// This file is part of OpenCV project. // It is subject to the license terms in the LICENSE file found in the top-level directory // of this distribution and at http://opencv.org/license.html #include "test_precomp.hpp" namespace opencv_test { namespace { using namespace cv; // that was easier than using CV_ENUM() macro namespace { using namespace cv; struct CullingModeEnum { static const std::array vals; static const std::array svals; CullingModeEnum(TriangleCullingMode v = RASTERIZE_CULLING_NONE) : val(v) {} operator TriangleCullingMode() const { return val; } void PrintTo(std::ostream *os) const { int v = int(val); if (v >= 0 && v < (int)vals.size()) { *os << svals[v]; } else { *os << "UNKNOWN"; } } static ::testing::internal::ParamGenerator all() { return ::testing::Values(CullingModeEnum(vals[0]), CullingModeEnum(vals[1]), CullingModeEnum(vals[2])); } private: TriangleCullingMode val; }; const std::array CullingModeEnum::vals { RASTERIZE_CULLING_NONE, RASTERIZE_CULLING_CW, RASTERIZE_CULLING_CCW }; const std::array CullingModeEnum::svals { std::string("None"), std::string("CW"), std::string("CCW") }; static inline void PrintTo(const CullingModeEnum &t, std::ostream *os) { t.PrintTo(os); } } // that was easier than using CV_ENUM() macro namespace { using namespace cv; struct ShadingTypeEnum { static const std::array vals; static const std::array svals; ShadingTypeEnum(TriangleShadingType v = RASTERIZE_SHADING_WHITE) : val(v) {} operator TriangleShadingType() const { return val; } void PrintTo(std::ostream *os) const { int v = int(val); if (v >= 0 && v < (int)vals.size()) { *os << svals[v]; } else { *os << "UNKNOWN"; } } static ::testing::internal::ParamGenerator all() { return ::testing::Values(ShadingTypeEnum(vals[0]), ShadingTypeEnum(vals[1]), ShadingTypeEnum(vals[2])); } private: TriangleShadingType val; }; const std::array ShadingTypeEnum::vals { RASTERIZE_SHADING_WHITE, RASTERIZE_SHADING_FLAT, RASTERIZE_SHADING_SHADED }; const std::array ShadingTypeEnum::svals { std::string("White"), std::string("Flat"), std::string("Shaded") }; static inline void PrintTo(const ShadingTypeEnum &t, std::ostream *os) { t.PrintTo(os); } } enum class ModelType { Empty = 0, File = 1, Clipping = 2, Color = 3, Centered = 4 }; // that was easier than using CV_ENUM() macro namespace { using namespace cv; struct ModelTypeEnum { static const std::array vals; static const std::array svals; ModelTypeEnum(ModelType v = ModelType::Empty) : val(v) {} operator ModelType() const { return val; } void PrintTo(std::ostream *os) const { int v = int(val); if (v >= 0 && v < (int)vals.size()) { *os << svals[v]; } else { *os << "UNKNOWN"; } } static ::testing::internal::ParamGenerator all() { return ::testing::Values(ModelTypeEnum(vals[0]), ModelTypeEnum(vals[1]), ModelTypeEnum(vals[2]), ModelTypeEnum(vals[3]), ModelTypeEnum(vals[4])); } private: ModelType val; }; const std::array ModelTypeEnum::vals { ModelType::Empty, ModelType::File, ModelType::Clipping, ModelType::Color, ModelType::Centered }; const std::array ModelTypeEnum::svals { std::string("Empty"), std::string("File"), std::string("Clipping"), std::string("Color"), std::string("Centered") }; static inline void PrintTo(const ModelTypeEnum &t, std::ostream *os) { t.PrintTo(os); } } template std::string printEnum(T v) { std::ostringstream