123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591 |
- /*
- ---------------------------------------------------------------------------
- Open Asset Import Library (assimp)
- ---------------------------------------------------------------------------
- Copyright (c) 2006-2022, assimp team
- All rights reserved.
- Redistribution and use of this software in source and binary forms,
- with or without modification, are permitted provided that the following
- conditions are met:
- * Redistributions of source code must retain the above
- copyright notice, this list of conditions and the
- following disclaimer.
- * Redistributions in binary form must reproduce the above
- copyright notice, this list of conditions and the
- following disclaimer in the documentation and/or other
- materials provided with the distribution.
- * Neither the name of the assimp team, nor the names of its
- contributors may be used to endorse or promote products
- derived from this software without specific prior
- written permission of the assimp team.
- THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- ---------------------------------------------------------------------------
- */
- /** @file TriangulateProcess.cpp
- * @brief Implementation of the post processing step to split up
- * all faces with more than three indices into triangles.
- *
- *
- * The triangulation algorithm will handle concave or convex polygons.
- * Self-intersecting or non-planar polygons are not rejected, but
- * they're probably not triangulated correctly.
- *
- * DEBUG SWITCHES - do not enable any of them in release builds:
- *
- * AI_BUILD_TRIANGULATE_COLOR_FACE_WINDING
- * - generates vertex colors to represent the face winding order.
- * the first vertex of a polygon becomes red, the last blue.
- * AI_BUILD_TRIANGULATE_DEBUG_POLYS
- * - dump all polygons and their triangulation sequences to
- * a file
- */
- #ifndef ASSIMP_BUILD_NO_TRIANGULATE_PROCESS
- #include "PostProcessing/TriangulateProcess.h"
- #include "PostProcessing/ProcessHelper.h"
- #include "Common/PolyTools.h"
- #include <memory>
- #include <cstdint>
- //#define AI_BUILD_TRIANGULATE_COLOR_FACE_WINDING
- //#define AI_BUILD_TRIANGULATE_DEBUG_POLYS
- #define POLY_GRID_Y 40
- #define POLY_GRID_X 70
- #define POLY_GRID_XPAD 20
- #define POLY_OUTPUT_FILE "assimp_polygons_debug.txt"
- using namespace Assimp;
- namespace {
- /**
- * @brief Helper struct used to simplify NGON encoding functions.
- */
- struct NGONEncoder {
- NGONEncoder() : mLastNGONFirstIndex((unsigned int)-1) {}
- /**
- * @brief Encode the current triangle, and make sure it is recognized as a triangle.
- *
- * This method will rotate indices in tri if needed in order to avoid tri to be considered
- * part of the previous ngon. This method is to be used whenever you want to emit a real triangle,
- * and make sure it is seen as a triangle.
- *
- * @param tri Triangle to encode.
- */
- void ngonEncodeTriangle(aiFace * tri) {
- ai_assert(tri->mNumIndices == 3);
- // Rotate indices in new triangle to avoid ngon encoding false ngons
- // Otherwise, the new triangle would be considered part of the previous NGON.
- if (isConsideredSameAsLastNgon(tri)) {
- std::swap(tri->mIndices[0], tri->mIndices[2]);
- std::swap(tri->mIndices[1], tri->mIndices[2]);
- }
- mLastNGONFirstIndex = tri->mIndices[0];
- }
- /**
- * @brief Encode a quad (2 triangles) in ngon encoding, and make sure they are seen as a single ngon.
- *
- * @param tri1 First quad triangle
- * @param tri2 Second quad triangle
- *
- * @pre Triangles must be properly fanned from the most appropriate vertex.
- */
- void ngonEncodeQuad(aiFace *tri1, aiFace *tri2) {
- ai_assert(tri1->mNumIndices == 3);
- ai_assert(tri2->mNumIndices == 3);
- ai_assert(tri1->mIndices[0] == tri2->mIndices[0]);
- // If the selected fanning vertex is the same as the previously
- // emitted ngon, we use the opposite vertex which also happens to work
- // for tri-fanning a concave quad.
