Merge remote-tracking branch 'upstream/3.4' into merge-3.4

modules/core/include/opencv2/core/fast_math.hpp

@@ -102,7 +102,7 @@
     #define CV_INLINE_ROUND_DBL(value) \
         int out; \
         double temp; \
-        __asm__( "fctiw %[temp],%[in]\n\tmffprwz %[out],%[temp]\n\t" : [out] "=r" (out), [temp] "=d" (temp) : [in] "d" ((double)(value)) : ); \
+        __asm__( "fctiw %[temp],%[in]\n\tmfvsrwz %[out],%[temp]\n\t" : [out] "=r" (out), [temp] "=d" (temp) : [in] "d" ((double)(value)) : ); \
         return out;
 
     // FP32 also works with FP64 routine above

modules/core/include/opencv2/core/hal/intrin_avx.hpp

@@ -1231,19 +1231,16 @@ inline int v_signmask(const v_int16x16& a)
 inline int v_signmask(const v_uint16x16& a)
 { return v_signmask(v_reinterpret_as_s16(a)); }
 
-inline int v_signmask(const v_int32x8& a)
-{
-    v_int16x16 a16 = v_pack(a, a);
-    return v_signmask(v_pack(a16, a16)) & 0xFF;
-}
-inline int v_signmask(const v_uint32x8& a)
-{ return v_signmask(v_reinterpret_as_s32(a)); }
-
 inline int v_signmask(const v_float32x8& a)
 { return _mm256_movemask_ps(a.val); }
 inline int v_signmask(const v_float64x4& a)
 { return _mm256_movemask_pd(a.val); }
 
+inline int v_signmask(const v_int32x8& a)
+{ return v_signmask(v_reinterpret_as_f32(a)); }
+inline int v_signmask(const v_uint32x8& a)
+{ return v_signmask(v_reinterpret_as_f32(a)); }
+
 inline int v_scan_forward(const v_int8x32& a) { return trailingZeros32(v_signmask(v_reinterpret_as_s8(a))); }
 inline int v_scan_forward(const v_uint8x32& a) { return trailingZeros32(v_signmask(v_reinterpret_as_s8(a))); }
 inline int v_scan_forward(const v_int16x16& a) { return trailingZeros32(v_signmask(v_reinterpret_as_s8(a))) / 2; }
@@ -1270,10 +1267,10 @@ inline int v_scan_forward(const v_float64x4& a) { return trailingZeros32(v_signm
 
 OPENCV_HAL_IMPL_AVX_CHECK(v_uint8x32,  OPENCV_HAL_1ST, -1)
 OPENCV_HAL_IMPL_AVX_CHECK(v_int8x32,   OPENCV_HAL_1ST, -1)
-OPENCV_HAL_IMPL_AVX_CHECK(v_uint16x16, OPENCV_HAL_AND, (int)0xaaaa)
-OPENCV_HAL_IMPL_AVX_CHECK(v_int16x16,  OPENCV_HAL_AND, (int)0xaaaa)
-OPENCV_HAL_IMPL_AVX_CHECK(v_uint32x8,  OPENCV_HAL_AND, (int)0x8888)
-OPENCV_HAL_IMPL_AVX_CHECK(v_int32x8,   OPENCV_HAL_AND, (int)0x8888)
+OPENCV_HAL_IMPL_AVX_CHECK(v_uint16x16, OPENCV_HAL_AND, (int)0xaaaaaaaa)
+OPENCV_HAL_IMPL_AVX_CHECK(v_int16x16,  OPENCV_HAL_AND, (int)0xaaaaaaaa)
+OPENCV_HAL_IMPL_AVX_CHECK(v_uint32x8,  OPENCV_HAL_AND, (int)0x88888888)
+OPENCV_HAL_IMPL_AVX_CHECK(v_int32x8,   OPENCV_HAL_AND, (int)0x88888888)
 
 #define OPENCV_HAL_IMPL_AVX_CHECK_FLT(_Tpvec, allmask) \
     inline bool v_check_all(const _Tpvec& a)           \

modules/core/include/opencv2/core/hal/intrin_vsx.hpp

@@ -764,10 +764,10 @@ inline scalartype v_reduce_##suffix(const _Tpvec& a)
     rs = func(rs, vec_sld(rs, rs, 2));                                             \
     return vec_extract(func(rs, vec_sld(rs, rs, 1)), 0);                           \
 }
-OPENCV_HAL_IMPL_VSX_REDUCE_OP_8(v_uint8x16, vec_uchar16, uchar, max, vec_max)
-OPENCV_HAL_IMPL_VSX_REDUCE_OP_8(v_uint8x16, vec_uchar16, uchar, min, vec_min)
-OPENCV_HAL_IMPL_VSX_REDUCE_OP_8(v_int8x16, vec_char16, schar, max, vec_max)
-OPENCV_HAL_IMPL_VSX_REDUCE_OP_8(v_int8x16, vec_char16, schar, min, vec_min)
+OPENCV_HAL_IMPL_VSX_REDUCE_OP_16(v_uint8x16, vec_uchar16, uchar, max, vec_max)
+OPENCV_HAL_IMPL_VSX_REDUCE_OP_16(v_uint8x16, vec_uchar16, uchar, min, vec_min)
+OPENCV_HAL_IMPL_VSX_REDUCE_OP_16(v_int8x16, vec_char16, schar, max, vec_max)
+OPENCV_HAL_IMPL_VSX_REDUCE_OP_16(v_int8x16, vec_char16, schar, min, vec_min)
 