ss; v.PrintTo(&ss); return ss.str(); } static Matx44d lookAtMatrixCal(const Vec3d& position, const Vec3d& lookat, const Vec3d& upVector) { Vec3d w = cv::normalize(position - lookat); Vec3d u = cv::normalize(upVector.cross(w)); Vec3d v = w.cross(u); Matx44d res(u[0], u[1], u[2], 0, v[0], v[1], v[2], 0, w[0], w[1], w[2], 0, 0, 0, 0, 1.0); Matx44d translate(1.0, 0, 0, -position[0], 0, 1.0, 0, -position[1], 0, 0, 1.0, -position[2], 0, 0, 0, 1.0); res = res * translate; return res; } static void generateNormals(const std::vector& points, const std::vector>& indices, std::vector& normals) { std::vector> preNormals(points.size(), std::vector()); for (const auto& tri : indices) { Vec3f p0 = points[tri[0]]; Vec3f p1 = points[tri[1]]; Vec3f p2 = points[tri[2]]; Vec3f cross = cv::normalize((p1 - p0).cross(p2 - p0)); for (int i = 0; i < 3; i++) { preNormals[tri[i]].push_back(cross); } } normals.reserve(points.size()); for (const auto& pn : preNormals) { Vec3f sum { }; for (const auto& n : pn) { sum += n; } normals.push_back(cv::normalize(sum)); } } // load model once and keep it in static memory static void getModelOnce(const std::string& objectPath, std::vector& vertices, std::vector& indices, std::vector& colors) { static bool load = false; static std::vector vert, col; static std::vector ind; if (!load) { std::vector> indvec; // using per-vertex normals as colors loadMesh(objectPath, vert, indvec); generateNormals(vert, indvec, col); for (const auto &vec : indvec) { ind.push_back({vec[0], vec[1], vec[2]}); } for (auto &color : col) { color = Vec3f(abs(color[0]), abs(color[1]), abs(color[2])); } load = true; } vertices = vert; colors = col; indices = ind; } class ModelData { public: ModelData(ModelType type = ModelType::Empty) { switch (type) { case ModelType::Empty: { position = Vec3d(0.0, 0.0, 0.0); lookat = Vec3d(0.0, 0.0, 0.0); upVector = Vec3d(0.0, 1.0, 0.0); fovy = 45.0; vertices = std::vector(4, {2.0f, 0, -2.0f}); colors = std::vector(4, {0, 0, 1.0f}); indices = { }; } break; case ModelType::File: { string objectPath = findDataFile("viz/dragon.ply"); position = Vec3d( 1.9, 0.4, 1.3); lookat = Vec3d( 0.0, 0.0, 0.0); upVector = Vec3d( 0.0, 1.0, 0.0); fovy = 45.0; getModelOnce(objectPath, vertices, indices, colors); } break; case ModelType::Clipping: { position = Vec3d(0.0, 0.0, 5.0); lookat = Vec3d(0.0, 0.0, 0.0); upVector = Vec3d(0.0, 1.0, 0.0); fovy = 45.0; vertices = { { 2.0, 0.0, -2.0}, { 0.0, -6.0, -2.0}, {-2.0, 0.0, -2.0}, { 3.5, -1.0, -5.0}, { 2.5, -2.5, -5.0}, {-1.0, 1.0, -5.0}, {-6.5, -1.0, -3.0}, {-2.5, -2.0, -3.0}, { 1.0, 1.0, -5.0}, }; indices = { {0, 1, 2}, {3, 4, 5}, {6, 7, 8} }; Vec3f col1(217.0, 238.0, 185.0); Vec3f col2(185.0, 217.0, 238.0); Vec3f col3(150.0, 10.0, 238.0); col1 *= (1.f / 255.f); col2 *= (1.f / 255.f); col3 *= (1.f / 255.f); colors = { col1, col2, col3, col2, col3, col1, col3, col1, col2, }; } break; case ModelType::Centered: { position = Vec3d(0.0, 0.0, 5.0); lookat = Vec3d(0.0, 0.0, 0.0); upVector = Vec3d(0.0, 1.0, 0.0); fovy = 45.0; vertices = { { 2.0, 0.0, -2.0}, { 0.0, -2.0, -2.0}, {-2.0, 0.0, -2.0}, { 3.5, -1.0, -5.0}, { 2.5, -1.5, -5.0}, {-1.0, 0.5, -5.0}, }; indices = { {0, 1, 2}, {3, 4, 5} }; Vec3f col1(217.0, 238.0, 185.0); Vec3f col2(185.0, 217.0, 238.0); col1 *= (1.f / 255.f); col2 *= (1.f / 255.f); colors = { col1, col2, col1, col2, col1, col2, }; } break; case