Support ConjugateTranspose for TF (openvinotoolkit#21586)

* conj transpose * Update * Fix Mistakes * Update src/frontends/tensorflow_common/src/op/conj_transpose.cpp Co-authored-by: Anastasiia Pnevskaia <[email protected]> * Update src/frontends/tensorflow/src/op_table.cpp Co-authored-by: Roman Kazantsev <[email protected]> * """Test the conjugate transpose layer.""" * u * Resolve Issues * Revert "conj transpose" This reverts commit 27833f3. * Update ConjTranspose * Update src/frontends/tensorflow_common/src/op/conj_transpose.cpp * update clang-format | fix test errors * Update src/frontends/tensorflow_common/src/op/conj_transpose.cpp * fix test error * Update src/frontends/tensorflow_common/src/op/conj_transpose.cpp --------- Co-authored-by: Anastasiia Pnevskaia <[email protected]> Co-authored-by: Roman Kazantsev <[email protected]>
xufang-lisa · Dec 30, 2023 · ded4938 · ded4938
1 parent 98c309f
commit ded4938
Show file tree

Hide file tree

Showing 4 changed files with 189 additions and 0 deletions.
diff --git a/src/frontends/tensorflow/src/op_table.cpp b/src/frontends/tensorflow/src/op_table.cpp
@@ -157,6 +157,7 @@ const std::map<std::string, CreatorFunction> get_supported_ops() {
         {"ClipByValue", CreatorFunction(translate_clip_by_value_op)},
         {"Complex", CreatorFunction(translate_complex_op)},
         {"ComplexAbs", CreatorFunction(translate_complex_abs_op)},
+        {"ConjugateTranspose", CreatorFunction(translate_conj_transpose_op)},
         {"Concat", CreatorFunction(translate_concat_op)},
         {"ConcatV2", CreatorFunction(translate_concat_op)},
         {"Const", CreatorFunction(translate_const_op)},

diff --git a/src/frontends/tensorflow_common/include/common_op_table.hpp b/src/frontends/tensorflow_common/include/common_op_table.hpp
@@ -50,6 +50,7 @@ OP_CONVERTER(translate_clip_by_value_op);
 OP_CONVERTER(translate_complex_op);
 OP_CONVERTER(translate_complex_abs_op);
 OP_CONVERTER(translate_concat_op);
+OP_CONVERTER(translate_conj_transpose_op);
 OP_CONVERTER(translate_const_op);
 OP_CONVERTER(translate_conv_2d_op);
 OP_CONVERTER(translate_conv_2d_backprop_input_op);

