeafNodeType::DIM);
552 if (remain > 0) {
553 const int searches = voxelizer->run(remain);
554 if (searches == 0) voxelizer.reset();
555 else leafs.rebuild(serial);
556 iter += searches;
557 }
558 }
559
560 mTask = std::bind(&Filter::doBoxX, std::placeholders::_1, std::placeholders::_2, w);
561 this->cook(leafs);
562 // note that the order of the YZ passes are flipped to maintain backwards-compatibility
563 // with an indexing typo in the original logic
564 mTask = std::bind(&Filter::doBoxZ, std::placeholders::_1, std::placeholders::_2, w);
565 this->cook(leafs);
566 mTask = std::bind(&Filter::doBoxY, std::placeholders::_1, std::placeholders::_2, w);
567 this->cook(leafs);
568 }
569
570 if (mInterrupter) mInterrupter->end();
571}
572
573
574template<typename GridT, typename MaskT, typename InterruptT>
575void
576Filter<GridT, MaskT, InterruptT>::gaussian(int width, int iterations, const MaskType* mask)
577{
578 if (iterations <= 0) return;
579 mMask = mask;
580 const int w = std::max(1, width);
581 const bool serial = mGrainSize == 0;
582
583 if (mInterrupter) mInterrupter->start("Applying Gaussian filter");
584
585 std::unique_ptr<filter_internal::Voxelizer<TreeType>> voxelizer;
586 if (this->getProcessTiles()) {
587 // if performing multiple iterations, also search edge/vertex
588 // neighbors for difference topology.
589 const bool allNeighbors = iterations > 1;
590 // If processing tiles, create a voxelizer and run a single
591 // width based search for tiles that need to be voxelized
592 // @note account for sub iteration due to gaussian filter
593 voxelizer.reset(new filter_internal::Voxelizer<TreeType>
594 (mGrid->tree(), allNeighbors, mGrainSize));
595 if (!voxelizer->run(w*4)) voxelizer.reset();
596 }
597
598 LeafManagerType leafs(mGrid->tree(), 1, serial);
599
600 int iter = 1; // num of leaf level neighbor based searches performed
601 int dist = w*4; // kernel distance of the current iteration
602 for (int i=0; i<iterations; ++i, dist+=(w*4)) {
603 if (i > 0 && voxelizer) {
604 // the total influence distance in voxels of this iteration
605 // minus how far we've already accounted for
606 const int remain = dist - iter * int(TreeType::LeafNodeType::DIM);
607 if (remain > 0) {
608 const int searches = voxelizer->run(remain);
609 if (searches == 0) voxelizer.reset();
610 else leafs.rebuild(serial);
611 iter += searches;
612 }
613 }
614
615 for (int n=0; n<4 && !this->wasInterrupted(); ++n) {
616 mTask = std::bind(&Filter::doBoxX, std::placeholders::_1, std::placeholders::_2, w);
617 this->cook(leafs);
618 // note that the order of the YZ passes are flipped to maintain backwards-compatibility
619 // with an indexing typo in the original logic
620 mTask = std::bind(&Filter::doBoxZ, std::placeholders::_1, std::placeholders::_2, w);
621 this->cook(leafs);
622 mTask = std::bind(&Filter::doBoxY, std::placeholders::_1, std::placeholders::_2, w);
623 this->cook(leafs);
624 }
625 }
626
627 if (mInterrupter) mInterrupter->end();
628}
629
630
631template<typename GridT, typename MaskT, typename InterruptT>
632void
633Filter<GridT, MaskT, InterruptT>::median(int width, int iterations, const MaskType* mask)
634{
635 if (iterations <= 0) return;
636 mMask = mask;
637 const int w = std::max(1, width);
638 const bool serial = mGrainSize == 0;
639
640 if (mInterrupter) mInterrupter->start("Applying median filter");
641
642 std::unique_ptr<filter_internal::Voxelizer<TreeType>> voxelizer;
643 if (this->getProcessTiles()) {
644 // If processing tiles, create a voxelizer and run a single
645 // width based search for tiles that need to be voxelized
646 voxelizer.reset(new filter_internal::Voxelizer<TreeType>
647 (mGrid->tree(), /*allNeighbors*/true, mGrainSize));
648 if (!voxelizer->run(w)) voxelizer.reset();
649 }
650
651 LeafManagerType leafs(mGrid->tree(), 1, serial);
652
653 mTask = std::bind(&Filter::doMedian, std::placeholders::_1, std::placeholders::_2, w);