ModelType::Color: { position = Vec3d(0.0, 0.0, 5.0); lookat = Vec3d(0.0, 0.0, 0.0); upVector = Vec3d(0.0, 1.0, 0.0); fovy = 60.0; vertices = { { 2.0, 0.0, -2.0}, { 0.0, 2.0, -3.0}, {-2.0, 0.0, -2.0}, { 0.0, -2.0, 1.0}, }; indices = { {0, 1, 2}, {0, 2, 3} }; colors = { { 0.0f, 0.0f, 1.0f}, { 0.0f, 1.0f, 0.0f}, { 1.0f, 0.0f, 0.0f}, { 0.0f, 1.0f, 0.0f}, }; } break; default: CV_Error(Error::StsBadArg, "Unknown model type"); break; } } Vec3d position; Vec3d lookat; Vec3d upVector; double fovy; std::vector vertices; std::vector indices; std::vector colors; }; void compareDepth(const cv::Mat& gt, const cv::Mat& mat, cv::Mat& diff, double zFar, double scale, double maskThreshold, double normInfThreshold, double normL2Threshold) { ASSERT_EQ(CV_16UC1, gt.type()); ASSERT_EQ(CV_16UC1, mat.type()); ASSERT_EQ(gt.size(), mat.size()); Mat gtMask = gt < zFar*scale; Mat matMask = mat < zFar*scale; Mat diffMask = gtMask != matMask; int nzDepthDiff = cv::countNonZero(diffMask); EXPECT_LE(nzDepthDiff, maskThreshold); Mat jointMask = gtMask & matMask; int nzJointMask = cv::countNonZero(jointMask); double normInfDepth = cv::norm(gt, mat, cv::NORM_INF, jointMask); EXPECT_LE(normInfDepth, normInfThreshold); double normL2Depth = nzJointMask ? (cv::norm(gt, mat, cv::NORM_L2, jointMask) / nzJointMask) : 0; EXPECT_LE(normL2Depth, normL2Threshold); // add --test_debug to output differences if (debugLevel > 0) { std::cout << "nzDepthDiff: " << nzDepthDiff << " vs " << maskThreshold << std::endl; std::cout << "normInfDepth: " << normInfDepth << " vs " << normInfThreshold << std::endl; std::cout << "normL2Depth: " << normL2Depth << " vs " << normL2Threshold << std::endl; } diff = (gt - mat) + (1 << 15); } void compareRGB(const cv::Mat& gt, const cv::Mat& mat, cv::Mat& diff, double normInfThreshold, double normL2Threshold) { ASSERT_EQ(CV_32FC3, gt.type()); ASSERT_EQ(CV_32FC3, mat.type()); ASSERT_EQ(gt.size(), mat.size()); double normInfRgb = cv::norm(gt, mat, cv::NORM_INF); EXPECT_LE(normInfRgb, normInfThreshold); double normL2Rgb = cv::norm(gt, mat, cv::NORM_L2) / gt.total(); EXPECT_LE(normL2Rgb, normL2Threshold); // add --test_debug to output differences if (debugLevel > 0) { std::cout << "normInfRgb: " << normInfRgb << " vs " << normInfThreshold << std::endl; std::cout << "normL2Rgb: " << normL2Rgb << " vs " << normL2Threshold << std::endl; } diff = (gt - mat) * 0.5 + 0.5; } struct RenderTestThresholds { RenderTestThresholds( double _rgbInfThreshold, double _rgbL2Threshold, double _depthMaskThreshold, double _depthInfThreshold, double _depthL2Threshold) : rgbInfThreshold(_rgbInfThreshold), rgbL2Threshold(_rgbL2Threshold), depthMaskThreshold(_depthMaskThreshold), depthInfThreshold(_depthInfThreshold), depthL2Threshold(_depthL2Threshold) { } double rgbInfThreshold; double rgbL2Threshold; double depthMaskThreshold; double depthInfThreshold; double depthL2Threshold; }; // resolution, shading type, culling mode, model type, float type, index type typedef std::tuple, ShadingTypeEnum, CullingModeEnum, ModelTypeEnum, MatDepth, MatDepth> RenderTestParamType; class RenderingTest : public ::testing::TestWithParam { protected: void SetUp() override { params = GetParam(); auto wh = std::get<0>(params); width = std::get<0>(wh); height = std::get<1>(wh); shadingType = std::get<1>(params); cullingMode = std::get<2>(params); modelType = std::get<3>(params); modelData = ModelData(modelType); ftype = std::get<4>(params); itype = std::get<5>(params); zNear = 0.1, zFar = 50.0; depthScale = 1000.0; depth_buf = Mat(height, width, ftype, zFar); color_buf = Mat(height, width, CV_MAKETYPE(ftype, 3), Scalar::all(0)); cameraPose = lookAtMatrixCal(modelData.position, modelData.lookat, modelData.upVector); fovYradians = modelData.fovy * (CV_PI / 180.0); verts = Mat(modelData.vertices); verts.convertTo(verts, ftype); if (shadingType != RASTERIZE_SHADING_WHITE) { // let vertices be in BGR format to avoid later color conversions // mixChannels() does not support in-place operation colors = Mat(modelData.colors); colors.convertTo(colors, ftype); cv::mixChannels(colors.clone(), colors, {0, 2, 1, 1, 2, 0}); } indices = Mat(modelData.indices); if (itype != CV_32S) { indices.convertTo(indices, itype); } settings = TriangleRasterizeSettings().setCullingMode(cullingMode).setShadingType(shadingType); triangleRasterize(verts, indices, colors, color_buf, depth_buf, cameraPose, fovYradians, zNear, zFar, settings); } public: RenderTestParamType params; int width, height; double zNear, zFar, depthScale; Mat depth_buf, color_buf; Mat verts, colors, indices; Matx44d cameraPose; double fovYradians; TriangleRasterizeSettings settings; ModelData modelData; ModelTypeEnum modelType; ShadingTypeEnum shadingType; CullingModeEnum cullingMode; int ftype, itype; }; // depth-only or RGB-only rendering should produce the same result as usual rendering TEST_P(RenderingTest, noArrays) { Mat depthOnly(height, width, ftype, zFar); Mat colorOnly(height, width, CV_MAKETYPE(ftype, 3), Scalar::all(0)); triangleRasterizeDepth(verts, indices, depthOnly, cameraPose, fovYradians, zNear, zFar, settings); triangleRasterizeColor(verts, indices, colors, colorOnly, cameraPose, fovYradians, zNear, zFar, settings); Mat rgbDiff, depthDiff; compareRGB(color_buf, colorOnly, rgbDiff, 0, 0); depth_buf.convertTo(depth_buf, CV_16U, depthScale); depthOnly.convertTo(depthOnly, CV_16U, depthScale); compareDepth(depth_buf, depthOnly, depthDiff, zFar, depthScale, 0, 0, 0); // add --test_debug to output resulting images if (debugLevel > 0) { std::string modelName = printEnum(modelType); std::string shadingName = printEnum(shadingType); std::string cullingName = printEnum(cullingMode); std::string suffix = cv::format("%s_%dx%d_Cull%s", modelName.c_str(), width, height, cullingName.c_str()); std::string outColorPath = "noarray_color_image_" + suffix + "_" + shadingName + ".png"; std::string outDepthPath = "noarray_depth_image_" + suffix + "_" + shadingName + ".png"; imwrite(outColorPath, color_buf * 255.f); imwrite(outDepthPath, depth_buf); imwrite("diff_" + outColorPath, rgbDiff * 255.f); imwrite("diff_" + outDepthPath, depthDiff); } } // passing the same parameters in float should give the same result TEST_P(RenderingTest, floatParams) { Mat depth_buf2(height, width, ftype, zFar); Mat color_buf2(height, width, CV_MAKETYPE(ftype, 3), Scalar::all(0)); // cameraPose can also be float, checking it triangleRasterize(verts, indices, colors, color_buf2, depth_buf2, Matx44f(cameraPose), (float)fovYradians, (float)zNear, (float)zFar, settings); RenderTestThresholds thr(0, 0, 0, 0, 0); switch (modelType) { case ModelType::Empty: break; case