[CPU] Fix RegionYolo (v3) node and Ngraph reference implementations (o…

…penvinotoolkit#5748)
yekruglov · Jun 7, 2021 · 1e53b9d · 1e53b9d
1 parent 2d39720
commit 1e53b9d
Show file tree

Hide file tree

Showing 4 changed files with 71 additions and 14 deletions.
diff --git a/inference-engine/src/mkldnn_plugin/nodes/mkldnn_region_yolo_node.cpp b/inference-engine/src/mkldnn_plugin/nodes/mkldnn_region_yolo_node.cpp
@@ -369,31 +369,38 @@ inline void MKLDNNRegionYoloNode::calculate_logistic(size_t start_index, int cou
 void MKLDNNRegionYoloNode::execute(mkldnn::stream strm) {
     auto inputDesc = getParentEdgeAt(0)->getDesc();
     auto outputDesc = getChildEdgeAt(0)->getDesc();
-    size_t mask_size = mask.size();
 
-    size_t IW = (inputDesc.getDims().size() > 3) ? inputDesc.getDims()[3] : 1;
-    size_t IH = (inputDesc.getDims().size() > 2) ? inputDesc.getDims()[2] : 1;
-    size_t IC = (inputDesc.getDims().size() > 1) ? inputDesc.getDims()[1] : 1;
     size_t B = (inputDesc.getDims().size() > 0) ? inputDesc.getDims()[0] : 1;
+    size_t IC = (inputDesc.getDims().size() > 1) ? inputDesc.getDims()[1] : 1;
+    size_t IH = (inputDesc.getDims().size() > 2) ? inputDesc.getDims()[2] : 1;
+    size_t IW = (inputDesc.getDims().size() > 3) ? inputDesc.getDims()[3] : 1;
 
+    size_t mask_size = mask.size();
     int end_index = 0;
     int num_ = 0;
+    int output_size = 0;
     if (do_softmax) {
         // Region layer (Yolo v2)
         end_index = IW * IH;
         num_ = num;
+        output_size = B * IH * IW * IC; // different shape combinations with the same overall size;
     } else {
         // Yolo layer (Yolo v3)
         end_index = IW * IH * (classes + 1);
         num_ = mask_size;
+        output_size = B * IH * IW * mask_size * (classes + coords + 1);
     }
+
+    if (output_size != getChildEdgeAt(0)->getMemoryPtr()->GetElementsCount())
+        IE_THROW() << "Incorrect layer configuration or output dimensions. " << output_size << " != " << getChildEdgeAt(0)->getMemoryPtr()->GetElementsCount();
+
     size_t inputs_size = IH * IW * num_ * (classes + coords + 1);
     size_t total_size = 2 * IH * IW;
 
     const auto *src_data = reinterpret_cast<const uint8_t *>(getParentEdgeAt(0)->getMemoryPtr()->GetPtr());
     auto *dst_data = reinterpret_cast<uint8_t *>(getChildEdgeAt(0)->getMemoryPtr()->GetPtr());
 
-    cpu_convert(src_data, dst_data, inputDesc.getPrecision(), outputDesc.getPrecision(), B * IC * IH * IW);
+    cpu_convert(src_data, dst_data, inputDesc.getPrecision(), outputDesc.getPrecision(), output_size);
 
     for (int b = 0; b < B; b++) {
         for (int n = 0; n < num_; n++) {

diff --git a/inference-engine/tests/functional/plugin/cpu/single_layer_tests/region_yolo.cpp b/inference-engine/tests/functional/plugin/cpu/single_layer_tests/region_yolo.cpp
@@ -69,11 +69,14 @@ class RegionYoloCPULayerTest : public testing::WithParamInterface<regionYoloPara
 
         selectedType = getPrimitiveType() + "_" + inPrc.name();
 
-        auto ngPrc = FuncTestUtils::PrecisionUtils::convertIE2nGraphPrc(inPrc);
-        auto param = std::make_shared<ngraph::op::Parameter>(ngPrc, inputShape);
-        auto region_yolo = std::make_shared<ngraph::op::v0::RegionYolo>(param, attributes.coordinates, attributes.classes, attributes.num_regions,
-                                                                        attributes.do_softmax, mask, attributes.start_axis, attributes.end_axis);
-        function = std::make_shared<ngraph::Function>(std::make_shared<ngraph::opset1::Result>(region_yolo), ngraph::ParameterVector{param}, "RegionYolo");
+        const auto ngPrc = FuncTestUtils::PrecisionUtils::convertIE2nGraphPrc(inPrc);
+        auto paramRegionYolo = ngraph::builder::makeParams(ngPrc, {inputShape});
+
+        const auto region_yolo = std::make_shared<ngraph::op::v0::RegionYolo>(paramRegionYolo[0],
+                                                                              attributes.coordinates, attributes.classes, attributes.num_regions,
+                                                                              attributes.do_softmax, mask, attributes.start_axis, attributes.end_axis);
+
+        function = makeNgraphFunction(ngPrc, paramRegionYolo, region_yolo, "RegionYolo");
     }
 };
 