- // ref: https://github.com/assimp/assimp/pull/3695#issuecomment-805999760
- if (isConsideredSameAsLastNgon(tri1)) {
- // Right-rotate indices for tri1 (index 2 becomes the new fanning vertex)
- std::swap(tri1->mIndices[0], tri1->mIndices[2]);
- std::swap(tri1->mIndices[1], tri1->mIndices[2]);
- // Left-rotate indices for tri2 (index 2 becomes the new fanning vertex)
- std::swap(tri2->mIndices[1], tri2->mIndices[2]);
- std::swap(tri2->mIndices[0], tri2->mIndices[2]);
- ai_assert(tri1->mIndices[0] == tri2->mIndices[0]);
- }
- mLastNGONFirstIndex = tri1->mIndices[0];
- }
- /**
- * @brief Check whether this triangle would be considered part of the lastly emitted ngon or not.
- *
- * @param tri Current triangle.
- * @return true If used as is, this triangle will be part of last ngon.
- * @return false If used as is, this triangle is not considered part of the last ngon.
- */
- bool isConsideredSameAsLastNgon(const aiFace * tri) const {
- ai_assert(tri->mNumIndices == 3);
- return tri->mIndices[0] == mLastNGONFirstIndex;
- }
- private:
- unsigned int mLastNGONFirstIndex;
- };
- }
- // ------------------------------------------------------------------------------------------------
- // Returns whether the processing step is present in the given flag field.
- bool TriangulateProcess::IsActive( unsigned int pFlags) const {
- return (pFlags & aiProcess_Triangulate) != 0;
- }
- // ------------------------------------------------------------------------------------------------
- // Executes the post processing step on the given imported data.
- void TriangulateProcess::Execute( aiScene* pScene) {
- ASSIMP_LOG_DEBUG("TriangulateProcess begin");
- bool bHas = false;
- for( unsigned int a = 0; a < pScene->mNumMeshes; a++)
- {
- if (pScene->mMeshes[ a ]) {
- if ( TriangulateMesh( pScene->mMeshes[ a ] ) ) {
- bHas = true;
- }
- }
- }
- if ( bHas ) {
- ASSIMP_LOG_INFO( "TriangulateProcess finished. All polygons have been triangulated." );
- } else {
- ASSIMP_LOG_DEBUG( "TriangulateProcess finished. There was nothing to be done." );
- }
- }
- // ------------------------------------------------------------------------------------------------
- // Triangulates the given mesh.
- bool TriangulateProcess::TriangulateMesh( aiMesh* pMesh) {
- // Now we have aiMesh::mPrimitiveTypes, so this is only here for test cases
- if (!pMesh->mPrimitiveTypes) {
- bool bNeed = false;
- for( unsigned int a = 0; a < pMesh->mNumFaces; a++) {
- const aiFace& face = pMesh->mFaces[a];
- if( face.mNumIndices != 3) {
- bNeed = true;
- }
- }
- if (!bNeed)
- return false;
- }
- else if (!(pMesh->mPrimitiveTypes & aiPrimitiveType_POLYGON)) {
- return false;
- }
- // Find out how many output faces we'll get
- uint32_t numOut = 0, max_out = 0;
- bool get_normals = true;
- for( unsigned int a = 0; a < pMesh->mNumFaces; a++) {
- aiFace& face = pMesh->mFaces[a];
- if (face.mNumIndices <= 4) {
- get_normals = false;
- }
- if( face.mNumIndices <= 3) {
- numOut++;
- } else {
- numOut += face.mNumIndices-2;
- max_out = std::max(max_out,face.mNumIndices);
- }
- }
- // Just another check whether aiMesh::mPrimitiveTypes is correct
- ai_assert(numOut != pMesh->mNumFaces);
- aiVector3D *nor_out = nullptr;
- // if we don't have normals yet, but expect them to be a cheap side
- // product of triangulation anyway, allocate storage for them.
- if (!pMesh->mNormals && get_normals) {
- // XXX need a mechanism to inform the GenVertexNormals process to treat these normals as preprocessed per-face normals
- // nor_out = pMesh->mNormals = new aiVector3D[pMesh->mNumVertices];
- }
- // the output mesh will contain triangles, but no polys anymore
- pMesh->mPrimitiveTypes |= aiPrimitiveType_TRIANGLE;
- pMesh->mPrimitiveTypes &= ~aiPrimitiveType_POLYGON;
- // The mesh becomes NGON encoded now, during the triangulation process.