diff --git a/src/frontends/tensorflow_common/src/op/conj_transpose.cpp b/src/frontends/tensorflow_common/src/op/conj_transpose.cpp
@@ -0,0 +1,65 @@
+// Copyright (C) 2018-2023 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#include "common_op_table.hpp"
+#include "helper_ops/complex_type_mark.hpp"
+#include "openvino/op/concat.hpp"
+#include "openvino/op/constant.hpp"
+#include "openvino/op/gather.hpp"
+#include "openvino/op/negative.hpp"
+#include "openvino/op/shape_of.hpp"
+#include "openvino/op/transpose.hpp"
+
+using namespace std;
+using namespace ov::op;
+
+namespace ov {
+namespace frontend {
+namespace tensorflow {
+namespace op {
+
+OutputVector translate_conj_transpose_op(const NodeContext& node) {
+    default_op_checks(node, 2, {"ConjugateTranspose"}, true);
+
+    auto x = node.get_input(0);
+    auto perm = node.get_input(1);
+
+    auto complex_type_mark = as_type_ptr<ComplexTypeMark>(x.get_node_shared_ptr());
+    if (complex_type_mark) {
+        element::Type complex_part_type = complex_type_mark->get_complex_part_type();
+        auto x = complex_type_mark->input_value(0);
+
+        auto real_index = make_shared<v0::Constant>(element::i32, Shape{1}, 0);
+        auto imag_index = make_shared<v0::Constant>(element::i32, Shape{1}, 1);
+
+        auto gather_axis = make_shared<v0::Constant>(element::i32, Shape{1}, -1);
+
+        auto real = make_shared<v8::Gather>(x, real_index, gather_axis)->output(0);
+        auto imag = make_shared<v8::Gather>(x, imag_index, gather_axis)->output(0);
+
+        imag = make_shared<v0::Negative>(imag);
+
+        auto conj_tensor = make_shared<v0::Concat>(OutputVector{real, imag}, -1)->output(0);
+
+        OutputVector concat_inputs;
+        concat_inputs.push_back(perm);
+        concat_inputs.push_back(make_shared<v3::ShapeOf>(perm, perm.get_element_type()));
+
+        auto concat = make_shared<v0::Concat>(concat_inputs, 0);
+        auto conj_transpose = make_shared<v1::Transpose>(conj_tensor, concat);
+
+        set_node_name(node.get_name(), conj_transpose);
+        auto complex_transpose = make_shared<ComplexTypeMark>(conj_transpose, complex_part_type);
+        return {complex_transpose->output(0)};
+    }
+
+    auto conj_transpose = make_shared<v1::Transpose>(x, perm);
+    set_node_name(node.get_name(), conj_transpose);
+    return {conj_transpose};
+}
+
+}  // namespace op
+}  // namespace tensorflow
+}  // namespace frontend
+}  // namespace ov
diff --git a/tests/layer_tests/tensorflow_tests/test_tf_ConjugateTranspose.py b/tests/layer_tests/tensorflow_tests/test_tf_ConjugateTranspose.py
@@ -0,0 +1,122 @@
+# Copyright (C) 2018-2023 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+import pytest
+import numpy as np
+import tensorflow as tf
+from common.tf_layer_test_class import CommonTFLayerTest
+
+# Testing operation ConjugateTranspose
+# Documentation: https://www.tensorflow.org/api_docs/python/tf/raw_ops/ConjugateTranspose
+
+
+class TestComplexConjugateTranspose(CommonTFLayerTest):
+
+    def _prepare_input(self, inputs_info):
+
+        rng = np.random.default_rng()
+        assert 'real_part' in inputs_info
+        real_part_shape = inputs_info['real_part']
+        assert 'imag_part' in inputs_info
+        imag_part_shape = inputs_info['imag_part']
+
+        inputs_data = {}
+        inputs_data['real_part'] = 4 * rng.random(real_part_shape).astype(np.float32) - 2
+        inputs_data['imag_part'] = 4 * rng.random(imag_part_shape).astype(np.float32) - 2
+
+        return inputs_data
+
+    def create_complex_conjugate_transpose_net(self, input_shape, perm):
+        """
+            TensorFlow net                                 IR net
+
+            Placeholder->ConjugateTranspose    =>       Placeholder->Transpose->Conjugate->Transpose
+        """
+
+        tf.compat.v1.reset_default_graph()
+
+        # Create the graph and model
+        with tf.compat.v1.Session() as sess:
+            real_part = tf.compat.v1.placeholder(np.float32, input_shape, 'real_part')
+            imag_part = tf.compat.v1.placeholder(np.float32, input_shape, 'imag_part')
+
+            complex_input = tf.raw_ops.Complex(real=real_part, imag=imag_part)
+
+            conj_tranpose = tf.raw_ops.ConjugateTranspose(x=complex_input, perm=perm, name = "Operation")
+            real = tf.raw_ops.Real(input=conj_tranpose)
+            img = tf.raw_ops.Imag(input=conj_tranpose)
+
+            tf.compat.v1.global_variables_initializer()
+            tf_net = sess.graph_def
+
+        ref_net = None
+
+        return tf_net, ref_net
+
+
+    test_data = [
+        (dict(input_shape=[1, 2], perm=[1, 0])),
+        (dict(input_shape=[1, 2, 3], perm=[2, 1, 0])),
+        (dict(input_shape=[1, 2, 3, 4], perm=[0, 3, 2, 1])),
+        (dict(input_shape=[1, 2, 3, 4, 5, 6], perm=[0, 2, 1, 3, 4, 5])),
+    ]
+
+    @pytest.mark.parametrize("params", test_data)
+    @pytest.mark.precommit_tf_fe
+    @pytest.mark.nightly
+    def test_conjugate_transpose(self, params, ie_device, precision, ir_version, temp_dir,
+                                 use_new_frontend, use_old_api):
+        self._test(*self.create_complex_conjugate_transpose_net(**params),
+                   ie_device, precision, ir_version, temp_dir=temp_dir,
+                   use_new_frontend=use_new_frontend, use_old_api=use_old_api)
+
+
+class TestConjugateTranspose(CommonTFLayerTest):
+
+    def _prepare_input(self, inputs_info):
+
+        assert 'input' in inputs_info
+        input_shape = inputs_info['input']
+
+        inputs_data = {}
+        inputs_data['input'] = np.random.default_rng().random(input_shape).astype(np.float32)
+
+        return inputs_data
+
+    def create_conjugate_transpose_net(self, input_shape, perm):
+        """
+            TensorFlow net                                 IR net
+
+            Placeholder->ConjugateTranspose    =>       Placeholder->Transpose->Conjugate->Transpose
+        """
+
+        tf.compat.v1.reset_default_graph()
+
+        # Create the graph and model
+        with tf.compat.v1.Session() as sess:
+            input = tf.compat.v1.placeholder(np.float32, input_shape, 'input')
+
+            tf.raw_ops.ConjugateTranspose(x=input, perm=perm, name = "Operation")
+
+            tf.compat.v1.global_variables_initializer()
+            tf_net = sess.graph_def
+
+        ref_net = None
+
+        return tf_net, ref_net
+
+    test_data = [
+        (dict(input_shape=[1, 2], perm=[1, 0])),
+        (dict(input_shape=[1, 2, 3], perm=[2, 1, 0])),
+        (dict(input_shape=[1, 2, 3, 4], perm=[0, 3, 2, 1])),
+        (dict(input_shape=[1, 2, 3, 4, 5, 6], perm=[0, 2, 1, 3, 4, 5])),
+    ]
+
+    @pytest.mark.parametrize("params", test_data)
+    @pytest.mark.precommit_tf_fe
+    @pytest.mark.nightly
+    def test_conjugate_transpose(self, params, ie_device, precision, ir_version, temp_dir,
+                                 use_new_frontend, use_old_api):
+        self._test(*self.create_conjugate_transpose_net(**params),
+                   ie_device, precision, ir_version, temp_dir=temp_dir,
+                   use_new_frontend=use_new_frontend, use_old_api=use_old_api)