654
655 int iter = 1; // num of leaf level neighbor based searches performed
656 int dist = w; // kernel distance of the current iteration
657 for (int i=0; i<iterations && !this->wasInterrupted(); ++i, dist+=w) {
658 if (i > 0 && voxelizer) {
659 // the total influence distance in voxels of this iteration
660 // minus how far we've already accounted for
661 const int remain = dist - iter * int(TreeType::LeafNodeType::DIM);
662 if (remain > 0) {
663 const int searches = voxelizer->run(remain);
664 if (searches == 0) voxelizer.reset();
665 else leafs.rebuild(serial);
666 iter += searches;
667 }
668 }
669
670 this->cook(leafs);
671 }
672
673 if (mInterrupter) mInterrupter->end();
674}
675
676
677template<typename GridT, typename MaskT, typename InterruptT>
678void
680{
681 mMask = mask;
682
683 if (mInterrupter) mInterrupter->start("Applying offset");
684
685 if (this->getProcessTiles()) {
686 // Don't process leaf nodes with the node manager - we'll do them
687 // separately to allow for cleaner branching
688 using NodeManagerT = tree::NodeManager<TreeType, TreeType::RootNodeType::LEVEL-1>;
689 NodeManagerT manager(mGrid->tree());
690
691 if (mask) {
692 manager.foreachBottomUp([&](auto& node) {
693 this->wasInterrupted();
694 AlphaMaskT alpha(*mGrid, *mMask, mMinMask, mMaxMask, mInvertMask);
695 typename AlphaMaskT::FloatType a, b;
696 for (auto iter = node.beginValueOn(); iter; ++iter) {
697 if (!alpha(iter.getCoord(), a, b)) continue;
699 iter.modifyValue([&](ValueType& v) { v += a*value; });
701 }
702 });
703 }
704 else {
705 manager.foreachBottomUp([&](auto& node) {
706 this->wasInterrupted();
707 for (auto iter = node.beginValueOn(); iter; ++iter) {
708 iter.modifyValue([&](ValueType& v) { v += value; });
709 }
710 });
711 }
712 }
713
714 LeafManagerType leafs(mGrid->tree(), 0, mGrainSize==0);
715 mTask = std::bind(&Filter::doOffset, std::placeholders::_1, std::placeholders::_2, value);
716 this->cook(leafs);
717
718 if (mInterrupter) mInterrupter->end();
719}
720
721
722////////////////////////////////////////
723
724
725/// Private method to perform the task (serial or threaded) and
726/// subsequently swap the leaf buffers.
727template<typename GridT, typename MaskT, typename InterruptT>
728void
729Filter<GridT, MaskT, InterruptT>::cook(LeafManagerType& leafs)
730{
731 if (mGrainSize>0) {
732 tbb::parallel_for(leafs.leafRange(mGrainSize), *this);
733 } else {
734 (*this)(leafs.leafRange());
735 }
736 leafs.swapLeafBuffer(1, mGrainSize==0);
737}
738
739
740/// One dimensional convolution of a separable box filter
741template<typename GridT, typename MaskT, typename InterruptT>
742template <typename AvgT>
743void
744Filter<GridT, MaskT, InterruptT>::doBox(const RangeType& range, Int32 w)
745{
746 this->wasInterrupted();
747 AvgT avg(mGrid, w);
748 if (mMask) {
749 typename AlphaMaskT::FloatType a, b;
750 AlphaMaskT alpha(*mGrid, *mMask, mMinMask, mMaxMask, mInvertMask);
751 for (LeafIterT leafIter=range.begin(); leafIter; ++leafIter) {
752 BufferT& buffer = leafIter.buffer(1);
753 for (VoxelCIterT iter = leafIter->cbeginValueOn(); iter; ++iter) {
754 const Coord xyz = iter.getCoord();
755 if (alpha(xyz, a, b)) {
757 const ValueType value(b*(*iter) + a*avg(xyz));
759 buffer.setValue(iter.pos(), value);
760 }
761 }
762 }
763 } else {
764 for (LeafIterT leafIter=range.begin(); leafIter; ++leafIter) {
765 BufferT& buffer = leafIter.buffer(1);
766 for (VoxelCIterT iter = leafIter->cbeginValueOn(); iter; ++iter) {
767 buffer.setValue(iter.pos(), avg(iter.getCoord()));
768 }
769 }
770 }
771}
772
773
774/// Performs simple but slow median-value diffusion
775template<typename GridT, typename MaskT, typename InterruptT>
776void
777Filter<GridT, MaskT, InterruptT>::doMedian(const RangeType& range, int width)
778{
779 this->wasInterrupted();
780 typename math::DenseStencil<GridType> stencil(*mGrid, width);//creates local cache!