 inline v_float32x4 v_reduce_sum4(const v_float32x4& a, const v_float32x4& b,
                                  const v_float32x4& c, const v_float32x4& d)

modules/core/include/opencv2/core/vsx_utils.hpp

@@ -363,10 +363,12 @@ VSX_FINLINE(Tvec) vec_popcntu(const Tvec2& a)  \
 VSX_IMPL_POPCNTU(vec_uchar16, vec_char16, vec_uchar16_c);
 VSX_IMPL_POPCNTU(vec_ushort8, vec_short8, vec_ushort8_c);
 VSX_IMPL_POPCNTU(vec_uint4,   vec_int4,   vec_uint4_c);
+VSX_IMPL_POPCNTU(vec_udword2, vec_dword2, vec_udword2_c);
 // redirect unsigned types
 VSX_REDIRECT_1RG(vec_uchar16, vec_uchar16, vec_popcntu, vec_popcnt)
 VSX_REDIRECT_1RG(vec_ushort8, vec_ushort8, vec_popcntu, vec_popcnt)
 VSX_REDIRECT_1RG(vec_uint4,   vec_uint4,   vec_popcntu, vec_popcnt)
+VSX_REDIRECT_1RG(vec_udword2, vec_udword2, vec_popcntu, vec_popcnt)
 
 // converts between single and double precision
 VSX_REDIRECT_1RG(vec_float4,  vec_double2, vec_cvfo, __builtin_vsx_xvcvdpsp)

modules/core/test/test_intrin_utils.hpp

@@ -804,11 +804,14 @@ template<typename R> struct TheTest
         all1s;
         all1s.ui = (uint_type)-1;
         LaneType mask_one = all1s.l;
+        dataB[R::nlanes - 1] = mask_one;
+        R l = dataB;
         dataB[1] = mask_one;
         dataB[R::nlanes / 2] = mask_one;
-        dataB[R::nlanes - 1] = mask_one;
         dataC *= (LaneType)-1;
         R a = dataA, b = dataB, c = dataC, d = dataD, e = dataE;
+        dataC[R::nlanes - 1] = 0;
+        R nl = dataC;
 
         EXPECT_EQ(2, v_signmask(a));
 #if CV_SIMD_WIDTH <= 32
@@ -818,11 +821,12 @@ template<typename R> struct TheTest
         EXPECT_EQ(false, v_check_all(a));
         EXPECT_EQ(false, v_check_all(b));
         EXPECT_EQ(true, v_check_all(c));
+        EXPECT_EQ(false, v_check_all(nl));
 
         EXPECT_EQ(true, v_check_any(a));
         EXPECT_EQ(true, v_check_any(b));
         EXPECT_EQ(true, v_check_any(c));
-
+        EXPECT_EQ(true, v_check_any(l));
         R f = v_select(b, d, e);
         Data<R> resF = f;
         for (int i = 0; i < R::nlanes; ++i)

modules/dnn/src/layers/fully_connected_layer.cpp

@@ -449,7 +449,7 @@ public:
 
         InferenceEngine::Builder::Layer l = ieLayer;
         addConstantData("weights", wrapToInfEngineBlob(blobs[0], {(size_t)blobs[0].size[0], (size_t)blobs[0].size[1], 1, 1}, InferenceEngine::Layout::OIHW), l);
-        if (blobs.size() > 1)
+        if (bias)
             addConstantData("biases", wrapToInfEngineBlob(blobs[1], {(size_t)outNum}, InferenceEngine::Layout::C), l);
 
         return Ptr<BackendNode>(new InfEngineBackendNode(l));

modules/dnn/src/tensorflow/tf_graph_simplifier.cpp

@@ -787,7 +787,7 @@ void RemoveIdentityOps(tensorflow::GraphDef& net)
         const tensorflow::NodeDef &layer = net.node(li);
         String type = layer.op();
 
-        if (type == "Identity" || type == "Dropout") {
+        if (type == "Identity" || type == "Dropout" || type == "PlaceholderWithDefault") {
             identity_ops_idx.push_back(li);
             identity_ops[layer.name()] = layer.input(0);
         }
@@ -1031,7 +1031,7 @@ void removePhaseSwitches(tensorflow::GraphDef& net)
                 }
             }
             nodesToRemove.push_back(i);
-            if (node.op() == "Merge")
+            if (node.op() == "Merge" || node.op() == "Switch")
                 mergeOpSubgraphNodes.push(i);
         }
     }

modules/dnn/test/test_halide_layers.cpp

@@ -386,8 +386,11 @@ TEST_P(FullyConnected, Accuracy)
     bool hasBias = get<3>(GetParam());
     Backend backendId = get<0>(get<4>(GetParam()));
     Target targetId = get<1>(get<4>(GetParam()));
-    if (backendId == DNN_BACKEND_INFERENCE_ENGINE)
-        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE);
+    if (backendId == DNN_BACKEND_INFERENCE_ENGINE && (targetId == DNN_TARGET_OPENCL_FP16 ||
+       (targetId == DNN_TARGET_MYRIAD && getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X))) {
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16);
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X);
+    }
 
     Mat weights(outChannels, inChannels * inSize.height * inSize.width, CV_32F);
     randu(weights, -1.0f, 1.0f);

modules/dnn/test/test_tf_importer.cpp

@@ -729,6 +729,7 @@ TEST_P(Test_TensorFlow_layers, subpixel)
 TEST_P(Test_TensorFlow_layers, keras_mobilenet_head)
 {
     runTensorFlowNet("keras_mobilenet_head");
+    runTensorFlowNet("keras_learning_phase");
 }
 