- pMesh->mPrimitiveTypes |= aiPrimitiveType_NGONEncodingFlag;
- aiFace* out = new aiFace[numOut](), *curOut = out;
- std::vector<aiVector3D> temp_verts3d(max_out+2); /* temporary storage for vertices */
- std::vector<aiVector2D> temp_verts(max_out+2);
- NGONEncoder ngonEncoder;
- // Apply vertex colors to represent the face winding?
- #ifdef AI_BUILD_TRIANGULATE_COLOR_FACE_WINDING
- if (!pMesh->mColors[0])
- pMesh->mColors[0] = new aiColor4D[pMesh->mNumVertices];
- else
- new(pMesh->mColors[0]) aiColor4D[pMesh->mNumVertices];
- aiColor4D* clr = pMesh->mColors[0];
- #endif
- #ifdef AI_BUILD_TRIANGULATE_DEBUG_POLYS
- FILE* fout = fopen(POLY_OUTPUT_FILE,"a");
- #endif
- const aiVector3D* verts = pMesh->mVertices;
- // use std::unique_ptr to avoid slow std::vector<bool> specialiations
- std::unique_ptr<bool[]> done(new bool[max_out]);
- for( unsigned int a = 0; a < pMesh->mNumFaces; a++) {
- aiFace& face = pMesh->mFaces[a];
- unsigned int* idx = face.mIndices;
- int num = (int)face.mNumIndices, ear = 0, tmp, prev = num-1, next = 0, max = num;
- // Apply vertex colors to represent the face winding?
- #ifdef AI_BUILD_TRIANGULATE_COLOR_FACE_WINDING
- for (unsigned int i = 0; i < face.mNumIndices; ++i) {
- aiColor4D& c = clr[idx[i]];
- c.r = (i+1) / (float)max;
- c.b = 1.f - c.r;
- }
- #endif
- aiFace* const last_face = curOut;
- // if it's a simple point,line or triangle: just copy it
- if( face.mNumIndices <= 3)
- {
- aiFace& nface = *curOut++;
- nface.mNumIndices = face.mNumIndices;
- nface.mIndices = face.mIndices;
- face.mIndices = nullptr;
- // points and lines don't require ngon encoding (and are not supported either!)
- if (nface.mNumIndices == 3) ngonEncoder.ngonEncodeTriangle(&nface);
- continue;
- }
- // optimized code for quadrilaterals
- else if ( face.mNumIndices == 4) {
- // quads can have at maximum one concave vertex. Determine
- // this vertex (if it exists) and start tri-fanning from
- // it.
- unsigned int start_vertex = 0;
- for (unsigned int i = 0; i < 4; ++i) {
- const aiVector3D& v0 = verts[face.mIndices[(i+3) % 4]];
- const aiVector3D& v1 = verts[face.mIndices[(i+2) % 4]];
- const aiVector3D& v2 = verts[face.mIndices[(i+1) % 4]];
- const aiVector3D& v = verts[face.mIndices[i]];
- aiVector3D left = (v0-v);
- aiVector3D diag = (v1-v);
- aiVector3D right = (v2-v);
- left.Normalize();
- diag.Normalize();
- right.Normalize();
- const float angle = std::acos(left*diag) + std::acos(right*diag);
- if (angle > AI_MATH_PI_F) {
- // this is the concave point
- start_vertex = i;
- break;
- }
- }
- const unsigned int temp[] = {face.mIndices[0], face.mIndices[1], face.mIndices[2], face.mIndices[3]};
- aiFace& nface = *curOut++;
- nface.mNumIndices = 3;
- nface.mIndices = face.mIndices;
- nface.mIndices[0] = temp[start_vertex];
- nface.mIndices[1] = temp[(start_vertex + 1) % 4];
- nface.mIndices[2] = temp[(start_vertex + 2) % 4];
- aiFace& sface = *curOut++;
- sface.mNumIndices = 3;
- sface.mIndices = new unsigned int[3];
- sface.mIndices[0] = temp[start_vertex];
- sface.mIndices[1] = temp[(start_vertex + 2) % 4];
- sface.mIndices[2] = temp[(start_vertex + 3) % 4];
- // prevent double deletion of the indices field
- face.mIndices = nullptr;
- ngonEncoder.ngonEncodeQuad(&nface, &sface);
- continue;
- }
- else
- {
- // A polygon with more than 3 vertices can be either concave or convex.
