Merge remote-tracking branch 'upstream/master' into mk/ov_pybind_poc

inference-engine/ie_bridges/python/tests/test_Blob.py

@@ -54,7 +54,7 @@ def test_get_buffer():
     ("BIN", np.int8),
     ("BF16", np.float16),
 ])
-def test_writes_to_buffer(precision, numpy_precision):
+def test_write_to_buffer(precision, numpy_precision):
     tensor_desc = TensorDesc(precision, [1, 3, 127, 127], "NCHW")
     array = np.zeros(shape=(1, 3, 127, 127), dtype=numpy_precision)
     blob = Blob(tensor_desc, array)
@@ -63,62 +63,6 @@ def test_writes_to_buffer(precision, numpy_precision):
     assert np.array_equal(blob.buffer, ones_arr)
 
 
-def write_to_buffer(precision, numpy_precision):
-    tensor_desc = TensorDesc(precision, [1, 3, 127, 127], "NCHW")
-    array = np.zeros(shape=(1, 3, 127, 127), dtype=numpy_precision)
-    blob = Blob(tensor_desc, array)
-    ones_arr = np.ones(shape=(1, 3, 127, 127), dtype=numpy_precision)
-    blob.buffer[:] = ones_arr
-    assert np.array_equal(blob.buffer, ones_arr)
-
-def test_write_to_buffer_fp32():
-    write_to_buffer("FP32", np.float32)
-
-
-def test_write_to_buffer_fp64():
-    write_to_buffer("FP64", np.float64)
-
-
-def test_write_to_buffer_fp16():
-    write_to_buffer("FP16", np.float16)
-
-
-def test_write_to_buffer_int8():
-    write_to_buffer("I8", np.int8)
-
-
-def test_write_to_buffer_uint8():
-    write_to_buffer("U8", np.uint8)
-
-
-def test_write_to_buffer_int32():
-    write_to_buffer("I32", np.int32)
-
-
-def test_write_to_buffer_int16():
-    write_to_buffer("I16", np.int16)
-
-
-def test_write_to_buffer_uint16():
-    write_to_buffer("U16", np.uint16)
-
-
-def test_write_to_buffer_int64():
-    write_to_buffer("I64", np.int64)
-
-
-def test_write_to_buffer_bool():
-    write_to_buffer("BOOL", np.uint8)
-
-
-def test_write_to_buffer_bin():
-    write_to_buffer("BIN", np.int8)
-
-
-def test_write_to_buffer_bf16():
-    write_to_buffer("BF16", np.float16)
-
-
 def test_write_numpy_scalar_int64():
     tensor_desc = TensorDesc("I64", [], "SCALAR")
     scalar = np.array(0, dtype=np.int64)

inference-engine/ie_bridges/python/tests/test_ExecutableNetwork.py

@@ -234,10 +234,11 @@ def test_plugin_accessible_after_deletion(device):
     del ie_core
 
 
-@pytest.mark.skipif(os.environ.get("TEST_DEVICE", "CPU") == "ARM",
-                    reason=f"Cannot run test on device {os.environ.get('TEST_DEVICE')}")
 def test_exec_graph(device):
     ie_core = ie.IECore()
+    if device == "CPU":
+        if ie_core.get_metric(device, "FULL_DEVICE_NAME") == "arm_compute::NEON":
+            pytest.skip("Can't run on ARM plugin due-to get_exec_graph_info method isn't implemented")
     net = ie_core.read_network(model=test_net_xml, weights=test_net_bin)
     exec_net = ie_core.load_network(net, device)
     img = read_image()
@@ -294,9 +295,11 @@ def test_get_metric(device):
     assert network_name == "test_model"
 