 TEST_P(Test_TensorFlow_layers, resize_bilinear)

modules/features2d/src/fast.cpp

@@ -159,15 +159,12 @@ void FAST_t(InputArray _img, std::vector<KeyPoint>& keypoints, int threshold, bo
                             }
 
                             max0 = K16 < v_max(max0, max1);
-                            int m = -v_reduce_sum(v_reinterpret_as_s8(max0));
-                            uchar mflag[16];
-                            v_store(mflag, max0);
+                            unsigned int m = v_signmask(v_reinterpret_as_s8(max0));
 
-                            for( k = 0; m > 0 && k < 16; k++ )
+                            for( k = 0; m > 0 && k < 16; k++, m >>= 1 )
                             {
-                                if(mflag[k])
+                                if( m & 1 )
                                 {
-                                    --m;
                                     cornerpos[ncorners++] = j+k;
                                     if(nonmax_suppression)
                                     {

modules/imgproc/include/opencv2/imgproc.hpp

@@ -4208,7 +4208,8 @@ enum ColormapTypes
     COLORMAP_VIRIDIS = 16, //!< ![viridis](pics/colormaps/colorscale_viridis.jpg)
     COLORMAP_CIVIDIS = 17, //!< ![cividis](pics/colormaps/colorscale_cividis.jpg)
     COLORMAP_TWILIGHT = 18, //!< ![twilight](pics/colormaps/colorscale_twilight.jpg)
-    COLORMAP_TWILIGHT_SHIFTED = 19 //!< ![twilight shifted](pics/colormaps/colorscale_twilight_shifted.jpg)
+    COLORMAP_TWILIGHT_SHIFTED = 19, //!< ![twilight shifted](pics/colormaps/colorscale_twilight_shifted.jpg)
+    COLORMAP_TURBO = 20 //!< ![turbo](pics/colormaps/colorscale_turbo.jpg)
 };
 
 /** @example samples/cpp/falsecolor.cpp

modules/imgproc/src/colormap.cpp

@@ -657,6 +657,36 @@ namespace colormap
         }
     };
 