781 if (mMask) {
782 typename AlphaMaskT::FloatType a, b;
783 AlphaMaskT alpha(*mGrid, *mMask, mMinMask, mMaxMask, mInvertMask);
784 for (LeafIterT leafIter=range.begin(); leafIter; ++leafIter) {
785 BufferT& buffer = leafIter.buffer(1);
786 for (VoxelCIterT iter = leafIter->cbeginValueOn(); iter; ++iter) {
787 if (alpha(iter.getCoord(), a, b)) {
788 stencil.moveTo(iter);
790 ValueType value(b*(*iter) + a*stencil.median());
792 buffer.setValue(iter.pos(), value);
793 }
794 }
795 }
796 } else {
797 for (LeafIterT leafIter=range.begin(); leafIter; ++leafIter) {
798 BufferT& buffer = leafIter.buffer(1);
799 for (VoxelCIterT iter = leafIter->cbeginValueOn(); iter; ++iter) {
800 stencil.moveTo(iter);
801 buffer.setValue(iter.pos(), stencil.median());
802 }
803 }
804 }
805}
806
807
808/// Offsets the values by a constant
809template<typename GridT, typename MaskT, typename InterruptT>
810void
811Filter<GridT, MaskT, InterruptT>::doOffset(const RangeType& range, ValueType offset)
812{
813 this->wasInterrupted();
814 if (mMask) {
815 typename AlphaMaskT::FloatType a, b;
816 AlphaMaskT alpha(*mGrid, *mMask, mMinMask, mMaxMask, mInvertMask);
817 for (LeafIterT leafIter=range.begin(); leafIter; ++leafIter) {
818 for (VoxelIterT iter = leafIter->beginValueOn(); iter; ++iter) {
819 if (alpha(iter.getCoord(), a, b)) {
821 ValueType value(*iter + a*offset);
823 iter.setValue(value);
824 }
825 }
826 }
827 } else {
828 for (LeafIterT leafIter=range.begin(); leafIter; ++leafIter) {
829 for (VoxelIterT iter = leafIter->beginValueOn(); iter; ++iter) {
830 iter.setValue(*iter + offset);
831 }
832 }
833 }
834}
835
836
837template<typename GridT, typename MaskT, typename InterruptT>
838inline bool
839Filter<GridT, MaskT, InterruptT>::wasInterrupted()
840{
841 if (util::wasInterrupted(mInterrupter)) {
842 thread::cancelGroupExecution();
843 return true;
844 }
845 return false;
846}
847
848
849////////////////////////////////////////
850
851
852// Explicit Template Instantiation
853
854#ifdef OPENVDB_USE_EXPLICIT_INSTANTIATION
855
856#ifdef OPENVDB_INSTANTIATE_FILTER
858#endif
859
860OPENVDB_INSTANTIATE_CLASS Filter<FloatGrid, FloatGrid, util::NullInterrupter>;
861OPENVDB_INSTANTIATE_CLASS Filter<DoubleGrid, FloatGrid, util::NullInterrupter>;
862
863#endif // OPENVDB_USE_EXPLICIT_INSTANTIATION
864
865
866} // namespace tools
867} // namespace OPENVDB_VERSION_NAME
868} // namespace openvdb
869
870#endif // OPENVDB_TOOLS_FILTER_HAS_BEEN_INCLUDED
ValueT value
Definition: GridBuilder.h:1290
A LeafManager manages a linear array of pointers to a given tree's leaf nodes, as well as optional au...