@@ -99,7 +102,10 @@ const std::vector<ngraph::Shape> inShapes_mxnet = {
         {1, 75, 26, 26},
         {1, 75, 16, 16},
         {1, 75, 13, 13},
-        {1, 75, 8, 8}
+        {1, 75, 8, 8},
+        {1, 303, 7, 7},
+        {1, 303, 14, 14},
+        {1, 303, 28, 28},
 };
 
 const std::vector<ngraph::Shape> inShapes_v3 = {
@@ -158,4 +164,4 @@ INSTANTIATE_TEST_CASE_P(smoke_TestsRegionYolov3CPU, RegionYoloCPULayerTest, test
 INSTANTIATE_TEST_CASE_P(smoke_TestsRegionYoloMxnetCPU, RegionYoloCPULayerTest, testCase_yolov3_mxnet, RegionYoloCPULayerTest::getTestCaseName);
 INSTANTIATE_TEST_CASE_P(smoke_TestsRegionYoloCaffeCPU, RegionYoloCPULayerTest, testCase_yolov2_caffe, RegionYoloCPULayerTest::getTestCaseName);
 } // namespace
-} // namespace CPULayerTestsDefinitions
+} // namespace CPULayerTestsDefinitions
diff --git a/ngraph/core/reference/include/ngraph/runtime/reference/region_yolo.hpp b/ngraph/core/reference/include/ngraph/runtime/reference/region_yolo.hpp
@@ -86,24 +86,27 @@ namespace ngraph
 
                 const auto mask_size = mask.size();
 
-                std::copy(input, input + shape_size(input_shape), output);
-
                 int num_regions = 0;
                 int end_index = 0;
+                int output_size = 0;
 
                 if (do_softmax)
                 {
                     // Region layer (Yolo v2)
                     num_regions = regions;
                     end_index = width * height;
+                    output_size = shape_size(input_shape);
                 }
                 else
                 {
                     // Yolo layer (Yolo v3)
                     num_regions = mask_size;
                     end_index = width * height * (classes + 1);
+                    output_size = width * height * num_regions * (classes + coords + 1);
                 }
 
+                std::copy(input, input + output_size, output);
+
                 const int inputs_size = width * height * num_regions * (classes + coords + 1);
 
                 for (int batch_idx = 0; batch_idx < batches; batch_idx++)

diff --git a/ngraph/test/backend/region_yolo.in.cpp b/ngraph/test/backend/region_yolo.in.cpp
@@ -70,3 +70,44 @@ NGRAPH_TEST(${BACKEND_NAME}, region_yolo_v3_mxnet)
         shape, TEST_FILES, "region_out_yolov3_mxnet.data");
     test_case.run_with_tolerance_as_fp(1.0e-4f);
 }
+
+NGRAPH_TEST(${BACKEND_NAME}, region_yolo_v3_mxnet_2)
+{
+    const size_t num = 1;
+    const size_t coords = 4;
+    const size_t classes = 1;
+    const size_t batch = 1;
+    const size_t channels = 8;
+    const size_t width = 2;
+    const size_t height = 2;
+    const std::vector<int64_t> mask{0};
+    const int axis = 1;
+    const int end_axis = 3;
+
+    Shape input_shape{batch, channels, height, width};
+    Shape output_shape{batch, (classes + coords + 1) * mask.size(), height, width};
+
+    const auto A = make_shared<op::Parameter>(element::f32, input_shape);
+    const auto R = make_shared<op::v0::RegionYolo>(A, coords, classes, num, false, mask, axis, end_axis);
+    const auto f = make_shared<Function>(R, ParameterVector{A});
+
+    EXPECT_EQ(R->get_output_shape(0), output_shape);
+
+    auto test_case = test::TestCase<TestEngine>(f);
+    std::vector<float> input {
+        0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.6f, 0.7f, 0.8f,
+        0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.6f, 0.7f, 0.8f,
+        0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.6f, 0.7f, 0.8f,
+        0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.6f, 0.7f, 0.8f
+    };
+
+    std::vector<float> output {
+        0.52497f, 0.54983f, 0.57444f, 0.59868f, 0.62245f, 0.64565f, 0.66818f, 0.68997f,
+        0.1f,     0.2f,     0.3f,     0.4f,     0.5f,     0.6f,     0.7f,     0.8f,
+        0.52497f, 0.54983f, 0.57444f, 0.59868f, 0.62245f, 0.64565f, 0.66818f, 0.68997f,
+    };
+
+    test_case.add_input<float>(input);
+    test_case.add_expected_output<float>(output);
+    test_case.run_with_tolerance_as_fp(1.0e-4f);
+}