/* Open Asset Import Library (assimp) ---------------------------------------------------------------------- Copyright (c) 2006-2021, 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. ---------------------------------------------------------------------- */ #ifndef ASSIMP_BUILD_NO_EXPORT #ifndef ASSIMP_BUILD_NO_GLTF_EXPORTER #include "AssetLib/glTF2/glTF2Exporter.h" #include "AssetLib/glTF2/glTF2AssetWriter.h" #include "PostProcessing/SplitLargeMeshes.h" #include #include #include #include #include #include #include #include #include #include // Header files, standard library. #include #include #include using namespace rapidjson; using namespace Assimp; using namespace glTF2; namespace Assimp { // ------------------------------------------------------------------------------------------------ // Worker function for exporting a scene to GLTF. Prototyped and registered in Exporter.cpp void ExportSceneGLTF2(const char* pFile, IOSystem* pIOSystem, const aiScene* pScene, const ExportProperties* pProperties) { // invoke the exporter glTF2Exporter exporter(pFile, pIOSystem, pScene, pProperties, false); } // ------------------------------------------------------------------------------------------------ // Worker function for exporting a scene to GLB. Prototyped and registered in Exporter.cpp void ExportSceneGLB2(const char* pFile, IOSystem* pIOSystem, const aiScene* pScene, const ExportProperties* pProperties) { // invoke the exporter glTF2Exporter exporter(pFile, pIOSystem, pScene, pProperties, true); } } // end of namespace Assimp glTF2Exporter::glTF2Exporter(const char* filename, IOSystem* pIOSystem, const aiScene* pScene, const ExportProperties* pProperties, bool isBinary) : mFilename(filename) , mIOSystem(pIOSystem) , mProperties(pProperties) { mScene = pScene; mAsset.reset( new Asset( pIOSystem ) ); if (isBinary) { mAsset->SetAsBinary(); } ExportMetadata(); ExportMaterials(); if (mScene->mRootNode) { ExportNodeHierarchy(mScene->mRootNode); } ExportMeshes(); MergeMeshes(); ExportScene(); ExportAnimations(); // export extras if(mProperties->HasPropertyCallback("extras")) { std::function ExportExtras = mProperties->GetPropertyCallback("extras"); mAsset->extras = (rapidjson::Value*)ExportExtras(0); } AssetWriter writer(*mAsset); if (isBinary) { writer.WriteGLBFile(filename); } else { writer.WriteFile(filename); } } glTF2Exporter::~glTF2Exporter() { // empty } /* * Copy a 4x4 matrix from struct aiMatrix to typedef mat4. * Also converts from row-major to column-major storage. */ static void CopyValue(const aiMatrix4x4& v, mat4& o) { o[ 0] = v.a1; o[ 1] = v.b1; o[ 2] = v.c1; o[ 3] = v.d1; o[ 4] = v.a2; o[ 5] = v.b2; o[ 6] = v.c2; o[ 7] = v.d2; o[ 8] = v.a3; o[ 9] = v.b3; o[10] = v.c3; o[11] = v.d3; o[12] = v.a4; o[13] = v.b4; o[14] = v.c4; o[15] = v.d4; } static void CopyValue(const aiMatrix4x4& v, aiMatrix4x4& o) { memcpy(&o, &v, sizeof(aiMatrix4x4)); } static void IdentityMatrix4(mat4& o) { o[ 0] = 1; o[ 1] = 0; o[ 2] = 0; o[ 3] = 0; o[ 4] = 0; o[ 5] = 1; o[ 6] = 0; o[ 7] = 0; o[ 8] = 0; o[ 9] = 0; o[10] = 1; o[11] = 0; o[12] = 0; o[13] = 0; o[14] = 0; o[15] = 1; } template void SetAccessorRange(Ref acc, void* data, size_t count, unsigned int numCompsIn, unsigned int numCompsOut) { ai_assert(numCompsOut <= numCompsIn); // Allocate and initialize with large values. for (unsigned int i = 0 ; i < numCompsOut ; i++) { acc->min.push_back( std::numeric_limits::max()); acc->max.push_back(-std::numeric_limits::max()); } size_t totalComps = count * numCompsIn; T* buffer_ptr = static_cast(data); T* buffer_end = buffer_ptr + totalComps; // Search and set extreme values. for (; buffer_ptr < buffer_end ; buffer_ptr += numCompsIn) { for (unsigned int j = 0 ; j < numCompsOut ; j++) { double valueTmp = buffer_ptr[j]; // Gracefully tolerate rogue NaN's in buffer data // Any NaNs/Infs introduced in accessor bounds will end up in // document and prevent rapidjson from writing out valid JSON if (!std::isfinite(valueTmp)) { continue; } if (valueTmp < acc->min[j]) { acc->min[j] = valueTmp; } if (valueTmp > acc->max[j]) { acc->max[j] = valueTmp; } } } } inline void SetAccessorRange(ComponentType compType, Ref acc, void* data, size_t count, unsigned int numCompsIn, unsigned int numCompsOut) { switch (compType) { case ComponentType_SHORT: SetAccessorRange(acc, data, count, numCompsIn, numCompsOut); return; case ComponentType_UNSIGNED_SHORT: SetAccessorRange(acc, data, count, numCompsIn, numCompsOut); return; case ComponentType_UNSIGNED_INT: SetAccessorRange(acc, data, count, numCompsIn, numCompsOut); return; case ComponentType_FLOAT: SetAccessorRange(acc, data, count, numCompsIn, numCompsOut); return; case ComponentType_BYTE: SetAccessorRange(acc, data, count, numCompsIn, numCompsOut); return; case ComponentType_UNSIGNED_BYTE: SetAccessorRange(acc, data, count, numCompsIn, numCompsOut); return; } } // compute the (data-dataBase), store the non-zero data items template size_t NZDiff(void *data, void *dataBase, size_t count, unsigned int numCompsIn, unsigned int numCompsOut, void *&outputNZDiff, void *&outputNZIdx) { std::vector vNZDiff; std::vector vNZIdx; size_t totalComps = count * numCompsIn; T *bufferData_ptr = static_cast(data); T *bufferData_end = bufferData_ptr + totalComps; T *bufferBase_ptr = static_cast(dataBase); // Search and set extreme values. for (short idx = 0; bufferData_ptr < bufferData_end; idx += 1, bufferData_ptr += numCompsIn) { bool bNonZero = false; //for the data, check any component Non Zero for (unsigned int j = 0; j < numCompsOut; j++) { double