ModelType::Color: break; case ModelType::Clipping: if (width == 320 && height == 240 && shadingType == RASTERIZE_SHADING_FLAT && cullingMode == RASTERIZE_CULLING_CW) { thr.depthInfThreshold = 1; thr.depthL2Threshold = 0.00127; } else if (width == 320 && height == 240 && shadingType == RASTERIZE_SHADING_SHADED && cullingMode == RASTERIZE_CULLING_NONE) { thr.rgbInfThreshold = 3e-7; thr.rgbL2Threshold = 1.86e-10; thr.depthInfThreshold = 1; thr.depthL2Threshold = 0.000406; } else if (width == 256 && height == 256 && shadingType == RASTERIZE_SHADING_SHADED && cullingMode == RASTERIZE_CULLING_CW) { thr.rgbInfThreshold = 2.39e-07; thr.rgbL2Threshold = 1.86e-10; thr.depthInfThreshold = 1; thr.depthL2Threshold = 0.0016; } else if (width == 256 && height == 256 && shadingType == RASTERIZE_SHADING_FLAT && cullingMode == RASTERIZE_CULLING_CCW) { thr.rgbInfThreshold = 0.934; thr.rgbL2Threshold = 0.000102; thr.depthMaskThreshold = 21; } else if (width == 640 && height == 480 && shadingType == RASTERIZE_SHADING_WHITE && cullingMode == RASTERIZE_CULLING_NONE) { thr.rgbL2Threshold = 1; thr.depthInfThreshold = 1; thr.depthL2Threshold = 0.000248; } else if (width == 700 && height == 700 && shadingType == RASTERIZE_SHADING_FLAT && cullingMode == RASTERIZE_CULLING_CCW) { thr.rgbInfThreshold = 0.934; thr.rgbL2Threshold = 3.18e-5; thr.depthMaskThreshold = 114; } break; case ModelType::File: thr.depthInfThreshold = 1; if (width == 320 && height == 240 && shadingType == RASTERIZE_SHADING_SHADED && cullingMode == RASTERIZE_CULLING_CCW) { thr.rgbInfThreshold = 0.000229; thr.rgbL2Threshold = 6.37e-09; thr.depthL2Threshold = 0.000427; } else if (width == 700 && height == 700 && shadingType == RASTERIZE_SHADING_SHADED && cullingMode == RASTERIZE_CULLING_CW) { thr.rgbInfThreshold = 0.000277; thr.rgbL2Threshold = 1.8e-09; thr.depthL2Threshold = 0.000124; } else if (width == 700 && height == 700 && shadingType == RASTERIZE_SHADING_WHITE && cullingMode == RASTERIZE_CULLING_NONE) { thr.depthL2Threshold = 0.000124; } break; case ModelType::Centered: if (shadingType == RASTERIZE_SHADING_SHADED && cullingMode != RASTERIZE_CULLING_CW) { thr.rgbInfThreshold = 3.58e-07; thr.rgbL2Threshold = 1.51e-10; } break; } Mat rgbDiff, depthDiff; compareRGB(color_buf, color_buf2, rgbDiff, thr.rgbInfThreshold, thr.rgbL2Threshold); depth_buf.convertTo(depth_buf, CV_16U, depthScale); depth_buf2.convertTo(depth_buf2, CV_16U, depthScale); compareDepth(depth_buf, depth_buf2, depthDiff, zFar, depthScale, thr.depthMaskThreshold, thr.depthInfThreshold, thr.depthL2Threshold); // add --test_debug to output resulting images if (debugLevel > 0) { std::string modelName = printEnum(modelType); std::string shadingName = printEnum(shadingType); std::string cullingName = printEnum(cullingMode); std::string suffix = cv::format("%s_%dx%d_Cull%s", modelName.c_str(), width, height, cullingName.c_str()); std::string outColorPath = "float_color_image_" + suffix + "_" + shadingName + ".png"; std::string outDepthPath = "float_depth_image_" + suffix + "_" + shadingName + ".png"; imwrite(outColorPath, color_buf * 255.f); imwrite(outDepthPath, depth_buf); imwrite("diff_" + outColorPath, rgbDiff * 255.f); imwrite("diff_" + outDepthPath, depthDiff); } } // some culling options produce the same pictures, let's join them TriangleCullingMode findSameCulling(ModelType