+    // Equals the colormap "Turbo" proposed by Google.
+    // https://ai.googleblog.com/2019/08/turbo-improved-rainbow-colormap-for.html
+    // https://gist.github.com/mikhailov-work/6a308c20e494d9e0ccc29036b28faa7a
+    class Turbo : public ColorMap {
+    public:
+        Turbo() : ColorMap() {
+            init(256);
+        }
+        Turbo(int n) : ColorMap() {
+            init(n);
+        }
+
+        void init(int n) {
+            // define the basemap
+            static const float r[] = { 0.18995f,0.19483f,0.19956f,0.20415f,0.20860f,0.21291f,0.21708f,0.22111f,0.22500f,0.22875f,0.23236f,0.23582f,0.23915f,0.24234f,0.24539f,0.24830f,0.25107f,0.25369f,0.25618f,0.25853f,0.26074f,0.26280f,0.26473f,0.26652f,0.26816f,0.26967f,0.27103f,0.27226f,0.27334f,0.27429f,0.27509f,0.27576f,0.27628f,0.27667f,0.27691f,0.27701f,0.27698f,0.27680f,0.27648f,0.27603f,0.27543f,0.27469f,0.27381f,0.27273f,0.27106f,0.26878f,0.26592f,0.26252f,0.25862f,0.25425f,0.24946f,0.24427f,0.23874f,0.23288f,0.22676f,0.22039f,0.21382f,0.20708f,0.20021f,0.19326f,0.18625f,0.17923f,0.17223f,0.16529f,0.15844f,0.15173f,0.14519f,0.13886f,0.13278f,0.12698f,0.12151f,0.11639f,0.11167f,0.10738f,0.10357f,0.10026f,0.09750f,0.09532f,0.09377f,0.09287f,0.09267f,0.09320f,0.09451f,0.09662f,0.09958f,0.10342f,0.10815f,0.11374f,0.12014f,0.12733f,0.13526f,0.14391f,0.15323f,0.16319f,0.17377f,0.18491f,0.19659f,0.20877f,0.22142f,0.23449f,0.24797f,0.26180f,0.27597f,0.29042f,0.30513f,0.32006f,0.33517f,0.35043f,0.36581f,0.38127f,0.39678f,0.41229f,0.42778f,0.44321f,0.45854f,0.47375f,0.48879f,0.50362f,0.51822f,0.53255f,0.54658f,0.56026f,0.57357f,0.58646f,0.59891f,0.61088f,0.62233f,0.63323f,0.64362f,0.65394f,0.66428f,0.67462f,0.68494f,0.69525f,0.70553f,0.71577f,0.72596f,0.73610f,0.74617f,0.75617f,0.76608f,0.77591f,0.78563f,0.79524f,0.80473f,0.81410f,0.82333f,0.83241f,0.84133f,0.85010f,0.85868f,0.86709f,0.87530f,0.88331f,0.89112f,0.89870f,0.90605f,0.91317f,0.92004f,0.92666f,0.93301f,0.93909f,0.94489f,0.95039f,0.95560f,0.96049f,0.96507f,0.96931f,0.97323f,0.97679f,0.98000f,0.98289f,0.98549f,0.98781f,0.98986f,0.99163f,0.99314f,0.99438f,0.99535f,0.99607f,0.99654f,0.99675f,0.99672f,0.99644f,0.99593f,0.99517f,0.99419f,0.99297f,0.99153f,0.98987f,0.98799f,0.98590f,0.98360f,0.98108f,0.97837f,0.97545f,0.97234f,0.96904f,0.96555f,0.96187f,0.95801f,0.95398f,0.94977f,0.94538f,0.94084f,0.93612f,0.93125f,0.92623f,0.92105f,0.91572f,0.91024f,0.90463f,0.89888f,0.89298f,0.88691f,0.88066f,0.87422f,0.86760f,0.86079f,0.85380f,0.84662f,0.83926f,0.83172f,0.82399f,0.81608f,0.80799f,0.79971f,0.79125f,0.78260f,0.77377f,0.76476f,0.75556f,0.74617f,0.73661f,0.72686f,0.71692f,0.70680f,0.69650f,0.68602f,0.67535f,0.66449f,0.65345f,0.64223f,0.63082f,0.61923f,0.60746f,0.59550f,0.58336f,0.57103f,0.55852f,0.54583f,0.53295f,0.51989f,0.50664f,0.49321f,0.47960f };
+            static const float g[] = { 0.07176f,0.08339f,0.09498f,0.10652f,0.11802f,0.12947f,0.14087f,0.15223f,0.16354f,0.17481f,0.18603f,0.19720f,0.20833f,0.21941f,0.23044f,0.24143f,0.25237f,0.26327f,0.27412f,0.28492f,0.29568f,0.30639f,0.31706f,0.32768f,0.33825f,0.34878f,0.35926f,0.36970f,0.38008f,0.39043f,0.40072f,0.41097f,0.42118f,0.43134f,0.44145f,0.45152f,0.46153f,0.47151f,0.48144f,0.49132f,0.50115f,0.51094f,0.52069f,0.53040f,0.54015f,0.54995f,0.55979f,0.56967f,0.57958f,0.58950f,0.59943f,0.60937f,0.61931f,0.62923f,0.63913f,0.64901f,0.65886f,0.66866f,0.67842f,0.68812f,0.69775f,0.70732f,0.71680f,0.72620f,0.73551f,0.74472f,0.75381f,0.76279f,0.77165f,0.78037f,0.78896f,0.79740f,0.80569f,0.81381f,0.82177f,0.82955f,0.83714f,0.84455f,0.85175f,0.85875f,0.86554f,0.87211f,0.87844f,0.88454f,0.89040f,0.89600f,0.90142f,0.90673f,0.91193f,0.91701f,0.92197f,0.92680f,0.93151f,0.93609f,0.94053f,0.94484f,0.94901f,0.95304f,0.95692f,0.96065f,0.96423f,0.96765f,0.97092f,0.97403f,0.97697f,0.97974f,0.98234f,0.98477f,0.98702f,0.98909f,0.99098f,0.99268f,0.99419f,0.99551f,0.99663f,0.99755f,0.99828f,0.99879f,0.99910f,0.99919f,0.99907f,0.99873f,0.99817f,0.99739f,0.99638f,0.99514f,0.99366f,0.99195f,0.98999f,0.98775f,0.98524f,0.98246f,0.97941f,0.97610f,0.97255f,0.96875f,0.96470f,0.96043f,0.95593f,0.95121f,0.94627f,0.94113f,0.93579f,0.93025f,0.92452f,0.91861f,0.91253f,0.90627f,0.89986f,0.89328f,0.88655f,0.87968f,0.87267f,0.86553f,0.85826f,0.85087f,0.84337f,0.83576f,0.82806f,0.82025f,0.81236f,0.80439f,0.79634f,0.78823f,0.78005f,0.77181