valueData = bufferData_ptr[j]; double valueBase = bufferBase_ptr ? bufferBase_ptr[j] : 0; if ((valueData - valueBase) != 0) { bNonZero = true; break; } } //all zeros, continue if (!bNonZero) continue; //non zero, store the data for (unsigned int j = 0; j < numCompsOut; j++) { T valueData = bufferData_ptr[j]; T valueBase = bufferBase_ptr ? bufferBase_ptr[j] : 0; vNZDiff.push_back(valueData - valueBase); } vNZIdx.push_back(idx); } //avoid all-0, put 1 item if (vNZDiff.size() == 0) { for (unsigned int j = 0; j < numCompsOut; j++) vNZDiff.push_back(0); vNZIdx.push_back(0); } //process data outputNZDiff = new T[vNZDiff.size()]; memcpy(outputNZDiff, vNZDiff.data(), vNZDiff.size() * sizeof(T)); outputNZIdx = new unsigned short[vNZIdx.size()]; memcpy(outputNZIdx, vNZIdx.data(), vNZIdx.size() * sizeof(unsigned short)); return vNZIdx.size(); } inline size_t NZDiff(ComponentType compType, void *data, void *dataBase, size_t count, unsigned int numCompsIn, unsigned int numCompsOut, void *&nzDiff, void *&nzIdx) { switch (compType) { case ComponentType_SHORT: return NZDiff(data, dataBase, count, numCompsIn, numCompsOut, nzDiff, nzIdx); case ComponentType_UNSIGNED_SHORT: return NZDiff(data, dataBase, count, numCompsIn, numCompsOut, nzDiff, nzIdx); case ComponentType_UNSIGNED_INT: return NZDiff(data, dataBase, count, numCompsIn, numCompsOut, nzDiff, nzIdx); case ComponentType_FLOAT: return NZDiff(data, dataBase, count, numCompsIn, numCompsOut, nzDiff, nzIdx); case ComponentType_BYTE: return NZDiff(data, dataBase, count, numCompsIn, numCompsOut, nzDiff, nzIdx); case ComponentType_UNSIGNED_BYTE: return NZDiff(data, dataBase, count, numCompsIn, numCompsOut, nzDiff, nzIdx); } return 0; } inline Ref ExportDataSparse(Asset &a, std::string &meshName, Ref &buffer, size_t count, void *data, AttribType::Value typeIn, AttribType::Value typeOut, ComponentType compType, BufferViewTarget target = BufferViewTarget_NONE, void *dataBase = 0) { if (!count || !data) { return Ref(); } unsigned int numCompsIn = AttribType::GetNumComponents(typeIn); unsigned int numCompsOut = AttribType::GetNumComponents(typeOut); unsigned int bytesPerComp = ComponentTypeSize(compType); // accessor Ref acc = a.accessors.Create(a.FindUniqueID(meshName, "accessor")); // if there is a basic data vector if (dataBase) { size_t base_offset = buffer->byteLength; size_t base_padding = base_offset % bytesPerComp; base_offset += base_padding; size_t base_length = count * numCompsOut * bytesPerComp; buffer->Grow(base_length + base_padding); Ref bv = a.bufferViews.Create(a.FindUniqueID(meshName, "view")); bv->buffer = buffer; bv->byteOffset = base_offset; bv->byteLength = base_length; //! The target that the WebGL buffer should be bound to. bv->byteStride = 0; bv->target = target; acc->bufferView = bv; acc->WriteData(count, dataBase, numCompsIn * bytesPerComp); } acc->byteOffset = 0; acc->componentType = compType; acc->count = count; acc->type = typeOut; if (data) { void *nzDiff = 0, *nzIdx = 0; size_t nzCount = NZDiff(compType, data, dataBase, count, numCompsIn, numCompsOut, nzDiff, nzIdx); acc->sparse.reset(new Accessor::Sparse); acc->sparse->count = nzCount; //indices unsigned int bytesPerIdx = sizeof(unsigned short); size_t indices_offset = buffer->byteLength; size_t indices_padding = indices_offset % bytesPerIdx; indices_offset += indices_padding; size_t indices_length = nzCount * 1 * bytesPerIdx; buffer->Grow(indices_length + indices_padding); Ref indicesBV = a.bufferViews.Create(a.FindUniqueID(meshName, "view")); indicesBV->buffer = buffer; indicesBV->byteOffset = indices_offset; indicesBV->byteLength = indices_length; indicesBV->byteStride = 0; acc->sparse->indices = indicesBV; acc->sparse->indicesType = ComponentType_UNSIGNED_SHORT; acc->sparse->indicesByteOffset = 0; acc->WriteSparseIndices(nzCount, nzIdx, 1 * bytesPerIdx); //values size_t values_offset = buffer->byteLength; size_t values_padding = values_offset % bytesPerComp; values_offset += values_padding; size_t values_length = nzCount * numCompsOut * bytesPerComp; buffer->Grow(values_length + values_padding); Ref valuesBV = a.bufferViews.Create(a.FindUniqueID(meshName, "view")); valuesBV->buffer = buffer; valuesBV->byteOffset = values_offset; valuesBV->byteLength = values_length; valuesBV->byteStride = 0; acc->sparse->values = valuesBV; acc->sparse->valuesByteOffset = 0; acc->WriteSparseValues(nzCount, nzDiff, numCompsIn * bytesPerComp); //clear delete[] (char*)nzDiff; delete[] (char*)nzIdx; } return acc; } inline Ref ExportData(Asset& a, std::string& meshName, Ref& buffer, size_t count, void* data, AttribType::Value typeIn, AttribType::Value typeOut, ComponentType compType, BufferViewTarget target = BufferViewTarget_NONE) { if (!count || !data) { return Ref(); } unsigned int numCompsIn = AttribType::GetNumComponents(typeIn); unsigned int numCompsOut = AttribType::GetNumComponents(typeOut); unsigned int bytesPerComp = ComponentTypeSize(compType); size_t offset = buffer->byteLength; // make sure offset is correctly byte-aligned, as required by spec size_t padding = offset % bytesPerComp; offset += padding; size_t length = count * numCompsOut * bytesPerComp; buffer->Grow(length + padding); // bufferView Ref bv = a.bufferViews.Create(a.FindUniqueID(meshName, "view")); bv->buffer = buffer; bv->byteOffset = offset; bv->byteLength = length; //! The target that the WebGL buffer should be bound to. bv->byteStride = 0; bv->target = target; // accessor Ref acc = a.accessors.Create(a.FindUniqueID(meshName, "accessor")); acc->bufferView = bv; acc->byteOffset = 0; acc->componentType = compType; acc->count = count; acc->type = typeOut; // calculate min and max values SetAccessorRange(compType, acc, data, count, numCompsIn, numCompsOut); // copy the data acc->WriteData(count, data, numCompsIn*bytesPerComp); return acc; } inline void SetSamplerWrap(SamplerWrap& wrap, aiTextureMapMode map) { switch (map) { case aiTextureMapMode_Clamp: wrap = SamplerWrap::Clamp_To_Edge; break; case aiTextureMapMode_Mirror: wrap = SamplerWrap::Mirrored_Repeat; break; case aiTextureMapMode_Wrap: case aiTextureMapMode_Decal: default: wrap = SamplerWrap::Repeat; break; }; } void glTF2Exporter::GetTexSampler(const aiMaterial* mat, Ref texture, aiTextureType tt, unsigned int slot) { aiString aId; std::string id; if (aiGetMaterialString(mat, AI_MATKEY_GLTF_MAPPINGID(tt, slot), &aId) == AI_SUCCESS) { id = aId.C_Str(); } if (Ref ref = mAsset->samplers.Get(id.c_str())) { texture->sampler = ref; } else { id = mAsset->FindUniqueID(id, "sampler"); texture->sampler = mAsset->samplers.Create(id.c_str()); aiTextureMapMode mapU, mapV; SamplerMagFilter filterMag; SamplerMinFilter filterMin; if (aiGetMaterialInteger(mat, AI_MATKEY_MAPPINGMODE_U(tt, slot), (int*)&mapU) == AI_SUCCESS) { SetSamplerWrap(texture->sampler->wrapS, mapU); } if (aiGetMaterialInteger(mat, AI_MATKEY_MAPPINGMODE_V(tt, slot), (int*)&mapV) == AI_SUCCESS) { SetSamplerWrap(texture->sampler->wrapT, mapV); } if (aiGetMaterialInteger(mat, AI_MATKEY_GLTF_MAPPINGFILTER_MAG(tt, slot), (int*)&filterMag) == AI_SUCCESS) { texture->sampler->magFilter = filterMag; } if (aiGetMaterialInteger(mat, AI_MATKEY_GLTF_MAPPINGFILTER_MIN(tt, slot), (int*)&filterMin) == AI_SUCCESS) { texture->sampler->minFilter = filterMin; } aiString name; if (aiGetMaterialString(mat, AI_MATKEY_GLTF_MAPPINGNAME(tt, slot), &name) == AI_SUCCESS) { texture->sampler->name = name.C_Str(); } } } void glTF2Exporter::GetMatTexProp(const aiMaterial* mat, unsigned int& prop, const char* propName, aiTextureType tt, unsigned int slot) { std::string textureKey = std::string(_AI_MATKEY_TEXTURE_BASE) + "." + propName; mat->Get(textureKey.c_str(), tt, slot, prop); } void glTF2Exporter::GetMatTexProp(const aiMaterial* mat, float& prop, const char* propName, aiTextureType tt, unsigned int slot) { std::string textureKey = std::string(_AI_MATKEY_TEXTURE_BASE) + "." + propName; mat->Get(textureKey.c_str(), tt, slot, prop); } void glTF2Exporter::GetMatTex(const aiMaterial* mat, Ref& texture, aiTextureType tt, unsigned int slot = 0) { if (mat->GetTextureCount(tt) > 0) { aiString tex; if (mat->Get(AI_MATKEY_TEXTURE(tt, slot), tex) == AI_SUCCESS) { std::string path = tex.C_Str(); if (path.size() > 0) { std::map::iterator it = mTexturesByPath.find(path); if (it != mTexturesByPath.end()) { texture = mAsset->textures.Get(it->second); } if (!texture) { std::string texId = mAsset->FindUniqueID("", "texture"); texture = mAsset->textures.Create(texId); mTexturesByPath[path] = texture.GetIndex(); std::string imgId = mAsset->FindUniqueID("", "image"); texture->source = mAsset->images.Create(imgId); if (path[0] == '*') { // embedded aiTexture* curTex = mScene->mTextures[atoi(&path[1])]; texture->source->name = curTex->mFilename.C_Str(); // The asset has its own buffer, see Image::SetData texture->source->SetData(reinterpret_cast(curTex->pcData), curTex->mWidth, *mAsset); if (curTex->achFormatHint[0]) { std::string mimeType = "image/"; mimeType += (memcmp(curTex->achFormatHint, "jpg", 3) == 0) ? "jpeg" : curTex->achFormatHint; texture->source->mimeType = mimeType; } } else { texture->source->uri = path; } GetTexSampler(mat, texture, tt, slot); } } } } } void glTF2Exporter::GetMatTex(const aiMaterial* mat, TextureInfo& prop, aiTextureType tt, unsigned int slot = 0) { Ref& texture = prop.texture; GetMatTex(mat, texture, tt, slot); if (texture) { GetMatTexProp(mat, prop.texCoord, "texCoord", tt, slot); } } void glTF2Exporter::GetMatTex(const aiMaterial* mat, NormalTextureInfo& prop, aiTextureType tt, unsigned int slot = 0) { Ref& texture = prop.texture; GetMatTex(mat, texture, tt, slot); if (texture) { GetMatTexProp(mat, prop.texCoord, "texCoord", tt, slot); GetMatTexProp(mat, prop.scale, "scale", tt, slot); } } void glTF2Exporter::GetMatTex(const aiMaterial* mat, OcclusionTextureInfo& prop, aiTextureType tt, unsigned int slot = 0) { Ref& texture = prop.texture; GetMatTex(mat, texture, tt, slot); if (texture) { GetMatTexProp(mat, prop.texCoord, "texCoord", tt, slot); GetMatTexProp(mat, prop.strength, "strength", tt, slot); } } aiReturn glTF2Exporter::GetMatColor(const aiMaterial* mat, vec4& prop, const char* propName, int type, int idx) { aiColor4D col; aiReturn result = mat->Get(propName, type, idx, col); if (result == AI_SUCCESS) { prop[0] = col.r; prop[1] = col.g; prop[2] = col.b; prop[3] = col.a; } return result; } aiReturn glTF2Exporter::GetMatColor(const aiMaterial* mat, vec3& prop, const char* propName, int type, int idx) { aiColor3D col; aiReturn result = mat->Get(propName, type, idx, col); if (result == AI_SUCCESS) { prop[0] = col.r; prop[1] = col.g; prop[2] = col.b; } return result; } void glTF2Exporter::ExportMaterials() { aiString aiName; for (unsigned int i = 0; i < mScene->mNumMaterials; ++i) { const aiMaterial* mat = mScene->mMaterials[i]; std::string id = "material_" + ai_to_string(i); Ref m = mAsset->materials.Create(id); std::string name; if (mat->Get(AI_MATKEY_NAME, aiName) == AI_SUCCESS) { name = aiName.C_Str(); } name = mAsset->FindUniqueID(name, "material"); m->name = name; GetMatTex(mat, m->pbrMetallicRoughness.baseColorTexture, AI_MATKEY_GLTF_PBRMETALLICROUGHNESS_BASE_COLOR_TEXTURE); if (!m->pbrMetallicRoughness.baseColorTexture.texture) { //if there wasn't a baseColorTexture defined in the source, fallback to any diffuse texture GetMatTex(mat, m->pbrMetallicRoughness.baseColorTexture, aiTextureType_DIFFUSE); } GetMatTex(mat, m->pbrMetallicRoughness.metallicRoughnessTexture, AI_MATKEY_GLTF_PBRMETALLICROUGHNESS_METALLICROUGHNESS_TEXTURE); if (GetMatColor(mat, m->pbrMetallicRoughness.baseColorFactor, AI_MATKEY_GLTF_PBRMETALLICROUGHNESS_BASE_COLOR_FACTOR) != AI_SUCCESS) { // if baseColorFactor wasn't defined, then the source is likely not a metallic roughness material. //a fallback to any diffuse color should be used instead GetMatColor(mat, m->pbrMetallicRoughness.baseColorFactor, AI_MATKEY_COLOR_DIFFUSE); } if (mat->Get(AI_MATKEY_GLTF_PBRMETALLICROUGHNESS_METALLIC_FACTOR, m->pbrMetallicRoughness.metallicFactor) != AI_SUCCESS) { //if metallicFactor wasn't defined, then the source is likely not a PBR file, and the metallicFactor should be 0 m->pbrMetallicRoughness.metallicFactor = 0; } // get roughness if source is gltf2 file if (mat->Get(AI_MATKEY_GLTF_PBRMETALLICROUGHNESS_ROUGHNESS_FACTOR, m->pbrMetallicRoughness.roughnessFactor) != AI_SUCCESS) { // otherwise, try to derive and convert from specular + shininess values aiColor4D specularColor; ai_real shininess; if ( mat->Get(AI_MATKEY_COLOR_SPECULAR, specularColor) == AI_SUCCESS && mat->Get(AI_MATKEY_SHININESS, shininess) == AI_SUCCESS ) { // convert specular color to luminance float specularIntensity = specularColor[0] * 0.2125f + specularColor[1] * 0.7154f + specularColor[2] * 0.0721f; //normalize shininess (assuming max is 1000) with an inverse exponentional curve float normalizedShininess = std::sqrt(shininess / 1000); //clamp the shininess value between 0 and 1 normalizedShininess = std::min(std::max(normalizedShininess, 0.0f), 1.0f); // low specular intensity values should produce a rough material even if shininess is high. normalizedShininess = normalizedShininess * specularIntensity; m->pbrMetallicRoughness.roughnessFactor = 1 - normalizedShininess; } } GetMatTex(mat, m->normalTexture, aiTextureType_NORMALS); GetMatTex(mat, m->occlusionTexture, aiTextureType_LIGHTMAP); GetMatTex(mat, m->emissiveTexture, aiTextureType_EMISSIVE); GetMatColor(mat, m->emissiveFactor, AI_MATKEY_COLOR_EMISSIVE); mat->Get(AI_MATKEY_TWOSIDED, m->doubleSided); mat->Get(AI_MATKEY_GLTF_ALPHACUTOFF, m->alphaCutoff); aiString alphaMode; if (mat->Get(AI_MATKEY_GLTF_ALPHAMODE, alphaMode) == AI_SUCCESS) { m->alphaMode = alphaMode.C_Str(); } else { float opacity; if (mat->Get(AI_MATKEY_OPACITY, opacity) == AI_SUCCESS) { if (opacity < 1) { m->alphaMode = "BLEND"; m->pbrMetallicRoughness.baseColorFactor[3] *= opacity; } } } bool hasPbrSpecularGlossiness = false; mat->Get(AI_MATKEY_GLTF_PBRSPECULARGLOSSINESS, hasPbrSpecularGlossiness); if (hasPbrSpecularGlossiness) { if (!mAsset->extensionsUsed.KHR_materials_pbrSpecularGlossiness) { mAsset->extensionsUsed.KHR_materials_pbrSpecularGlossiness = true; } PbrSpecularGlossiness pbrSG; GetMatColor(mat, pbrSG.diffuseFactor, AI_MATKEY_COLOR_DIFFUSE); GetMatColor(mat, pbrSG.specularFactor, AI_MATKEY_COLOR_SPECULAR); if (mat->Get(AI_MATKEY_GLTF_PBRSPECULARGLOSSINESS_GLOSSINESS_FACTOR, pbrSG.glossinessFactor) != AI_SUCCESS) { float shininess; if (mat->Get(AI_MATKEY_SHININESS, shininess) == AI_SUCCESS) { pbrSG.glossinessFactor = shininess / 1000; } } GetMatTex(mat, pbrSG.diffuseTexture, aiTextureType_DIFFUSE); GetMatTex(mat, pbrSG.specularGlossinessTexture, aiTextureType_SPECULAR); m->pbrSpecularGlossiness = Nullable(pbrSG); } bool unlit; if (mat->Get(AI_MATKEY_GLTF_UNLIT, unlit) == AI_SUCCESS && unlit) { mAsset->extensionsUsed.KHR_materials_unlit = true; m->unlit = true; } bool hasMaterialSheen = false; mat->Get(AI_MATKEY_GLTF_MATERIAL_SHEEN, hasMaterialSheen); if (hasMaterialSheen) { mAsset->extensionsUsed.KHR_materials_sheen = true; MaterialSheen sheen; GetMatColor(mat, sheen.sheenColorFactor, AI_MATKEY_GLTF_MATERIAL_SHEEN_COLOR_FACTOR); mat->Get(AI_MATKEY_GLTF_MATERIAL_SHEEN_ROUGHNESS_FACTOR, sheen.sheenRoughnessFactor); GetMatTex(mat, sheen.sheenColorTexture, AI_MATKEY_GLTF_MATERIAL_SHEEN_COLOR_TEXTURE); GetMatTex(mat, sheen.sheenRoughnessTexture, AI_MATKEY_GLTF_MATERIAL_SHEEN_ROUGHNESS_TEXTURE); m->materialSheen = Nullable(sheen); } bool hasMaterialClearcoat = false; mat->Get(AI_MATKEY_GLTF_MATERIAL_CLEARCOAT, hasMaterialClearcoat); if (hasMaterialClearcoat) { mAsset->extensionsUsed.KHR_materials_clearcoat= true; MaterialClearcoat clearcoat; mat->Get(AI_MATKEY_GLTF_MATERIAL_CLEARCOAT_FACTOR, clearcoat.clearcoatFactor); mat->Get(AI_MATKEY_GLTF_MATERIAL_CLEARCOAT_ROUGHNESS_FACTOR, clearcoat.clearcoatRoughnessFactor); GetMatTex(mat, clearcoat.clearcoatTexture, AI_MATKEY_GLTF_MATERIAL_CLEARCOAT_TEXTURE); GetMatTex(mat, clearcoat.clearcoatRoughnessTexture, AI_MATKEY_GLTF_MATERIAL_CLEARCOAT_ROUGHNESS_TEXTURE); GetMatTex(mat, clearcoat.clearcoatNormalTexture, AI_MATKEY_GLTF_MATERIAL_CLEARCOAT_NORMAL_TEXTURE); m->materialClearcoat = Nullable(clearcoat); } bool hasMaterialTransmission = false; mat->Get(AI_MATKEY_GLTF_MATERIAL_TRANSMISSION, hasMaterialTransmission); if (hasMaterialTransmission) { mAsset->extensionsUsed.KHR_materials_transmission = true; MaterialTransmission transmission; mat->Get(AI_MATKEY_GLTF_MATERIAL_TRANSMISSION_FACTOR, transmission.transmissionFactor); GetMatTex(mat, transmission.transmissionTexture, AI_MATKEY_GLTF_MATERIAL_TRANSMISSION_TEXTURE); m->materialTransmission = Nullable(transmission); } } } /* * Search through node hierarchy and find the node containing the given meshID. * Returns true on success, and false otherwise. */ bool FindMeshNode(Ref& nodeIn, Ref& meshNode, std::string meshID) { for (unsigned int i = 0; i < nodeIn->meshes.size(); ++i) { if (meshID.compare(nodeIn->meshes[i]->id) == 0) { meshNode = nodeIn; return true; } } for (unsigned int i = 0; i < nodeIn->children.size(); ++i) { if(FindMeshNode(nodeIn->children[i], meshNode, meshID)) { return true; } } return false; } /* * Find the root joint of the skeleton. * Starts will any joint node and traces up the tree, * until a parent is found that does not have a jointName. * Returns the first parent Ref found that does not have a jointName. */ Ref FindSkeletonRootJoint(Ref& skinRef) { Ref startNodeRef; Ref parentNodeRef; // Arbitrarily use the first joint to start the search. startNodeRef = skinRef->jointNames[0]; parentNodeRef = skinRef->jointNames[0]; do { startNodeRef = parentNodeRef; parentNodeRef = startNodeRef->parent; } while (!parentNodeRef->jointName.empty()); return parentNodeRef; } void ExportSkin(Asset& mAsset, const aiMesh* aimesh, Ref& meshRef, Ref& bufferRef, Ref& skinRef, std::vector& inverseBindMatricesData) { if (aimesh->mNumBones < 1) { return; } // Store the vertex joint and weight data. const size_t NumVerts( aimesh->mNumVertices ); vec4* vertexJointData = new vec4[ NumVerts ]; vec4* vertexWeightData = new vec4[ NumVerts ]; int* jointsPerVertex = new int[ NumVerts ]; for (size_t i = 0; i < NumVerts; ++i) { jointsPerVertex[i] = 0; for (size_t j = 0; j < 4; ++j) { vertexJointData[i][j] = 0; vertexWeightData[i][j] = 0; } } for (unsigned int idx_bone = 0; idx_bone < aimesh->mNumBones; ++idx_bone) { const aiBone* aib = aimesh->mBones[idx_bone]; // aib->mName =====> skinRef->jointNames // Find the node with id = mName. Ref nodeRef = mAsset.nodes.Get(aib->mName.C_Str()); nodeRef->jointName = nodeRef->name; unsigned int jointNamesIndex = 0; bool addJointToJointNames = true; for ( unsigned int idx_joint = 0; idx_joint < skinRef->jointNames.size(); ++idx_joint) { if (skinRef->jointNames[idx_joint]->jointName.compare(nodeRef->jointName) == 0) { addJointToJointNames = false; jointNamesIndex = idx_joint; } } if (addJointToJointNames) { skinRef->jointNames.push_back(nodeRef); // aib->mOffsetMatrix =====> skinRef->inverseBindMatrices aiMatrix4x4 tmpMatrix4; CopyValue(aib->mOffsetMatrix, tmpMatrix4); inverseBindMatricesData.push_back(tmpMatrix4); jointNamesIndex = static_cast(inverseBindMatricesData.size() - 1); } // aib->mWeights =====> vertexWeightData for (unsigned int idx_weights = 0; idx_weights < aib->mNumWeights; ++idx_weights) { unsigned int vertexId = aib->mWeights[idx_weights].mVertexId; float vertWeight = aib->mWeights[idx_weights].mWeight; // A vertex can only have at most four joint weights. Ignore all others. if (jointsPerVertex[vertexId] > 3) { continue; } vertexJointData[vertexId][jointsPerVertex[vertexId]] = static_cast(jointNamesIndex); vertexWeightData[vertexId][jointsPerVertex[vertexId]] = vertWeight; jointsPerVertex[vertexId] += 1; } } // End: for-loop mNumMeshes Mesh::Primitive& p = meshRef->primitives.back(); Ref vertexJointAccessor = ExportData(mAsset, skinRef->id, bufferRef, aimesh->mNumVertices, vertexJointData, AttribType::VEC4, AttribType::VEC4, ComponentType_FLOAT); if ( vertexJointAccessor ) { size_t offset = vertexJointAccessor->bufferView->byteOffset; size_t bytesLen = vertexJointAccessor->bufferView->byteLength; unsigned int s_bytesPerComp= ComponentTypeSize(ComponentType_UNSIGNED_SHORT); unsigned int bytesPerComp = ComponentTypeSize(vertexJointAccessor->componentType); size_t s_bytesLen = bytesLen * s_bytesPerComp / bytesPerComp; Ref buf = vertexJointAccessor->bufferView->buffer; uint8_t* arrys = new uint8_t[bytesLen]; unsigned int i = 0; for ( unsigned int j = 0; j <= bytesLen; j += bytesPerComp ){ size_t len_p = offset + j; float f_value = *(float *)&buf->GetPointer()[len_p]; unsigned short c = static_cast(f_value); memcpy(&arrys[i*s_bytesPerComp], &c, s_bytesPerComp); ++i; } buf->ReplaceData_joint(offset, bytesLen, arrys, bytesLen); vertexJointAccessor->componentType = ComponentType_UNSIGNED_SHORT; vertexJointAccessor->bufferView->byteLength = s_bytesLen; p.attributes.joint.push_back( vertexJointAccessor ); delete[] arrys; } Ref vertexWeightAccessor = ExportData(mAsset, skinRef->id, bufferRef, aimesh->mNumVertices, vertexWeightData, AttribType::VEC4, AttribType::VEC4, ComponentType_FLOAT); if ( vertexWeightAccessor ) { p.attributes.weight.push_back( vertexWeightAccessor ); } delete[] jointsPerVertex; delete[] vertexWeightData; delete[] vertexJointData; } void glTF2Exporter::ExportMeshes() { typedef decltype(aiFace::mNumIndices) IndicesType; std::string fname = std::string(mFilename); std::string bufferIdPrefix = fname.substr(0, fname.rfind(".gltf")); std::string bufferId = mAsset->FindUniqueID("", bufferIdPrefix.c_str()); Ref b = mAsset->GetBodyBuffer(); if (!b) { b = mAsset->buffers.Create(bufferId); } //---------------------------------------- // Initialize variables for the skin bool createSkin = false; for (unsigned int idx_mesh = 0; idx_mesh < mScene->mNumMeshes; ++idx_mesh) { const aiMesh* aim = mScene->mMeshes[idx_mesh]; if(aim->HasBones()) { createSkin = true; break; } } Ref skinRef; std::string skinName = mAsset->FindUniqueID("skin", "skin"); std::vector inverseBindMatricesData; if(createSkin) { skinRef = mAsset->skins.Create(skinName); skinRef->name = skinName; } //---------------------------------------- for (unsigned int idx_mesh = 0; idx_mesh < mScene->mNumMeshes; ++idx_mesh) { const aiMesh* aim = mScene->mMeshes[idx_mesh]; std::string name = aim->mName.C_Str(); std::string meshId = mAsset->FindUniqueID(name, "mesh"); Ref m = mAsset->meshes.Create(meshId); m->primitives.resize(1); Mesh::Primitive& p = m->primitives.back(); m->name = name; p.material = mAsset->materials.Get(aim->mMaterialIndex); /******************* Vertices ********************/ Ref v = ExportData(*mAsset, meshId, b, aim->mNumVertices, aim->mVertices, AttribType::VEC3, AttribType::VEC3, ComponentType_FLOAT, BufferViewTarget_ARRAY_BUFFER); if (v) p.attributes.position.push_back(v); /******************** Normals ********************/ // Normalize all normals as the validator can emit a warning otherwise if ( nullptr != aim->mNormals) { for ( auto i = 0u; i < aim->mNumVertices; ++i ) { aim->mNormals[ i ].NormalizeSafe(); } } Ref n = ExportData(*mAsset, meshId, b, aim->mNumVertices, aim->mNormals, AttribType::VEC3, AttribType::VEC3, ComponentType_FLOAT, BufferViewTarget_ARRAY_BUFFER); if (n) p.attributes.normal.push_back(n); /************** Texture coordinates **************/ for (int i = 0; i < AI_MAX_NUMBER_OF_TEXTURECOORDS; ++i) { if (!aim->HasTextureCoords(i)) continue; // Flip UV y coords if (aim -> mNumUVComponents[i] > 1) { for (unsigned int j = 0; j < aim->mNumVertices; ++j) { aim->mTextureCoords[i][j].y = 1 - aim->mTextureCoords[i][j].y; } } if (aim->mNumUVComponents[i] > 0) { AttribType::Value type = (aim->mNumUVComponents[i] == 2) ? AttribType::VEC2 : AttribType::VEC3; Ref tc = ExportData(*mAsset, meshId, b, aim->mNumVertices, aim->mTextureCoords[i], AttribType::VEC3, type, ComponentType_FLOAT, BufferViewTarget_ARRAY_BUFFER); if (tc) p.attributes.texcoord.push_back(tc); } } /*************** Vertex colors ****************/ for (unsigned int indexColorChannel = 0; indexColorChannel < aim->GetNumColorChannels(); ++indexColorChannel) { Ref c = ExportData(*mAsset, meshId, b, aim->mNumVertices, aim->mColors[indexColorChannel], AttribType::VEC4, AttribType::VEC4, ComponentType_FLOAT, BufferViewTarget_ARRAY_BUFFER); if (c) p.attributes.color.push_back(c); } /*************** Vertices indices ****************/ if (aim->mNumFaces > 0) { std::vector indices; unsigned int nIndicesPerFace = aim->mFaces[0].mNumIndices; indices.resize(aim->mNumFaces * nIndicesPerFace); for (size_t i = 0; i < aim->mNumFaces; ++i) { for (size_t j = 0; j < nIndicesPerFace; ++j) { indices[i*nIndicesPerFace + j] = IndicesType(aim->mFaces[i].mIndices[j]); } } p.indices = ExportData(*mAsset, meshId, b, indices.size(), &indices[0], AttribType::SCALAR, AttribType::SCALAR, ComponentType_UNSIGNED_INT, BufferViewTarget_ELEMENT_ARRAY_BUFFER); } switch (aim->mPrimitiveTypes) { case aiPrimitiveType_POLYGON: p.mode = PrimitiveMode_TRIANGLES; break; // TODO implement this case aiPrimitiveType_LINE: p.mode = PrimitiveMode_LINES; break; case aiPrimitiveType_POINT: p.mode = PrimitiveMode_POINTS; break; default: // aiPrimitiveType_TRIANGLE p.mode = PrimitiveMode_TRIANGLES; } /*************** Skins ****************/ if(aim->HasBones()) { ExportSkin(*mAsset, aim, m, b, skinRef, inverseBindMatricesData); } /*************** Targets for blendshapes ****************/ if (aim->mNumAnimMeshes > 0) { bool bUseSparse = this->mProperties->HasPropertyBool("GLTF2_SPARSE_ACCESSOR_EXP") && this->mProperties->GetPropertyBool("GLTF2_SPARSE_ACCESSOR_EXP"); bool bIncludeNormal = this->mProperties->HasPropertyBool("GLTF2_TARGET_NORMAL_EXP") && this->mProperties->GetPropertyBool("GLTF2_TARGET_NORMAL_EXP"); bool bExportTargetNames = this->mProperties->HasPropertyBool("GLTF2_TARGETNAMES_EXP") && this->mProperties->GetPropertyBool("GLTF2_TARGETNAMES_EXP"); p.targets.resize(aim->mNumAnimMeshes); for (unsigned int am = 0; am < aim->mNumAnimMeshes; ++am) { aiAnimMesh *pAnimMesh = aim->mAnimMeshes[am]; if (bExportTargetNames) m->targetNames.push_back(pAnimMesh->mName.data); // position if (pAnimMesh->HasPositions()) { // NOTE: in gltf it is the diff stored aiVector3D *pPositionDiff = new aiVector3D[pAnimMesh->mNumVertices]; for (unsigned int vt = 0; vt < pAnimMesh->mNumVertices; ++vt) { pPositionDiff[vt] = pAnimMesh->mVertices[vt] - aim->mVertices[vt]; } Ref vec; if (bUseSparse) { vec = ExportDataSparse(*mAsset, meshId, b, pAnimMesh->mNumVertices, pPositionDiff, AttribType::VEC3, AttribType::VEC3, ComponentType_FLOAT); } else { vec = ExportData(*mAsset, meshId, b, pAnimMesh->mNumVertices, pPositionDiff, AttribType::VEC3, AttribType::VEC3, ComponentType_FLOAT); } if (vec) { p.targets[am].position.push_back(vec); } delete[] pPositionDiff; } // normal if (pAnimMesh->HasNormals() && bIncludeNormal) { aiVector3D *pNormalDiff = new aiVector3D[pAnimMesh->mNumVertices]; for (unsigned int vt = 0; vt < pAnimMesh->mNumVertices; ++vt) { pNormalDiff[vt] = pAnimMesh->mNormals[vt] - aim->mNormals[vt]; } Ref vec; if (bUseSparse) { vec = ExportDataSparse(*mAsset, meshId, b, pAnimMesh->mNumVertices, pNormalDiff, AttribType::VEC3, AttribType::VEC3, ComponentType_FLOAT); } else { vec = ExportData(*mAsset, meshId, b, pAnimMesh->mNumVertices, pNormalDiff, AttribType::VEC3, AttribType::VEC3, ComponentType_FLOAT); } if (vec) { p.targets[am].normal.push_back(vec); } delete[] pNormalDiff; } // tangent? } } } //---------------------------------------- // Finish the skin // Create the Accessor for skinRef->inverseBindMatrices bool bAddCustomizedProperty = this->mProperties->HasPropertyBool("GLTF2_CUSTOMIZE_PROPERTY"); if (createSkin) { mat4* invBindMatrixData = new