 
-@pytest.mark.skipif(os.environ.get("TEST_DEVICE", "CPU") != "CPU", reason="Device independent test")
+@pytest.mark.skipif(os.environ.get("TEST_DEVICE", "CPU") != "CPU", reason="Device dependent test")
 def test_get_config(device):
     ie_core = ie.IECore()
+    if ie_core.get_metric(device, "FULL_DEVICE_NAME") == "arm_compute::NEON":
+        pytest.skip("Can't run on ARM plugin due-to CPU dependent test")
     net = ie_core.read_network(model=test_net_xml, weights=test_net_bin)
     exec_net = ie_core.load_network(net, device)
     config = exec_net.get_config("PERF_COUNT")

inference-engine/ie_bridges/python/tests/test_IECore.py

@@ -4,7 +4,6 @@
 import os
 import pytest
 from sys import platform
-import numpy as np
 from pathlib import Path
 
 from openvino.inference_engine import IENetwork, IECore, ExecutableNetwork
@@ -61,8 +60,11 @@ def test_load_network_wrong_device():
 
 
 def test_query_network(device):
-    import ngraph as ng
     ie = IECore()
+    if device == "CPU":
+        if ie.get_metric(device, "FULL_DEVICE_NAME") == "arm_compute::NEON":
+            pytest.skip("Can't run on ARM plugin due-to ngraph")
+    import ngraph as ng
     net = ie.read_network(model=test_net_xml, weights=test_net_bin)
     query_res = ie.query_network(net, device)
     func_net = ng.function_from_cnn(net)
@@ -73,18 +75,22 @@ def test_query_network(device):
     assert next(iter(set(query_res.values()))) == device, "Wrong device for some layers"
 
 
-@pytest.mark.skipif(os.environ.get("TEST_DEVICE", "CPU") != "CPU", reason="Device independent test")
+@pytest.mark.skipif(os.environ.get("TEST_DEVICE", "CPU") != "CPU", reason="Device dependent test")
 def test_register_plugin():
     ie = IECore()
+    if ie.get_metric("CPU", "FULL_DEVICE_NAME") == "arm_compute::NEON":
+        pytest.skip("Can't run on ARM plugin due-to MKLDNNPlugin specific test")
     ie.register_plugin("MKLDNNPlugin", "BLA")
     net = ie.read_network(model=test_net_xml, weights=test_net_bin)
     exec_net = ie.load_network(net, "BLA")
     assert isinstance(exec_net, ExecutableNetwork), "Cannot load the network to the registered plugin with name 'BLA'"
 
 
-@pytest.mark.skipif(os.environ.get("TEST_DEVICE", "CPU") != "CPU", reason="Device independent test")
+@pytest.mark.skipif(os.environ.get("TEST_DEVICE", "CPU") != "CPU", reason="Device dependent test")
 def test_register_plugins():
     ie = IECore()
+    if ie.get_metric("CPU", "FULL_DEVICE_NAME") == "arm_compute::NEON":
+        pytest.skip("Can't run on ARM plugin due-to MKLDNNPlugin specific test")
     if platform == "linux" or platform == "linux2":
         ie.register_plugins(plugins_xml)
     elif platform == "darwin":
@@ -126,11 +132,12 @@ def test_get_metric_list_of_str():
                                                    "metric are strings!"
 
 
-
 @pytest.mark.skipif(os.environ.get("TEST_DEVICE", "CPU") != "CPU",
                     reason=f"Cannot run test on device {os.environ.get('TEST_DEVICE')}, Plugin specific test")
 def test_get_metric_tuple_of_two_ints():
     ie = IECore()
+    if ie.get_metric("CPU", "FULL_DEVICE_NAME") == "arm_compute::NEON":
+        pytest.skip("Can't run on ARM plugin due-to unsupported device metric")
     param = ie.get_metric("CPU", "RANGE_FOR_STREAMS")
     assert isinstance(param, tuple), "Parameter value for 'RANGE_FOR_STREAMS' " \
                                      f"metric must be tuple but {type(param)} is returned"
@@ -142,6 +149,8 @@ def test_get_metric_tuple_of_two_ints():
                     reason=f"Cannot run test on device {os.environ.get('TEST_DEVICE')}, Plugin specific test")
 def test_get_metric_tuple_of_three_ints():
     ie = IECore()
+    if ie.get_metric("CPU", "FULL_DEVICE_NAME") == "arm_compute::NEON":
+        pytest.skip("Can't run on ARM plugin due-to unsupported device metric")
     param = ie.get_metric("CPU", "RANGE_FOR_ASYNC_INFER_REQUESTS")
     assert isinstance(param, tuple), "Parameter value for 'RANGE_FOR_ASYNC_INFER_REQUESTS' " \
                                      f"metric must be tuple but {type(param)} is returned"
@@ -185,21 +194,25 @@ def test_read_network_from_onnx():
     net = ie.read_network(model=test_net_onnx)
     assert isinstance(net, IENetwork)
 