- // Usually everything we're getting is convex and we could easily
- // triangulate by tri-fanning. However, LightWave is probably the only
- // modeling suite to make extensive use of highly concave, monster polygons ...
- // so we need to apply the full 'ear cutting' algorithm to get it right.
- // REQUIREMENT: polygon is expected to be simple and *nearly* planar.
- // We project it onto a plane to get a 2d triangle.
- // Collect all vertices of of the polygon.
- for (tmp = 0; tmp < max; ++tmp) {
- temp_verts3d[tmp] = verts[idx[tmp]];
- }
- // Get newell normal of the polygon. Store it for future use if it's a polygon-only mesh
- aiVector3D n;
- NewellNormal<3,3,3>(n,max,&temp_verts3d.front().x,&temp_verts3d.front().y,&temp_verts3d.front().z);
- if (nor_out) {
- for (tmp = 0; tmp < max; ++tmp)
- nor_out[idx[tmp]] = n;
- }
- // Select largest normal coordinate to ignore for projection
- const float ax = (n.x>0 ? n.x : -n.x);
- const float ay = (n.y>0 ? n.y : -n.y);
- const float az = (n.z>0 ? n.z : -n.z);
- unsigned int ac = 0, bc = 1; /* no z coord. projection to xy */
- float inv = n.z;
- if (ax > ay) {
- if (ax > az) { /* no x coord. projection to yz */
- ac = 1; bc = 2;
- inv = n.x;
- }
- }
- else if (ay > az) { /* no y coord. projection to zy */
- ac = 2; bc = 0;
- inv = n.y;
- }
- // Swap projection axes to take the negated projection vector into account
- if (inv < 0.f) {
- std::swap(ac,bc);
- }
- for (tmp =0; tmp < max; ++tmp) {
- temp_verts[tmp].x = verts[idx[tmp]][ac];
- temp_verts[tmp].y = verts[idx[tmp]][bc];
- done[tmp] = false;
- }
- #ifdef AI_BUILD_TRIANGULATE_DEBUG_POLYS
- // plot the plane onto which we mapped the polygon to a 2D ASCII pic
- aiVector2D bmin,bmax;
- ArrayBounds(&temp_verts[0],max,bmin,bmax);
- char grid[POLY_GRID_Y][POLY_GRID_X+POLY_GRID_XPAD];
- std::fill_n((char*)grid,POLY_GRID_Y*(POLY_GRID_X+POLY_GRID_XPAD),' ');
- for (int i =0; i < max; ++i) {
- const aiVector2D& v = (temp_verts[i] - bmin) / (bmax-bmin);
- const size_t x = static_cast<size_t>(v.x*(POLY_GRID_X-1)), y = static_cast<size_t>(v.y*(POLY_GRID_Y-1));
- char* loc = grid[y]+x;
- if (grid[y][x] != ' ') {
- for(;*loc != ' '; ++loc);
- *loc++ = '_';
- }
- *(loc+::ai_snprintf(loc, POLY_GRID_XPAD,"%i",i)) = ' ';
- }
- for(size_t y = 0; y < POLY_GRID_Y; ++y) {
- grid[y][POLY_GRID_X+POLY_GRID_XPAD-1] = '\0';
- fprintf(fout,"%s\n",grid[y]);
- }
- fprintf(fout,"\ntriangulation sequence: ");
- #endif
- //
- // FIXME: currently this is the slow O(kn) variant with a worst case
- // complexity of O(n^2) (I think). Can be done in O(n).
- while (num > 3) {
- // Find the next ear of the polygon
- int num_found = 0;
- for (ear = next;;prev = ear,ear = next) {
- // break after we looped two times without a positive match
- for (next=ear+1;done[(next>=max?next=0:next)];++next);
- if (next < ear) {
- if (++num_found == 2) {
- break;
- }
- }
- const aiVector2D* pnt1 = &temp_verts[ear],
- *pnt0 = &temp_verts[prev],
- *pnt2 = &temp_verts[next];
- // Must be a convex point. Assuming ccw winding, it must be on the right of the line between p-1 and p+1.