General-purpose arithmetic and comparison routines, most of which accept arbitrary value types (or at...
#define OPENVDB_NO_TYPE_CONVERSION_WARNING_BEGIN
Bracket code with OPENVDB_NO_TYPE_CONVERSION_WARNING_BEGIN/_END, to inhibit warnings about type conve...
Definition: Platform.h:204
#define OPENVDB_NO_TYPE_CONVERSION_WARNING_END
Definition: Platform.h:205
Definition: Exceptions.h:65
Signed (x, y, z) 32-bit integer coordinates.
Definition: Coord.h:25
Definition: Interpolation.h:544
FloatT FloatType
Definition: Interpolation.h:551
Volume filtering (e.g., diffusion) with optional alpha masking.
Definition: Filter.h:46
bool getProcessTiles() const
Definition: Filter.h:95
typename MaskType::ValueType AlphaType
Definition: Filter.h:53
typename tree::LeafManager< TreeType > LeafManagerType
Definition: Filter.h:54
AlphaType minMask() const
Return the minimum value of the mask to be used for the derivation of a smooth alpha value.
Definition: Filter.h:106
void setProcessTiles(bool flag)
Set whether active tiles should also be processed.
Definition: Filter.h:102
void setGrainSize(int grainsize)
Set the grain-size used for multi-threading.
Definition: Filter.h:92
void invertMask(bool invert=true)
Invert the optional mask, i.e. min->max in the original mask maps to 1->0 in the inverted alpha mask.
Definition: Filter.h:128
int getGrainSize() const
Definition: Filter.h:89
typename GridType::ValueType ValueType
Definition: Filter.h:52
typename GridType::TreeType TreeType
Definition: Filter.h:50
void setMaskRange(AlphaType min, AlphaType max)
Define the range for the (optional) scalar mask.
Definition: Filter.h:116
AlphaType maxMask() const
Return the maximum value of the mask to be used for the derivation of a smooth alpha value.
Definition: Filter.h:109
Filter(const Filter &other)
Shallow copy constructor called by tbb::parallel_for() threads during filtering.
Definition: Filter.h:77
void operator()(const RangeType &range) const
Used internally by tbb::parallel_for()
Definition: Filter.h:162
typename LeafManagerType::LeafRange RangeType
Definition: Filter.h:55
typename LeafManagerType::BufferType BufferType
Definition: Filter.h:56
MaskT MaskType
Definition: Filter.h:49
typename TreeType::LeafNodeType LeafType
Definition: Filter.h:51
bool isMaskInverted() const
Return true if the mask is inverted, i.e. min->max in the original mask maps to 1->0 in the inverted ...
Definition: Filter.h:125
GridT GridType
Definition: Filter.h:48
Filter(GridT &grid, InterruptT *interrupt=nullptr)
Definition: Filter.h:63
This class manages a linear array of pointers to a given tree's leaf nodes, as well as optional auxil...
Definition: LeafManager.h:85
typename CopyConstness< TreeType, NonConstBufferType >::Type BufferType
Definition: LeafManager.h:95
To facilitate threading over the nodes of a tree, cache node pointers in linear arrays,...
Definition: NodeManager.h:531
GridType
List of types that are currently supported by NanoVDB.
Definition: NanoVDB.h:243
OPENVDB_AX_API void run(const char *ax, openvdb::GridBase &grid, const AttributeBindings &bindings={})
Run a full AX pipeline (parse, compile and execute) on a single OpenVDB Grid.
Axis
Definition: Math.h:901
bool wasInterrupted(T *i, int percent=-1)
Definition: NullInterrupter.h:49
Index32 Index
Definition: Types.h:54
int32_t Int32
Definition: Types.h:56
Definition: Exceptions.h:13
#define OPENVDB_THROW(exception, message)
Definition: Exceptions.h:74
Defines various finite difference stencils by means of the "curiously recurring template pattern" on ...
NodeManager produces linear arrays of all tree nodes allowing for efficient threading and bottom-up p...
#define OPENVDB_VERSION_NAME
The version namespace name for this library version.
Definition: version.h.in:121
#define OPENVDB_USE_VERSION_NAMESPACE
Definition: version.h.in:212
#define OPENVDB_INSTANTIATE_CLASS
Definition: version.h.in:153