+
 def test_read_network_from_onnx_as_path():
     ie = IECore()
     net = ie.read_network(model=Path(test_net_onnx))
     assert isinstance(net, IENetwork)
 
+
 def test_read_network_from_prototxt():
     ie = IECore()
     net = ie.read_network(model=test_net_prototxt)
     assert isinstance(net, IENetwork)
 
+
 def test_read_network_from_prototxt_as_path():
     ie = IECore()
     net = ie.read_network(model=Path(test_net_prototxt))
     assert isinstance(net, IENetwork)
 
+
 def test_incorrect_xml():
     ie = IECore()
     with pytest.raises(Exception) as e:

inference-engine/ie_bridges/python/tests/test_IENetwork.py

@@ -4,7 +4,6 @@
 import os
 import pytest
 import warnings
-import numpy as np
 
 from openvino.inference_engine import IECore, IENetwork, DataPtr, InputInfoPtr, PreProcessInfo
 from conftest import model_path
@@ -183,9 +182,12 @@ def test_batch_size_after_reshape():
     assert net.input_info['data'].input_data.shape == [8, 3, 32, 32]
 
 
-def test_serialize():
-    import ngraph as ng
+def test_serialize(device):
     ie = IECore()
+    if device == "CPU":
+        if ie.get_metric(device, "FULL_DEVICE_NAME") == "arm_compute::NEON":
+            pytest.skip("Can't run on ARM plugin due-to ngraph")
+    import ngraph as ng
     net = ie.read_network(model=test_net_xml, weights=test_net_bin)
     net.serialize("./serialized_net.xml", "./serialized_net.bin")
     serialized_net = ie.read_network(model="./serialized_net.xml", weights="./serialized_net.bin")

inference-engine/ie_bridges/python/tests/test_InferRequest.py

@@ -376,6 +376,9 @@ def execute(self, input_data):
 
 def test_get_perf_counts(device):
     ie_core = ie.IECore()
+    if device == "CPU":
+        if ie_core.get_metric(device, "FULL_DEVICE_NAME") == "arm_compute::NEON":
+            pytest.skip("Can't run on ARM plugin due-to ngraph")
     net = ie_core.read_network(test_net_xml, test_net_bin)
     ie_core.set_config({"PERF_COUNT": "YES"}, device)
     exec_net = ie_core.load_network(net, device)
@@ -395,6 +398,8 @@ def test_get_perf_counts(device):
                             "Dynamic batch fully supported only on CPU")
 def test_set_batch_size(device):
     ie_core = ie.IECore()
+    if ie_core.get_metric(device, "FULL_DEVICE_NAME") == "arm_compute::NEON":
+        pytest.skip("Can't run on ARM plugin due-to dynamic batch isn't supported")
     ie_core.set_config({"DYN_BATCH_ENABLED": "YES"}, device)
     net = ie_core.read_network(test_net_xml, test_net_bin)
     net.batch_size = 10
@@ -438,6 +443,9 @@ def test_set_negative_batch_size(device):
 
 def test_blob_setter(device):
     ie_core = ie.IECore()
+    if device == "CPU":
+        if ie_core.get_metric(device, "FULL_DEVICE_NAME") == "arm_compute::NEON":
+            pytest.skip("Can't run on ARM plugin")
     net = ie_core.read_network(test_net_xml, test_net_bin)
     exec_net_1 = ie_core.load_network(network=net, device_name=device, num_requests=1)
 

inference-engine/src/cldnn_engine/cldnn_engine.cpp

@@ -32,6 +32,7 @@
 #include <transformations/common_optimizations/lin_op_sequence_fusion.hpp>
 #include <transformations/common_optimizations/weights_dequantize_to_fake_quantize.hpp>
 #include "transformations/common_optimizations/convert_quantize_dequantize.hpp"
+#include "transformations/common_optimizations/softmax_fusion.hpp"
 #include <transformations/op_conversions/convert_depth_to_space.hpp>
 #include <transformations/op_conversions/convert_space_to_depth.hpp>
 #include <transformations/op_conversions/convert_gelu.hpp>
@@ -323,6 +324,11 @@ InferenceEngine::CNNNetwork clDNNEngine::CloneAndTransformNetwork(const Inferenc
                     return false;
                 });
 
