[TF FE] Support Bincount operation (#23418)

### Details:
 - implemented Bincount op
 - simple test cases

### Tickets:
#22071

---------

Co-authored-by: Roman Kazantsev <roman.kazantsev@intel.com>

src/frontends/tensorflow/docs/supported_ops.md

@@ -139,7 +139,7 @@ A "supported operation" is one that TensorFlow Frontend can convert to the OpenV
 | BiasAdd                                                 | YES                           |                               |
 | BiasAddGrad                                             | NO                            |                               |
 | BiasAddV1                                               | NO                            |                               |
-| Bincount                                                | NO                            |                               |
+| Bincount                                                | YES                           |                               |
 | Bitcast                                                 | NO                            |                               |
 | BitwiseAnd                                              | YES                           |                               |
 | BitwiseOr                                               | YES                           |                               |

src/frontends/tensorflow/src/op_table.cpp

@@ -218,6 +218,7 @@ const std::map<std::string, CreatorFunction> get_supported_ops() {
         {"BroadcastTo", CreatorFunction(translate_broadcast_to_op)},
         {"Bucketize", CreatorFunction(translate_bucketize_op)},
         {"BiasAdd", CreatorFunction(translate_bias_add_op)},
+        {"Bincount", CreatorFunction(translate_bincount_op)},
         {"Cast", CreatorFunction(translate_cast_op)},
         {"CheckNumerics", CreatorFunction(translate_identity_op)},
         {"CheckNumericsV2", CreatorFunction(translate_identity_op)},

src/frontends/tensorflow_common/include/common_op_table.hpp

@@ -42,6 +42,7 @@ OP_CONVERTER(translate_batch_mat_mul_op);
 OP_CONVERTER(translate_batch_mat_mul_with_type_op);
 OP_CONVERTER(translate_batch_to_space_nd_op);
 OP_CONVERTER(translate_bias_add_op);
+OP_CONVERTER(translate_bincount_op);
 OP_CONVERTER(translate_broadcast_args_op);
 OP_CONVERTER(translate_broadcast_to_op);
 OP_CONVERTER(translate_bucketize_op);

src/frontends/tensorflow_common/src/op/bincount.cpp

@@ -0,0 +1,79 @@
+// Copyright (C) 2018-2024 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#include <memory>
+
+#include "common_op_table.hpp"
+#include "openvino/core/shape.hpp"
+#include "openvino/op/broadcast.hpp"
+#include "openvino/op/convert.hpp"
+#include "openvino/op/equal.hpp"
+#include "openvino/op/less.hpp"
+#include "openvino/op/multiply.hpp"
+#include "openvino/op/range.hpp"
+#include "openvino/op/reduce_sum.hpp"
+#include "openvino/op/select.hpp"
+#include "openvino/op/shape_of.hpp"
+#include "openvino/op/unsqueeze.hpp"
+
+using namespace std;
+using namespace ov;
+using namespace ov::op;
+
+namespace ov {
+namespace frontend {
+namespace tensorflow {
+namespace op {
+
+OutputVector translate_bincount_op(const NodeContext& node) {
+    default_op_checks(node, 3, {"Bincount"});
+    auto arr = node.get_input(0);
+    auto size = node.get_input(1);
+    auto weights = node.get_input(2);
+
+    auto scalar_shape = make_shared<v0::Constant>(element::i32, ov::Shape{0}, std::vector<int32_t>{});
+    size = make_shared<v1::Reshape>(size, scalar_shape, false);
+
+    auto weights_type = weights.get_element_type();
+
+    if (weights.get_partial_shape() == ov::Shape{0}) {
+        auto arr_shape = make_shared<v3::ShapeOf>(arr, element::i32);
+        weights = make_shared<v0::Constant>(weights_type, Shape{}, std::vector<int>{1});
+        weights = make_shared<v3::Broadcast>(weights, arr_shape);
+    }
+
+    // implementation
+    auto start = make_shared<v0::Constant>(element::i32, Shape{}, std::vector<int>{0});
+    auto step = make_shared<v0::Constant>(element::i32, Shape{}, std::vector<int>{1});
+    auto range = make_shared<v4::Range>(start, size, step, element::i32);
+
+    // Reshape arr and weights to 1D tensors
+    auto const_flatten_shape = make_shared<v0::Constant>(element::i32, Shape{1}, std::vector<int32_t>{-1});
+    auto arr_reshaped = make_shared<v1::Reshape>(arr, const_flatten_shape, false);
+    auto weights_reshaped = make_shared<v1::Reshape>(weights, const_flatten_shape, false);
+
+    // Unsqueeze range to [size, 1] shape and unsqueeze arr and weights to shapes [1, num]
+    auto const_axis_zero = make_shared<v0::Constant>(element::i32, Shape{1}, vector<int>({0}));
+    auto const_axis_one = make_shared<v0::Constant>(element::i32, Shape{1}, vector<int>({1}));
+    auto unsqueeze_range = make_shared<v0::Unsqueeze>(range, const_axis_one);
+    auto unsqueeze_arr = make_shared<v0::Unsqueeze>(arr_reshaped, const_axis_zero);
+    auto unsqueeze_weights = make_shared<v0::Unsqueeze>(weights_reshaped, const_axis_zero);
+
+    // Generate a mask [size, num] on range == arr
+    auto mask = make_shared<v1::Equal>(unsqueeze_range, unsqueeze_arr);
+    // Compute the weighted mask
+    auto mask_casted = make_shared<v0::Convert>(mask, weights_type);
+
+    auto to_sum = make_shared<v1::Multiply>(mask_casted, unsqueeze_weights);
+    auto reduce_axis = make_shared<v0::Constant>(element::i32, Shape{}, 1);
+    auto result = make_shared<v1::ReduceSum>(to_sum, reduce_axis);
+
+    set_node_name(node.get_name(), result);
+
+    return {result};
+}
+}  // namespace op
+}  // namespace tensorflow
+}  // namespace frontend
+}  // namespace ov