- if (OnLeftSideOfLine2D(*pnt0,*pnt2,*pnt1) == 1) {
- continue;
- }
- // Skip when three point is in a line
- aiVector2D left = *pnt0 - *pnt1;
- aiVector2D right = *pnt2 - *pnt1;
- left.Normalize();
- right.Normalize();
- auto mul = left * right;
- // if the angle is 0 or 180
- if (std::abs(mul - 1.f) < ai_epsilon || std::abs(mul + 1.f) < ai_epsilon) {
- // skip this ear
- ASSIMP_LOG_WARN("Skip a ear, due to its angle is near 0 or 180.");
- continue;
- }
- // and no other point may be contained in this triangle
- for ( tmp = 0; tmp < max; ++tmp) {
- // We need to compare the actual values because it's possible that multiple indexes in
- // the polygon are referring to the same position. concave_polygon.obj is a sample
- //
- // FIXME: Use 'epsiloned' comparisons instead? Due to numeric inaccuracies in
- // PointInTriangle() I'm guessing that it's actually possible to construct
- // input data that would cause us to end up with no ears. The problem is,
- // which epsilon? If we chose a too large value, we'd get wrong results
- const aiVector2D& vtmp = temp_verts[tmp];
- if ( vtmp != *pnt1 && vtmp != *pnt2 && vtmp != *pnt0 && PointInTriangle2D(*pnt0,*pnt1,*pnt2,vtmp)) {
- break;
- }
- }
- if (tmp != max) {
- continue;
- }
- // this vertex is an ear
- break;
- }
- if (num_found == 2) {
- // Due to the 'two ear theorem', every simple polygon with more than three points must
- // have 2 'ears'. Here's definitely something wrong ... but we don't give up yet.
- //
- // Instead we're continuing with the standard tri-fanning algorithm which we'd
- // use if we had only convex polygons. That's life.
- ASSIMP_LOG_ERROR("Failed to triangulate polygon (no ear found). Probably not a simple polygon?");
- #ifdef AI_BUILD_TRIANGULATE_DEBUG_POLYS
- fprintf(fout,"critical error here, no ear found! ");
- #endif
- num = 0;
- break;
- }
- aiFace& nface = *curOut++;
- nface.mNumIndices = 3;
- if (!nface.mIndices) {
- nface.mIndices = new unsigned int[3];
- }
- // setup indices for the new triangle ...
- nface.mIndices[0] = prev;
- nface.mIndices[1] = ear;
- nface.mIndices[2] = next;
- // exclude the ear from most further processing
- done[ear] = true;
- --num;
- }
- if (num > 0) {
- // We have three indices forming the last 'ear' remaining. Collect them.
- aiFace& nface = *curOut++;
- nface.mNumIndices = 3;
- if (!nface.mIndices) {
- nface.mIndices = new unsigned int[3];
- }
- for (tmp = 0; done[tmp]; ++tmp);
- nface.mIndices[0] = tmp;
- for (++tmp; done[tmp]; ++tmp);
- nface.mIndices[1] = tmp;
- for (++tmp; done[tmp]; ++tmp);
- nface.mIndices[2] = tmp;
- }
- }
- #ifdef AI_BUILD_TRIANGULATE_DEBUG_POLYS
- for(aiFace* f = last_face; f != curOut; ++f) {
- unsigned int* i = f->mIndices;
- fprintf(fout," (%i %i %i)",i[0],i[1],i[2]);
- }
- fprintf(fout,"\n*********************************************************************\n");
- fflush(fout);
- #endif
- for(aiFace* f = last_face; f != curOut; ) {
- unsigned int* i = f->mIndices;
- i[0] = idx[i[0]];
- i[1] = idx[i[1]];
- i[2] = idx[i[2]];
- // IMPROVEMENT: Polygons are not supported yet by this ngon encoding + triangulation step.
- // So we encode polygons as regular triangles. No way to reconstruct the original
- // polygon in this case.
- ngonEncoder.ngonEncodeTriangle(f);
- ++f;
- }
- delete[] face.mIndices;
- face.mIndices = nullptr;
- }
- #ifdef AI_BUILD_TRIANGULATE_DEBUG_POLYS
- fclose(fout);
- #endif
- // kill the old faces
- delete [] pMesh->mFaces;
- // ... and store the new ones
- pMesh->mFaces = out;
- pMesh->mNumFaces = (unsigned int)(curOut-out); /* not necessarily equal to numOut */
- return true;
- }
- #endif // !! ASSIMP_BUILD_NO_TRIANGULATE_PROCESS
|