f,0.76352f,0.75519f,0.74682f,0.73842f,0.73000f,0.72140f,0.71250f,0.70330f,0.69382f,0.68408f,0.67408f,0.66386f,0.65341f,0.64277f,0.63193f,0.62093f,0.60977f,0.59846f,0.58703f,0.57549f,0.56386f,0.55214f,0.54036f,0.52854f,0.51667f,0.50479f,0.49291f,0.48104f,0.46920f,0.45740f,0.44565f,0.43399f,0.42241f,0.41093f,0.39958f,0.38836f,0.37729f,0.36638f,0.35566f,0.34513f,0.33482f,0.32473f,0.31489f,0.30530f,0.29599f,0.28696f,0.27824f,0.26981f,0.26152f,0.25334f,0.24526f,0.23730f,0.22945f,0.22170f,0.21407f,0.20654f,0.19912f,0.19182f,0.18462f,0.17753f,0.17055f,0.16368f,0.15693f,0.15028f,0.14374f,0.13731f,0.13098f,0.12477f,0.11867f,0.11268f,0.10680f,0.10102f,0.09536f,0.08980f,0.08436f,0.07902f,0.07380f,0.06868f,0.06367f,0.05878f,0.05399f,0.04931f,0.04474f,0.04028f,0.03593f,0.03169f,0.02756f,0.02354f,0.01963f,0.01583f };
+            static const float b[] = { 0.23217f,0.26149f,0.29024f,0.31844f,0.34607f,0.37314f,0.39964f,0.42558f,0.45096f,0.47578f,0.50004f,0.52373f,0.54686f,0.56942f,0.59142f,0.61286f,0.63374f,0.65406f,0.67381f,0.69300f,0.71162f,0.72968f,0.74718f,0.76412f,0.78050f,0.79631f,0.81156f,0.82624f,0.84037f,0.85393f,0.86692f,0.87936f,0.89123f,0.90254f,0.91328f,0.92347f,0.93309f,0.94214f,0.95064f,0.95857f,0.96594f,0.97275f,0.97899f,0.98461f,0.98930f,0.99303f,0.99583f,0.99773f,0.99876f,0.99896f,0.99835f,0.99697f,0.99485f,0.99202f,0.98851f,0.98436f,0.97959f,0.97423f,0.96833f,0.96190f,0.95498f,0.94761f,0.93981f,0.93161f,0.92305f,0.91416f,0.90496f,0.89550f,0.88580f,0.87590f,0.86581f,0.85559f,0.84525f,0.83484f,0.82437f,0.81389f,0.80342f,0.79299f,0.78264f,0.77240f,0.76230f,0.75237f,0.74265f,0.73316f,0.72393f,0.71500f,0.70599f,0.69651f,0.68660f,0.67627f,0.66556f,0.65448f,0.64308f,0.63137f,0.61938f,0.60713f,0.59466f,0.58199f,0.56914f,0.55614f,0.54303f,0.52981f,0.51653f,0.50321f,0.48987f,0.47654f,0.46325f,0.45002f,0.43688f,0.42386f,0.41098f,0.39826f,0.38575f,0.37345f,0.36140f,0.34963f,0.33816f,0.32701f,0.31622f,0.30581f,0.29581f,0.28623f,0.27712f,0.26849f,0.26038f,0.25280f,0.24579f,0.23937f,0.23356f,0.22835f,0.22370f,0.21960f,0.21602f,0.21294f,0.21032f,0.20815f,0.20640f,0.20504f,0.20406f,0.20343f,0.20311f,0.20310f,0.20336f,0.20386f,0.20459f,0.20552f,0.20663f,0.20788f,0.20926f,0.21074f,0.21230f,0.21391f,0.21555f,0.21719f,0.21880f,0.22038f,0.22188f,0.22328f,0.22456f,0.22570f,0.22667f,0.22744f,0.22800f,0.22831f,0.22836f,0.22811f,0.22754f,0.22663f,0.22536f,0.22369f,0.22161f,0.21918f,0.21650f,0.21358f,0.21043f,0.20706f,0.20348f,0.19971f,0.19577f,0.19165f,0.18738f,0.18297f,0.17842f,0.17376f,0.16899f,0.16412f,0.15918f,0.15417f,0.14910f,0.14398f,0.13883f,0.13367f,0.12849f,0.12332f,0.11817f,0.11305f,0.10797f,0.10294f,0.09798f,0.09310f,0.08831f,0.08362f,0.07905f,0.07461f,0.07031f,0.06616f,0.06218f,0.05837f,0.05475f,0.05134f,0.04814f,0.04516f,0.04243f,0.03993f,0.03753f,0.03521f,0.03297f,0.03082f,0.02875f,0.02677f,0.02487f,0.02305f,0.02131f,0.01966f,0.01809f,0.01660f,0.01520f,0.01387f,0.01264f,0.01148f,0.01041f,0.00942f,0.00851f,0.00769f,0.00695f,0.00629f,0.00571f,0.00522f,0.00481f,0.00449f,0.00424f,0.00408f,0.00401f,0.00401f,0.00410f,0.00427f,0.00453f,0.00486f,0.00529f,0.00579f,0.00638f,0.00705f,0.00780f,0.00863f,0.00955f,0.01055f };
+
+            // breakpoints
+            Mat X = linspace(0,1,256);
+
+            // now build lookup table
+            this->_lut = ColorMap::linear_colormap(X,
+                    Mat(256,1, CV_32FC1, (void*)r).clone(), // red
+                    Mat(256,1, CV_32FC1, (void*)g).clone(), // green
+                    Mat(256,1, CV_32FC1, (void*)b).clone(), // blue
+                    n);
+        }
+    };
+
     // UserColormap .
     class UserColorMap : public ColorMap {
     public:
@@ -724,6 +754,7 @@ namespace colormap
             colormap == COLORMAP_RAINBOW ? (colormap::ColorMap*)(new colormap::Rainbow) :
             colormap == COLORMAP_SPRING ? (colormap::ColorMap*)(new colormap::Spring) :
             colormap == COLORMAP_SUMMER ? (colormap::ColorMap*)(new colormap::Summer) :
+            colormap == COLORMAP_TURBO ? (colormap::ColorMap*)(new colormap::Turbo) :
             colormap == COLORMAP_TWILIGHT ? (colormap::ColorMap*)(new colormap::Twilight) :
             colormap == COLORMAP_TWILIGHT_SHIFTED ? (colormap::ColorMap*)(new colormap::TwilightShifted) :
             colormap == COLORMAP_VIRIDIS ? (colormap::ColorMap*)(new colormap::Viridis) :