modelType, TriangleShadingType shadingType, TriangleCullingMode cullingMode, bool forRgb) { TriangleCullingMode sameCullingMode = cullingMode; if ((modelType == ModelType::Centered && cullingMode == RASTERIZE_CULLING_CCW) || (modelType == ModelType::Color && cullingMode == RASTERIZE_CULLING_CW) || (modelType == ModelType::File && shadingType == RASTERIZE_SHADING_WHITE && forRgb) || (modelType == ModelType::File && cullingMode == RASTERIZE_CULLING_CW)) { sameCullingMode = RASTERIZE_CULLING_NONE; } return sameCullingMode; } // compare rendering results to the ones produced by samples/opengl/opengl_testdata_generator app TEST_P(RenderingTest, accuracy) { depth_buf.convertTo(depth_buf, CV_16U, depthScale); if (modelType == ModelType::Empty || (modelType == ModelType::Centered && cullingMode == RASTERIZE_CULLING_CW) || (modelType == ModelType::Color && cullingMode == RASTERIZE_CULLING_CCW)) { // empty image case EXPECT_EQ(0, cv::norm(color_buf, NORM_INF)); Mat depthDiff; absdiff(depth_buf, Scalar(zFar * depthScale), depthDiff); EXPECT_EQ(0, cv::norm(depthDiff, cv::NORM_INF)); } else { RenderTestThresholds thr(0, 0, 0, 0, 0); switch (modelType) { case ModelType::Centered: if (shadingType == RASTERIZE_SHADING_SHADED) { thr.rgbInfThreshold = 0.00218; thr.rgbL2Threshold = 2.85e-06; } break; case ModelType::Clipping: if (width == 320 && height == 240 && shadingType == RASTERIZE_SHADING_FLAT && cullingMode == RASTERIZE_CULLING_CW) { thr.depthInfThreshold = 1; thr.depthL2Threshold = 0.00163; } else if (width == 320 && height == 240 && shadingType == RASTERIZE_SHADING_SHADED && cullingMode == RASTERIZE_CULLING_NONE) { thr.rgbInfThreshold = 0.934; thr.rgbL2Threshold = 8.03E-05; thr.depthMaskThreshold = 23; thr.depthInfThreshold = 1; thr.depthL2Threshold = 0.000555; } else if (width == 256 && height == 256 && shadingType == RASTERIZE_SHADING_SHADED && cullingMode == RASTERIZE_CULLING_CW) { thr.rgbInfThreshold = 0.0022; thr.rgbL2Threshold = 2.54E-06; thr.depthInfThreshold = 1; thr.depthL2Threshold = 0.00175; } else if (width == 256 && height == 256 && shadingType == RASTERIZE_SHADING_FLAT && cullingMode == RASTERIZE_CULLING_CCW) { thr.rgbInfThreshold = 0.934; thr.rgbL2Threshold = 0.000102; thr.depthMaskThreshold = 21; } else if (width == 640 && height == 480 && shadingType == RASTERIZE_SHADING_WHITE && cullingMode == RASTERIZE_CULLING_NONE) { thr.rgbInfThreshold = 1; thr.rgbL2Threshold = 3.95E-05; thr.depthMaskThreshold = 49; thr.depthInfThreshold = 1; thr.depthL2Threshold = 0.000269; } else if (width == 700 && height == 700 && shadingType == RASTERIZE_SHADING_FLAT && cullingMode == RASTERIZE_CULLING_CCW) { thr.rgbInfThreshold = 0.934; thr.rgbL2Threshold = 3.27e-5; thr.depthMaskThreshold = 121; } break; case ModelType::Color: thr.depthInfThreshold = 1; if (width == 320 && height == 240) { thr.depthL2Threshold = 0.00103; } else if (width == 256 && height == 256) { thr.depthL2Threshold = 0.000785; } if (shadingType == RASTERIZE_SHADING_SHADED) { thr.rgbInfThreshold = 0.0022; thr.rgbL2Threshold = 3.13e-06; } break; case ModelType::File: if (width == 320 && height == 240 && shadingType == RASTERIZE_SHADING_SHADED && cullingMode == RASTERIZE_CULLING_CCW) { thr.rgbInfThreshold = 0.836; thr.rgbL2Threshold = 2.08e-05; thr.depthMaskThreshold = 1; thr.depthInfThreshold = 99; thr.depthL2Threshold = 0.00544; } else if (width == 700 && height == 700 && shadingType == RASTERIZE_SHADING_SHADED && cullingMode == RASTERIZE_CULLING_CW) { thr.rgbInfThreshold = 0.973; thr.rgbL2Threshold = 5.2e-06; thr.depthMaskThreshold = 4; thr.depthInfThreshold = 258; thr.depthL2Threshold = 0.00228; } else if (width == 700 && height == 700 && shadingType == RASTERIZE_SHADING_WHITE && cullingMode == RASTERIZE_CULLING_NONE) { thr.rgbInfThreshold = 1; thr.rgbL2Threshold = 7.07e-06; thr.depthMaskThreshold = 4; thr.depthInfThreshold = 258; thr.depthL2Threshold = 0.00228; } break; default: break; } CullingModeEnum cullingModeRgb = findSameCulling(modelType, shadingType, cullingMode, true); CullingModeEnum cullingModeDepth = findSameCulling(modelType, shadingType, cullingMode, false); std::string modelName = printEnum(modelType); std::string shadingName = printEnum(shadingType); std::string cullingName = printEnum(cullingMode); std::string cullingRgbName = printEnum(cullingModeRgb); std::string cullingDepthName = printEnum(cullingModeDepth); std::string path = findDataDirectory("rendering"); std::string suffix = cv::format("%s_%dx%d_Cull%s", modelName.c_str(), width, height, cullingName.c_str()); std::string suffixRgb = cv::format("%s_%dx%d_Cull%s", modelName.c_str(), width, height, cullingRgbName.c_str()); std::string suffixDepth = cv::format("%s_%dx%d_Cull%s", modelName.c_str(), width, height, cullingDepthName.c_str()); std::string gtPathColor = path + "/example_image_" + suffixRgb + "_" + shadingName + ".png"; std::string gtPathDepth = path + "/depth_image_" + suffixDepth + ".png"; Mat rgbDiff, depthDiff; Mat groundTruthColor = imread(gtPathColor); groundTruthColor.convertTo(groundTruthColor, CV_32F, (1.f / 255.f)); compareRGB(groundTruthColor, color_buf, rgbDiff, thr.rgbInfThreshold, thr.rgbL2Threshold); Mat groundTruthDepth = imread(gtPathDepth, cv::IMREAD_GRAYSCALE | cv::IMREAD_ANYDEPTH); compareDepth(groundTruthDepth, depth_buf, depthDiff, zFar, depthScale, thr.depthMaskThreshold, thr.depthInfThreshold, thr.depthL2Threshold); // add --test_debug to output resulting images if (debugLevel > 0) { std::string outColorPath = "color_image_" + suffix + "_" + shadingName + ".png"; std::string outDepthPath = "depth_image_" + suffix + "_" + shadingName + ".png"; imwrite(outColorPath, color_buf * 255.f); imwrite(outDepthPath, depth_buf); imwrite("diff_" + outColorPath, rgbDiff * 255.f); imwrite("diff_" + outDepthPath, depthDiff); } } } // drawing model as a whole or as two halves should give the same result TEST_P(RenderingTest, keepDrawnData) { if (modelType != ModelType::Empty) { Mat depth_buf2(height, width, ftype, zFar); Mat color_buf2(height, width, CV_MAKETYPE(ftype, 3), Scalar::all(0)); Mat idx1, idx2; int nTriangles = (int)indices.total(); idx1 = indices.reshape(3, 1)(Range::all(), Range(0, nTriangles / 2)); idx2 = indices.reshape(3, 1)(Range::all(), Range(nTriangles / 2, nTriangles)); triangleRasterize(verts, idx1, colors, color_buf2, depth_buf2, cameraPose, fovYradians, zNear, zFar, settings); triangleRasterize(verts, idx2, colors, color_buf2, depth_buf2, cameraPose, fovYradians, zNear, zFar, settings); Mat rgbDiff, depthDiff; compareRGB(color_buf, color_buf2, rgbDiff, 0, 0); depth_buf.convertTo(depth_buf, CV_16U, depthScale); depth_buf2.convertTo(depth_buf2, CV_16U, depthScale); compareDepth(depth_buf, depth_buf2, depthDiff, zFar, depthScale, 