mat4[inverseBindMatricesData.size()]; for ( unsigned int idx_joint = 0; idx_joint < inverseBindMatricesData.size(); ++idx_joint) { CopyValue(inverseBindMatricesData[idx_joint], invBindMatrixData[idx_joint]); } Ref invBindMatrixAccessor = ExportData(*mAsset, skinName, b, static_cast(inverseBindMatricesData.size()), invBindMatrixData, AttribType::MAT4, AttribType::MAT4, ComponentType_FLOAT); if (invBindMatrixAccessor) { skinRef->inverseBindMatrices = invBindMatrixAccessor; } // Identity Matrix =====> skinRef->bindShapeMatrix // Temporary. Hard-coded identity matrix here skinRef->bindShapeMatrix.isPresent = bAddCustomizedProperty; IdentityMatrix4(skinRef->bindShapeMatrix.value); // Find nodes that contain a mesh with bones and add "skeletons" and "skin" attributes to those nodes. Ref rootNode = mAsset->nodes.Get(unsigned(0)); Ref meshNode; for (unsigned int meshIndex = 0; meshIndex < mAsset->meshes.Size(); ++meshIndex) { Ref mesh = mAsset->meshes.Get(meshIndex); bool hasBones = false; for (unsigned int i = 0; i < mesh->primitives.size(); ++i) { if (!mesh->primitives[i].attributes.weight.empty()) { hasBones = true; break; } } if (!hasBones) { continue; } std::string meshID = mesh->id; FindMeshNode(rootNode, meshNode, meshID); Ref rootJoint = FindSkeletonRootJoint(skinRef); if(bAddCustomizedProperty) meshNode->skeletons.push_back(rootJoint); meshNode->skin = skinRef; } delete[] invBindMatrixData; } } // Merges a node's multiple meshes (with one primitive each) into one mesh with multiple primitives void glTF2Exporter::MergeMeshes() { for (unsigned int n = 0; n < mAsset->nodes.Size(); ++n) { Ref node = mAsset->nodes.Get(n); unsigned int nMeshes = static_cast(node->meshes.size()); //skip if it's 1 or less meshes per node if (nMeshes > 1) { Ref firstMesh = node->meshes.at(0); //loop backwards to allow easy removal of a mesh from a node once it's merged for (unsigned int m = nMeshes - 1; m >= 1; --m) { Ref mesh = node->meshes.at(m); //append this mesh's primitives to the first mesh's primitives firstMesh->primitives.insert( firstMesh->primitives.end(), mesh->primitives.begin(), mesh->primitives.end() ); //remove the mesh from the list of meshes unsigned int removedIndex = mAsset->meshes.Remove(mesh->id.c_str()); //find the presence of the removed mesh in other nodes for (unsigned int nn = 0; nn < mAsset->nodes.Size(); ++nn) { Ref curNode = mAsset->nodes.Get(nn); for (unsigned int mm = 0; mm < curNode->meshes.size(); ++mm) { Ref &meshRef = curNode->meshes.at(mm); unsigned int meshIndex = meshRef.GetIndex(); if (meshIndex == removedIndex) { curNode->meshes.erase(curNode->meshes.begin() + mm); } else if (meshIndex > removedIndex) { Ref newMeshRef = mAsset->meshes.Get(meshIndex - 1); meshRef = newMeshRef; } } } } //since we were looping backwards, reverse the order of merged primitives to their original order std::reverse(firstMesh->primitives.begin() + 1, firstMesh->primitives.end()); } } } /* * Export the root node of the node hierarchy. * Calls ExportNode for all children. */ unsigned int glTF2Exporter::ExportNodeHierarchy(const aiNode* n) { Ref node = mAsset->nodes.Create(mAsset->FindUniqueID(n->mName.C_Str(), "node")); node->name = n->mName.C_Str(); if (!n->mTransformation.IsIdentity()) { node->matrix.isPresent = true; CopyValue(n->mTransformation, node->matrix.value); } for (unsigned int i = 0; i < n->mNumMeshes; ++i) { node->meshes.push_back(mAsset->meshes.Get(n->mMeshes[i])); } for (unsigned int i = 0; i < n->mNumChildren; ++i) { unsigned int idx = ExportNode(n->mChildren[i], node); node->children.push_back(mAsset->nodes.Get(idx)); } return node.GetIndex(); } /* * Export node and recursively calls ExportNode for all children. * Since these nodes are not the root node, we also export the parent Ref */ unsigned int glTF2Exporter::ExportNode(const aiNode* n, Ref& parent) { std::string name = mAsset->FindUniqueID(n->mName.C_Str(), "node"); Ref node = mAsset->nodes.Create(name); node->parent = parent; node->name = name; if (!n->mTransformation.IsIdentity()) { if (mScene->mNumAnimations > 0 || (mProperties && mProperties->HasPropertyBool("GLTF2_NODE_IN_TRS"))) { aiQuaternion quaternion; n->mTransformation.Decompose(*reinterpret_cast(&node->scale.value), quaternion, *reinterpret_cast(&node->translation.value)); aiVector3D vector(static_cast(1.0f), static_cast(1.0f), static_cast(1.0f)); if (!reinterpret_cast(&node->scale.value)->Equal(vector)) { node->scale.isPresent = true; } if (!reinterpret_cast(&node->translation.value)->Equal(vector)) { node->translation.isPresent = true; } node->rotation.isPresent = true; node->rotation.value[0] = quaternion.x; node->rotation.value[1] = quaternion.y; node->rotation.value[2] = quaternion.z; node->rotation.value[3] = quaternion.w; node->matrix.isPresent = false; } else { node->matrix.isPresent = true; CopyValue(n->mTransformation, node->matrix.value); } } for (unsigned int i = 0; i < n->mNumMeshes; ++i) { node->meshes.push_back(mAsset->meshes.Get(n->mMeshes[i])); } for (unsigned int i = 0; i < n->mNumChildren; ++i) { unsigned int idx = ExportNode(n->mChildren[i], node); node->children.push_back(mAsset->nodes.Get(idx)); } return node.GetIndex(); } void glTF2Exporter::ExportScene() { const char* sceneName = "defaultScene"; Ref scene = mAsset->scenes.Create(sceneName); // root node will be the first one exported (idx 0) if (mAsset->nodes.Size() > 0) { scene->nodes.push_back(mAsset->nodes.Get(0u)); } // set as the default scene mAsset->scene = scene; } void glTF2Exporter::ExportMetadata() { AssetMetadata& asset = mAsset->asset; asset.version = "2.0"; char buffer[256]; ai_snprintf(buffer, 256, "Open Asset Import Library (assimp v%d.