modules/ts/src/ts.cpp

@@ -512,8 +512,9 @@ string TS::str_from_code( const TS::FailureCode code )
     return "Generic/Unknown";
 }
 
-static int tsErrorCallback( int status, const char* func_name, const char* err_msg, const char* file_name, int line, TS* ts )
+static int tsErrorCallback( int status, const char* func_name, const char* err_msg, const char* file_name, int line, void* data )
 {
+    TS* ts = (TS*)data;
     const char* delim = std::string(err_msg).find('\n') == std::string::npos ? "" : "\n";
     ts->printf(TS::LOG, "OpenCV Error:\n\t%s (%s%s) in %s, file %s, line %d\n", cvErrorStr(status), delim, err_msg, func_name[0] != 0 ? func_name : "unknown function", file_name, line);
     return 0;

modules/video/src/lkpyramid.cpp

@@ -237,13 +237,12 @@ void cv::detail::LKTrackerInvoker::operator()(const Range& range) const
         acctype iA11 = 0, iA12 = 0, iA22 = 0;
         float A11, A12, A22;
 
-#if CV_SSE2
-        __m128i qw0 = _mm_set1_epi32(iw00 + (iw01 << 16));
-        __m128i qw1 = _mm_set1_epi32(iw10 + (iw11 << 16));
-        __m128i z = _mm_setzero_si128();
-        __m128i qdelta_d = _mm_set1_epi32(1 << (W_BITS1-1));
-        __m128i qdelta = _mm_set1_epi32(1 << (W_BITS1-5-1));
-        __m128 qA11 = _mm_setzero_ps(), qA12 = _mm_setzero_ps(), qA22 = _mm_setzero_ps();
+#if CV_SIMD128 && !CV_NEON
+        v_int16x8 qw0((short)(iw00), (short)(iw01), (short)(iw00), (short)(iw01), (short)(iw00), (short)(iw01), (short)(iw00), (short)(iw01));
+        v_int16x8 qw1((short)(iw10), (short)(iw11), (short)(iw10), (short)(iw11), (short)(iw10), (short)(iw11), (short)(iw10), (short)(iw11));
+        v_int32x4 qdelta_d = v_setall_s32(1 << (W_BITS1-1));
+        v_int32x4 qdelta = v_setall_s32(1 << (W_BITS1-5-1));
+        v_float32x4 qA11 = v_setzero_f32(), qA12 = v_setzero_f32(), qA22 = v_setzero_f32();
 #endif
 
 #if CV_NEON
@@ -273,44 +272,75 @@ void cv::detail::LKTrackerInvoker::operator()(const Range& range) const
 
             x = 0;
 