tests/layer_tests/tensorflow_tests/test_tf_Bincount.py

@@ -0,0 +1,65 @@
+# Copyright (C) 2018-2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+import platform
+
+import numpy as np
+import pytest
+import tensorflow as tf
+from common.tf_layer_test_class import CommonTFLayerTest
+
+rng = np.random.default_rng()
+
+class TestBincount(CommonTFLayerTest):
+    def _prepare_input(self, inputs_info):
+        assert 'x:0' in inputs_info, "Test error: inputs_info must contain `x`"
+        x_shape = inputs_info['x:0']
+
+        inputs_data = {}
+        inputs_data['x:0'] = rng.integers(0, 8, x_shape).astype(np.int32)
+
+        if 'w:0' in inputs_info:
+            w_shape = inputs_info['w:0']
+            inputs_data['w:0'] = rng.uniform(-2.0, 2.0, w_shape).astype(self.weights_type)
+
+        return inputs_data
+
+    def create_bincount_net(self, input_shape, size, weights, weights_type):
+        tf.compat.v1.reset_default_graph()
+        # Create the graph and model
+        with tf.compat.v1.Session() as sess:
+            x = tf.compat.v1.placeholder(np.int32, input_shape, 'x')
+            s = tf.constant(size)
+            self.weights_type = weights_type
+            if weights is not None:
+              w = tf.compat.v1.placeholder(weights_type, input_shape, 'w')
+            else:
+              w = tf.constant([], dtype=weights_type)
+
+            tf.raw_ops.Bincount(arr=x, size=s, weights=w)
+            tf.compat.v1.global_variables_initializer()
+            tf_net = sess.graph_def
+
+        return tf_net, None
+
+    test_data = [
+        # with no weights
+        dict(input_shape=[], size=1, weights=None, weights_type=np.float32),
+        dict(input_shape=[2], size=2, weights=None, weights_type=np.float64),
+        dict(input_shape=[1,3], size=3, weights=None, weights_type=np.int32),
+        dict(input_shape=[3,1,4], size=4, weights=None, weights_type=np.int64),
+
+
+        # with weights
+        dict(input_shape=[], size=1, weights=True, weights_type=np.float32),
+        dict(input_shape=[2], size=2, weights=True, weights_type=np.float64),
+        dict(input_shape=[1,3], size=3, weights=True, weights_type=np.int32),
+        dict(input_shape=[3,1,4], size=4, weights=True, weights_type=np.int64),
+    ]
+
+    @pytest.mark.parametrize("params", test_data)
+    @pytest.mark.precommit_tf_fe
+    @pytest.mark.nightly
+    def test_bincount(self, params, ie_device, precision, ir_version, temp_dir):
+        self._test(*self.create_bincount_net(**params),
+                   ie_device, precision, ir_version, temp_dir=temp_dir)