+            pass_config->set_callback<ngraph::pass::SoftmaxFusion>(
+                [](const_node_ptr &node) -> bool {
+                    return node->input_value(0).get_partial_shape().rank().get_length() > 5;
+                });
+
             // List of enabled/disabled transformations
             pass_config->disable<ngraph::pass::ConvertGELU>();
             pass_config->disable<ngraph::pass::ConvertMod>();

inference-engine/src/mkldnn_plugin/mkldnn_plugin.cpp

@@ -35,6 +35,7 @@
 #include <transformations/common_optimizations/weights_dequantize_to_fake_quantize.hpp>
 #include "transformations/common_optimizations/convert_quantize_dequantize.hpp"
 #include <transformations/common_optimizations/depth_to_space_fusion.hpp>
+#include <transformations/common_optimizations/softmax_fusion.hpp>
 #include <transformations/op_conversions/convert_depth_to_space.hpp>
 #include <transformations/op_conversions/convert_space_to_depth.hpp>
 #include <transformations/op_conversions/convert_gelu.hpp>
@@ -260,6 +261,11 @@ static void Transformation(CNNNetwork& clonedNetwork, const Config& conf) {
                 return MKLDNNMVNNode::checkAxesSuitability(node);
             });
 
+    pass_config->set_callback<ngraph::pass::SoftmaxFusion>(
+            [](const_node_ptr &node) -> bool {
+                return node->input_value(0).get_partial_shape().rank().get_length() > 5;
+            });
+
     // List of enabled/disabled transformations
     pass_config->disable<ngraph::pass::ConvertGELU>();
     pass_config->disable<ngraph::pass::Gelu7Downgrade>();

inference-engine/src/transformations/src/transformations/common_optimizations/softmax_fusion.cpp

@@ -28,6 +28,9 @@ ngraph::pass::SoftmaxFusion::SoftmaxFusion() {
     auto div_pattern = ngraph::pattern::wrap_type<opset6::Divide>({exp_pattern, reduce_sum_pattern});
 
     ngraph::matcher_pass_callback callback = [=](pattern::Matcher& m) {
+        if (transformation_callback(m.get_match_root()))
+            return false;
+
         const auto& pattern_map = m.get_pattern_value_map();
 
         auto reduce_max_axes = std::dynamic_pointer_cast<opset6::Constant>(pattern_map.at(reduce_max_axes_pattern).get_node_shared_ptr());

inference-engine/tests/functional/shared_test_classes/include/shared_test_classes/single_layer/reduce_ops.hpp

@@ -32,6 +32,7 @@ class ReduceOpsLayerTest : public testing::WithParamInterface<reduceMeanParams>,
                            virtual public LayerTestsUtils::LayerTestsCommon {
 public:
     static std::string getTestCaseName(testing::TestParamInfo<reduceMeanParams> obj);
+    InferenceEngine::Blob::Ptr GenerateInput(const InferenceEngine::InputInfo &info) const override;
 
 protected:
     void SetUp() override;

inference-engine/tests/functional/shared_test_classes/src/single_layer/reduce_ops.cpp

@@ -69,6 +69,21 @@ void ReduceOpsLayerTest::SetUp() {
     const ngraph::ResultVector results{std::make_shared<ngraph::opset3::Result>(reduce)};
     function = std::make_shared<ngraph::Function>(results, params, "Reduce");
 }
+InferenceEngine::Blob::Ptr ReduceOpsLayerTest::GenerateInput(const InferenceEngine::InputInfo &info) const {
+    ngraph::helpers::ReductionType reductionType = std::get<3>(GetParam());
+    InferenceEngine::Precision netPrecision = std::get<4>(GetParam());
+    if (reductionType == ngraph::helpers::ReductionType::LogicalOr ||
+        reductionType == ngraph::helpers::ReductionType::LogicalAnd) {
+        return FuncTestUtils::createAndFillBlob(info.getTensorDesc(), 2, 0);
+    } else if (!netPrecision.is_float()) {
+        return FuncTestUtils::createAndFillBlob(info.getTensorDesc(), 5, 0);
+    }
+    auto td = info.getTensorDesc();
+    auto blob = make_blob_with_precision(td);
+    blob->allocate();
+    CommonTestUtils::fill_data_random_float<InferenceEngine::Precision::FP32>(blob, 5, 0, 1000);
+    return blob;
+}
 