%d.%x)", aiGetVersionMajor(), aiGetVersionMinor(), aiGetVersionRevision()); asset.generator = buffer; // Copyright aiString copyright_str; if (mScene->mMetaData != nullptr && mScene->mMetaData->Get(AI_METADATA_SOURCE_COPYRIGHT, copyright_str)) { asset.copyright = copyright_str.C_Str(); } } inline Ref GetSamplerInputRef(Asset& asset, std::string& animId, Ref& buffer, std::vector& times) { return ExportData(asset, animId, buffer, (unsigned int)times.size(), ×[0], AttribType::SCALAR, AttribType::SCALAR, ComponentType_FLOAT); } inline void ExtractTranslationSampler(Asset& asset, std::string& animId, Ref& buffer, const aiNodeAnim* nodeChannel, float ticksPerSecond, Animation::Sampler& sampler) { const unsigned int numKeyframes = nodeChannel->mNumPositionKeys; std::vector times(numKeyframes); std::vector values(numKeyframes * 3); for (unsigned int i = 0; i < numKeyframes; ++i) { const aiVectorKey& key = nodeChannel->mPositionKeys[i]; // mTime is measured in ticks, but GLTF time is measured in seconds, so convert. times[i] = static_cast(key.mTime / ticksPerSecond); values[(i * 3) + 0] = key.mValue.x; values[(i * 3) + 1] = key.mValue.y; values[(i * 3) + 2] = key.mValue.z; } sampler.input = GetSamplerInputRef(asset, animId, buffer, times); sampler.output = ExportData(asset, animId, buffer, numKeyframes, &values[0], AttribType::VEC3, AttribType::VEC3, ComponentType_FLOAT); sampler.interpolation = Interpolation_LINEAR; } inline void ExtractScaleSampler(Asset& asset, std::string& animId, Ref& buffer, const aiNodeAnim* nodeChannel, float ticksPerSecond, Animation::Sampler& sampler) { const unsigned int numKeyframes = nodeChannel->mNumScalingKeys; std::vector times(numKeyframes); std::vector values(numKeyframes * 3); for (unsigned int i = 0; i < numKeyframes; ++i) { const aiVectorKey& key = nodeChannel->mScalingKeys[i]; // mTime is measured in ticks, but GLTF time is measured in seconds, so convert. times[i] = static_cast(key.mTime / ticksPerSecond); values[(i * 3) + 0] = key.mValue.x; values[(i * 3) + 1] = key.mValue.y; values[(i * 3) + 2] = key.mValue.z; } sampler.input = GetSamplerInputRef(asset, animId, buffer, times); sampler.output = ExportData(asset, animId, buffer, numKeyframes, &values[0], AttribType::VEC3, AttribType::VEC3, ComponentType_FLOAT); sampler.interpolation = Interpolation_LINEAR; } inline void ExtractRotationSampler(Asset& asset, std::string& animId, Ref& buffer, const aiNodeAnim* nodeChannel, float ticksPerSecond, Animation::Sampler& sampler) { const unsigned int numKeyframes = nodeChannel->mNumRotationKeys; std::vector times(numKeyframes); std::vector values(numKeyframes * 4); for (unsigned int i = 0; i < numKeyframes; ++i) { const aiQuatKey& key = nodeChannel->mRotationKeys[i]; // mTime is measured in ticks, but GLTF time is measured in seconds, so convert. times[i] = static_cast(key.mTime / ticksPerSecond); values[(i * 4) + 0] = key.mValue.x; values[(i * 4) + 1] = key.mValue.y; values[(i * 4) + 2] = key.mValue.z; values[(i * 4) + 3] = key.mValue.w; } sampler.input = GetSamplerInputRef(asset, animId, buffer, times); sampler.output = ExportData(asset, animId, buffer, numKeyframes, &values[0], AttribType::VEC4, AttribType::VEC4, ComponentType_FLOAT); sampler.interpolation = Interpolation_LINEAR; } static void AddSampler(Ref& animRef, Ref& nodeRef, Animation::Sampler& sampler, AnimationPath path) { Animation::Channel channel; channel.sampler = static_cast(animRef->samplers.size()); channel.target.path = path; channel.target.node = nodeRef; animRef->channels.push_back(channel); animRef->samplers.push_back(sampler); } void glTF2Exporter::ExportAnimations() { Ref bufferRef = mAsset->buffers.Get(unsigned (0)); for (unsigned int i = 0; i < mScene->mNumAnimations; ++i) { const aiAnimation* anim = mScene->mAnimations[i]; const float ticksPerSecond = static_cast(anim->mTicksPerSecond); std::string nameAnim = "anim"; if (anim->mName.length > 0) { nameAnim = anim->mName.C_Str(); } Ref animRef = mAsset->animations.Create(nameAnim); animRef->name = nameAnim; for (unsigned int channelIndex = 0; channelIndex < anim->mNumChannels; ++channelIndex) { const aiNodeAnim* nodeChannel = anim->mChannels[channelIndex]; std::string name = nameAnim + "_" + ai_to_string(channelIndex); name = mAsset->FindUniqueID(name, "animation"); Ref animNode = mAsset->nodes.Get(nodeChannel->mNodeName.C_Str()); if (nodeChannel->mNumPositionKeys > 0) { Animation::Sampler translationSampler; ExtractTranslationSampler(*mAsset, name, bufferRef, nodeChannel, ticksPerSecond, translationSampler); AddSampler(animRef, animNode, translationSampler, AnimationPath_TRANSLATION); } if (nodeChannel->mNumRotationKeys > 0) { Animation::Sampler rotationSampler; ExtractRotationSampler(*mAsset, name, bufferRef, nodeChannel, ticksPerSecond, rotationSampler); AddSampler(animRef, animNode, rotationSampler, AnimationPath_ROTATION); } if (nodeChannel->mNumScalingKeys > 0) { Animation::Sampler scaleSampler; ExtractScaleSampler(*mAsset, name, bufferRef, nodeChannel, ticksPerSecond, scaleSampler); AddSampler(animRef, animNode, scaleSampler, AnimationPath_SCALE); } } // Assimp documentation staes this is not used (not implemented) // for (unsigned int channelIndex = 0; channelIndex < anim->mNumMeshChannels; ++channelIndex) { // const aiMeshAnim* meshChannel = anim->mMeshChannels[channelIndex]; // } } // End: for-loop mNumAnimations } #endif // ASSIMP_BUILD_NO_GLTF_EXPORTER #endif // ASSIMP_BUILD_NO_EXPORT