-#if CV_SSE2
-            for( ; x <= winSize.width*cn - 4; x += 4, dsrc += 4*2, dIptr += 4*2 )
+#if CV_SIMD128 && !CV_NEON
+            for( ; x <= winSize.width*cn - 8; x += 8, dsrc += 8*2, dIptr += 8*2 )
             {
-                __m128i v00, v01, v10, v11, t0, t1;
-
-                v00 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int*)(src + x)), z);
-                v01 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int*)(src + x + cn)), z);
-                v10 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int*)(src + x + stepI)), z);
-                v11 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int*)(src + x + stepI + cn)), z);
-
-                t0 = _mm_add_epi32(_mm_madd_epi16(_mm_unpacklo_epi16(v00, v01), qw0),
-                                   _mm_madd_epi16(_mm_unpacklo_epi16(v10, v11), qw1));
-                t0 = _mm_srai_epi32(_mm_add_epi32(t0, qdelta), W_BITS1-5);
-                _mm_storel_epi64((__m128i*)(Iptr + x), _mm_packs_epi32(t0,t0));
-
-                v00 = _mm_loadu_si128((const __m128i*)(dsrc));
-                v01 = _mm_loadu_si128((const __m128i*)(dsrc + cn2));
-                v10 = _mm_loadu_si128((const __m128i*)(dsrc + dstep));
-                v11 = _mm_loadu_si128((const __m128i*)(dsrc + dstep + cn2));
-
-                t0 = _mm_add_epi32(_mm_madd_epi16(_mm_unpacklo_epi16(v00, v01), qw0),
-                                   _mm_madd_epi16(_mm_unpacklo_epi16(v10, v11), qw1));
-                t1 = _mm_add_epi32(_mm_madd_epi16(_mm_unpackhi_epi16(v00, v01), qw0),
-                                   _mm_madd_epi16(_mm_unpackhi_epi16(v10, v11), qw1));
-                t0 = _mm_srai_epi32(_mm_add_epi32(t0, qdelta_d), W_BITS1);
-                t1 = _mm_srai_epi32(_mm_add_epi32(t1, qdelta_d), W_BITS1);
-                v00 = _mm_packs_epi32(t0, t1); // Ix0 Iy0 Ix1 Iy1 ...
-
-                _mm_storeu_si128((__m128i*)dIptr, v00);
-                t0 = _mm_srai_epi32(v00, 16); // Iy0 Iy1 Iy2 Iy3
-                t1 = _mm_srai_epi32(_mm_slli_epi32(v00, 16), 16); // Ix0 Ix1 Ix2 Ix3
-
-                __m128 fy = _mm_cvtepi32_ps(t0);
-                __m128 fx = _mm_cvtepi32_ps(t1);
-
-                qA22 = _mm_add_ps(qA22, _mm_mul_ps(fy, fy));
-                qA12 = _mm_add_ps(qA12, _mm_mul_ps(fx, fy));
-                qA11 = _mm_add_ps(qA11, _mm_mul_ps(fx, fx));
+                v_int32x4 t0, t1;
+                v_int16x8 v00, v01, v10, v11, t00, t01, t10, t11;
+
+                v00 = v_reinterpret_as_s16(v_load_expand(src + x));
+                v01 = v_reinterpret_as_s16(v_load_expand(src + x + cn));
+                v10 = v_reinterpret_as_s16(v_load_expand(src + x + stepI));
+                v11 = v_reinterpret_as_s16(v_load_expand(src + x + stepI + cn));
+
+                v_zip(v00, v01, t00, t01);
+                v_zip(v10, v11, t10, t11);
+
+                t0 = v_dotprod(t00, qw0, qdelta) + v_dotprod(t10, qw1);
+                t1 = v_dotprod(t01, qw0, qdelta) + v_dotprod(t11, qw1);
+                t0 = t0 >> (W_BITS1-5);
+                t1 = t1 >> (W_BITS1-5);
+                v_store(Iptr + x, v_pack(t0, t1));
+
+                v00 = v_reinterpret_as_s16(v_load(dsrc));
+                v01 = v_reinterpret_as_s16(v_load(dsrc + cn2));
+                v10 = v_reinterpret_as_s16(v_load(dsrc + dstep));
+                v11 = v_reinterpret_as_s16(v_load(dsrc + dstep + cn2));
+
+                v_zip(v00, v01, t00, t01);
+                v_zip(v10, v11, t10, t11);
+
+                t0 = v_dotprod(t00, qw0, qdelta_d) + v_dotprod(t10, qw1);
+                t1 = v_dotprod(t01, qw0, qdelta_d) + v_dotprod(t11, qw1);
+                t0 = t0 >> W_BITS1;
+                t1 = t1 >> W_BITS1;
+                v00 = v_pack(t0, t1); // Ix0 Iy0 Ix1 Iy1 ...
+                v_store(dIptr, v00);
+
+                v00 = v_reinterpret_as_s16(v_interleave_pairs(v_reinterpret_as_s32(v_interleave_pairs(v00))));
+                v_expand(v00, t1, t0);
+
+                v_float32x4 fy = v_cvt_f32(t0);
+                v_float32x4 fx = v_cvt_f32(t1);
+
+                qA22 = v_muladd(fy, fy, qA22);
+                qA12 = v_muladd(fx, fy, qA12);
+                qA11 = v_muladd(fx, fx, qA11);
+
+                v00 = v_reinterpret_as_s16(v_load(dsrc + 4*2));
+                v01 = v_reinterpret_as_s16(v_load(dsrc + 4*2 + cn2));
+                v10 = v_reinterpret_as_s16(v_load(dsrc + 4*2 + dstep));
+                v11 = v_reinterpret_as_s16(v_load(dsrc + 4*2 + dstep + cn2));
+
+                v_zip(v00, v01, t00, t01);
+                v_zip(v10, v11, t10, t11);
+
+                t0 = v_dotprod(t00, qw0, qdelta_d) + v_dotprod(t10, qw1);
+                t1 = v_dotprod(t01, qw0, qdelta_d) + v_dotprod(t11, qw1);
+                t0 = t0 >> W_BITS1;
+                t1 = t1 >> W_BITS1;
+                v00 = v_pack(t0, t1); // Ix0 Iy0 Ix1 Iy1 ...
+                v_store(dIptr + 4*2, v00);
+
+                v00 = v_reinterpret_as_s16(v_interleave_pairs(v_reinterpret_as_s32(v_interleave_pairs(v00))));
+                v_expand(v00, t1, t0);
+
+                fy = v_cvt_f32(t0);
+                fx = v_cvt_f32(t1);
+
+                qA22 = v_muladd(fy, fy, qA22);
+                qA12 = v_muladd(fx, fy, qA12);
+                qA11 = v_muladd(fx, fx, qA11);
             }
 #endif
 
@@ -417,14 +447,10 @@ void cv::detail::LKTrackerInvoker::operator()(const Range& range) const
             }
         }
 
-#if CV_SSE2
-        float CV_DECL_ALIGNED(16) A11buf[4], A12buf[4], A22buf[4];
-        _mm_store_ps(A11buf, qA11);
-        _mm_store_ps(A12buf, qA12);
-        _mm_store_ps(A22buf, qA22);
-        iA11 += A11buf[0] + A11buf[1] + A11buf[2] + A11buf[3];
-        iA12 += A12buf[0] + A12buf[1] + A12buf[2] + A12buf[3];
-        iA22 += A22buf[0] + A22buf[1] + A22buf[2] + A22buf[3];
+#if CV_SIMD128 && !CV_NEON
+        iA11 += v_reduce_sum(qA11);
+        iA12 += v_reduce_sum(qA12);
+        iA22 += v_reduce_sum(qA22);
 #endif
 
 #if CV_NEON
@@ -477,10 +503,10 @@ void cv::detail::LKTrackerInvoker::operator()(const Range& range) const
             iw11 = (1 << W_BITS) - iw00 - iw01 - iw10;
             acctype ib1 = 0, ib2 = 0;
             float b1, b2;
-#if CV_SSE2
-            qw0 = _mm_set1_epi32(iw00 + (iw01 << 16));
-            qw1 = _mm_set1_epi32(iw10 + (iw11 << 16));
-            __m128 qb0 = _mm_setzero_ps(), qb1 = _mm_setzero_ps();
+#if CV_SIMD128 && !CV_NEON
+            qw0 = v_int16x8((short)(iw00), (short)(iw01), (short)(iw00), (short)(iw01), (short)(iw00), (short)(iw01), (short)(iw00), (short)(iw01));
+            qw1 = v_int16x8((short)(iw10), (short)(iw11), (short)(iw10), (short)(iw11), (short)(iw10), (short)(iw11), (short)(iw10), (short)(iw11));
+            v_float32x4 qb0 = v_setzero_f32(), qb1 = v_setzero_f32();
 #endif
 