 InferenceEngine::Blob::Ptr ReduceOpsLayerWithSpecificInputTest::GenerateInput(const InferenceEngine::InputInfo &info) const {
     auto axis_vec = std::get<0>(GetParam());

inference-engine/tests/ie_test_utils/functional_test_utils/layer_tests_summary/merge_xmls.py

@@ -2,7 +2,6 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import xml.etree.ElementTree as ET
-from jinja2 import Environment, FileSystemLoader
 import argparse
 import os
 from datetime import datetime
@@ -81,7 +80,7 @@ def merge_xml(input_folder_paths: list, output_folder_paths: str):
             logger.error(f" {folder_path} is not a directory!")
             continue
 
-        xml_reports = glob.glob(os.path.join(folder_path, 'report*.xml'))
+        xml_reports = glob.glob(os.path.join(folder_path, '**/report*.xml'))
 
         xml_root = ET.parse(xml_reports[0]).getroot()
         for op in xml_root.find("ops_list"):
@@ -93,6 +92,8 @@ def merge_xml(input_folder_paths: list, output_folder_paths: str):
         summary.set("timestamp", timestamp)
         logger.info(f" Processing is finished")
 
+        if not os.path.exists(output_folder_paths):
+            os.mkdir(output_folder_paths)
         out_file_path = os.path.join(output_folder_paths, "report.xml")
         with open(out_file_path, "w") as xml_file:
             xml_file.write(ET.tostring(summary).decode('utf8'))

model-optimizer/automation/package_BOM.txt

@@ -797,6 +797,7 @@ mo/front/caffe/python_layer_extractor.py
 mo/front/caffe/register_custom_ops.py
 mo/front/common/__init__.py
 mo/front/common/custom_replacement_registry.py
+mo/front/common/extractors/__init__.py
 mo/front/common/extractors/utils.py
 mo/front/common/find_unsupported_ops.py
 mo/front/common/layout.py

model-optimizer/extensions/front/tf/AutomlEfficientDet.py

@@ -22,6 +22,7 @@
 
 class EfficientDet(FrontReplacementFromConfigFileGeneral):
     replacement_id = 'AutomlEfficientDet'
+    run_not_recursively = True
 
     def run_before(self):
         from extensions.front.ExpandDimsToUnsqueeze import ExpandDimsToUnsqueeze
@@ -57,10 +58,11 @@ def transform_graph(self, graph: Graph, replacement_descriptions: dict):
         # which includes padding and resizing from the model
         preprocessing_input_node_id = replacement_descriptions['preprocessing_input_node']
         assert preprocessing_input_node_id in graph.nodes, 'The node with name "{}" is not found in the graph. This ' \
-                                                           'node should provide scaled image output and is specified' \
+                                                           'should be a last node before image normalization and is specified' \
                                                            ' in the json file.'.format(preprocessing_input_node_id)
         preprocessing_input_node = Node(graph, preprocessing_input_node_id)
-        preprocessing_input_node.in_port(0).get_connection().set_source(parameter_node.out_port(0))
+        consumer_node = preprocessing_input_node.out_port(0).get_connection().get_destination().node
+        consumer_node.in_port(0).get_connection().set_source(parameter_node.out_port(0))
 
         preprocessing_output_node_id = replacement_descriptions['preprocessing_output_node']
         assert preprocessing_output_node_id in graph.nodes, 'The node with name "{}" is not found in the graph. This ' \

model-optimizer/extensions/front/tf/automl_efficientdet.json

@@ -2,7 +2,7 @@
   {
     "id": "AutomlEfficientDet",
     "custom_attributes": {
-      "preprocessing_input_node": "convert_image",
+      "preprocessing_input_node": "strided_slice_1",
       "preprocessing_output_node": "truediv",
       "aspect_ratios": [1.0, 1.0, 1.4, 0.7, 0.7, 1.4],
       "variance": [1.0, 1.0, 1.0, 1.0],