0, 0, 0); } } TEST_P(RenderingTest, glCompatibleDepth) { Mat depth_buf2(height, width, ftype, 1.0); triangleRasterizeDepth(verts, indices, depth_buf2, cameraPose, fovYradians, zNear, zFar, settings.setGlCompatibleMode(RASTERIZE_COMPAT_INVDEPTH)); Mat convertedDepth(height, width, ftype); // map from [0, 1] to [zNear, zFar] double scaleNear = (1.0 / zNear); double scaleFar = (1.0 / zFar); for (int y = 0; y < height; y++) { for (int x = 0; x < width; x++) { double z = (double)depth_buf2.at(y, x); convertedDepth.at(y, x) = (float)(1.0 / ((1.0 - z) * scaleNear + z * scaleFar )); } } double normL2Diff = cv::norm(depth_buf, convertedDepth, cv::NORM_L2) / (height * width); const double normL2Threshold = 5.53e-10; EXPECT_LE(normL2Diff, normL2Threshold); // add --test_debug to output differences if (debugLevel > 0) { std::cout << "normL2Diff: " << normL2Diff << " vs " << normL2Threshold << std::endl; } } INSTANTIATE_TEST_CASE_P(Rendering, RenderingTest, ::testing::Values( RenderTestParamType { std::make_tuple(320, 240), RASTERIZE_SHADING_SHADED, RASTERIZE_CULLING_NONE, ModelType::Centered, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(256, 256), RASTERIZE_SHADING_SHADED, RASTERIZE_CULLING_NONE, ModelType::Centered, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(256, 256), RASTERIZE_SHADING_WHITE, RASTERIZE_CULLING_NONE, ModelType::Centered, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(640, 480), RASTERIZE_SHADING_FLAT, RASTERIZE_CULLING_NONE, ModelType::Centered, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(320, 240), RASTERIZE_SHADING_FLAT, RASTERIZE_CULLING_CW, ModelType::Color, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(320, 240), RASTERIZE_SHADING_SHADED, RASTERIZE_CULLING_NONE, ModelType::Color, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(256, 256), RASTERIZE_SHADING_SHADED, RASTERIZE_CULLING_NONE, ModelType::Color, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(256, 256), RASTERIZE_SHADING_WHITE, RASTERIZE_CULLING_NONE, ModelType::Color, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(320, 240), RASTERIZE_SHADING_FLAT, RASTERIZE_CULLING_CW, ModelType::Clipping, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(320, 240), RASTERIZE_SHADING_SHADED, RASTERIZE_CULLING_NONE, ModelType::Clipping, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(256, 256), RASTERIZE_SHADING_FLAT, RASTERIZE_CULLING_CCW, ModelType::Clipping, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(256, 256), RASTERIZE_SHADING_SHADED, RASTERIZE_CULLING_CW, ModelType::Clipping, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(640, 480), RASTERIZE_SHADING_WHITE, RASTERIZE_CULLING_NONE, ModelType::Clipping, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(700, 700), RASTERIZE_SHADING_FLAT, RASTERIZE_CULLING_CCW, ModelType::Clipping, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(320, 240), RASTERIZE_SHADING_SHADED, RASTERIZE_CULLING_CCW, ModelType::File, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(700, 700), RASTERIZE_SHADING_SHADED, RASTERIZE_CULLING_CW, ModelType::File, CV_32F, CV_32S }, RenderTestParamType { std::make_tuple(700, 700), RASTERIZE_SHADING_WHITE, RASTERIZE_CULLING_NONE, ModelType::File, CV_32F, CV_32S } )); } // namespace :: } // namespace opencv_test