 #if CV_NEON
@@ -501,34 +527,32 @@ void cv::detail::LKTrackerInvoker::operator()(const Range& range) const
 
                 x = 0;
 
-#if CV_SSE2
+#if CV_SIMD128 && !CV_NEON
                 for( ; x <= winSize.width*cn - 8; x += 8, dIptr += 8*2 )
                 {
-                    __m128i diff0 = _mm_loadu_si128((const __m128i*)(Iptr + x)), diff1;
-                    __m128i v00 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i*)(Jptr + x)), z);
-                    __m128i v01 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i*)(Jptr + x + cn)), z);
-                    __m128i v10 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i*)(Jptr + x + stepJ)), z);
-                    __m128i v11 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i*)(Jptr + x + stepJ + cn)), z);
-
-                    __m128i t0 = _mm_add_epi32(_mm_madd_epi16(_mm_unpacklo_epi16(v00, v01), qw0),
-                                               _mm_madd_epi16(_mm_unpacklo_epi16(v10, v11), qw1));
-                    __m128i t1 = _mm_add_epi32(_mm_madd_epi16(_mm_unpackhi_epi16(v00, v01), qw0),
-                                               _mm_madd_epi16(_mm_unpackhi_epi16(v10, v11), qw1));
-                    t0 = _mm_srai_epi32(_mm_add_epi32(t0, qdelta), W_BITS1-5);
-                    t1 = _mm_srai_epi32(_mm_add_epi32(t1, qdelta), W_BITS1-5);
-                    diff0 = _mm_subs_epi16(_mm_packs_epi32(t0, t1), diff0);
-                    diff1 = _mm_unpackhi_epi16(diff0, diff0);
-                    diff0 = _mm_unpacklo_epi16(diff0, diff0); // It0 It0 It1 It1 ...
-                    v00 = _mm_loadu_si128((const __m128i*)(dIptr)); // Ix0 Iy0 Ix1 Iy1 ...
-                    v01 = _mm_loadu_si128((const __m128i*)(dIptr + 8));
-                    v10 = _mm_unpacklo_epi16(v00, v01);
-                    v11 = _mm_unpackhi_epi16(v00, v01);
-                    v00 = _mm_unpacklo_epi16(diff0, diff1);
-                    v01 = _mm_unpackhi_epi16(diff0, diff1);
-                    v00 = _mm_madd_epi16(v00, v10);
-                    v11 = _mm_madd_epi16(v01, v11);
-                    qb0 = _mm_add_ps(qb0, _mm_cvtepi32_ps(v00));
-                    qb1 = _mm_add_ps(qb1, _mm_cvtepi32_ps(v11));
+                    v_int16x8 diff0 = v_reinterpret_as_s16(v_load(Iptr + x)), diff1, diff2;
+                    v_int16x8 v00 = v_reinterpret_as_s16(v_load_expand(Jptr + x));
+                    v_int16x8 v01 = v_reinterpret_as_s16(v_load_expand(Jptr + x + cn));
+                    v_int16x8 v10 = v_reinterpret_as_s16(v_load_expand(Jptr + x + stepJ));
+                    v_int16x8 v11 = v_reinterpret_as_s16(v_load_expand(Jptr + x + stepJ + cn));
+
+                    v_int32x4 t0, t1;
+                    v_int16x8 t00, t01, t10, t11;
+                    v_zip(v00, v01, t00, t01);
+                    v_zip(v10, v11, t10, t11);
+
+                    t0 = v_dotprod(t00, qw0, qdelta) + v_dotprod(t10, qw1);
+                    t1 = v_dotprod(t01, qw0, qdelta) + v_dotprod(t11, qw1);
+                    t0 = t0 >> (W_BITS1-5);
+                    t1 = t1 >> (W_BITS1-5);
+                    diff0 = v_pack(t0, t1) - diff0;
+                    v_zip(diff0, diff0, diff2, diff1); // It0 It0 It1 It1 ...
+                    v00 = v_reinterpret_as_s16(v_load(dIptr)); // Ix0 Iy0 Ix1 Iy1 ...
+                    v01 = v_reinterpret_as_s16(v_load(dIptr + 8));
+                    v_zip(v00, v01, v10, v11);
+                    v_zip(diff2, diff1, v00, v01);
+                    qb0 += v_cvt_f32(v_dotprod(v00, v10));
+                    qb1 += v_cvt_f32(v_dotprod(v01, v11));
                 }
 #endif
 
@@ -614,11 +638,11 @@ void cv::detail::LKTrackerInvoker::operator()(const Range& range) const
                 }
             }
 
-#if CV_SSE2
-            float CV_DECL_ALIGNED(16) bbuf[4];
-            _mm_store_ps(bbuf, _mm_add_ps(qb0, qb1));
-            ib1 += bbuf[0] + bbuf[2];
-            ib2 += bbuf[1] + bbuf[3];
+#if CV_SIMD128 && !CV_NEON
+            v_float32x4 qf0, qf1;
+            v_recombine(v_interleave_pairs(qb0 + qb1), v_setzero_f32(), qf0, qf1);
+            ib1 += v_reduce_sum(qf0);
+            ib2 += v_reduce_sum(qf1);
 #endif
 
 #if CV_NEON