[MO] Updating condition in TF Const value extracting (#6322)

* update value extracting condition

* added a comment

* updated unittest

* added a scalar test, little fix for scalar values extracting

model-optimizer/mo/front/tf/extractors/utils.py

@@ -70,8 +70,18 @@ def tf_tensor_content(tf_dtype, shape, pb_tensor):
                 log.error(decode_err_msg, extra={'is_warning': True})
                 value = np.array(type_helper[1](pb_tensor))
 
-    if len(shape) == 0 or shape.prod() == 0:
-        if len(value) == 1:
+    # Ignore an empty value, if len(shape) > 1
+    # For example, value = [] and shape = [1, 1, 0]
+    # This is needed to reshape this value later and to return reshaped value = [[[]]]
+    # Otherwise there can be failures during partial inference, because we are storing an empty value with incorrect
+    # shape
+    if len(shape) == 0 or (len(shape) == 1 and shape.prod() == 0):
+        try:
+            value_length = len(value)
+        except TypeError:
+            # case, when value is a scalar
+            value_length = 0
+        if value_length == 1:
             # return scalar if shape is [] otherwise broadcast according to shape
             try:
                 return np.array(value[0], dtype=type_helper[0])

model-optimizer/unit_tests/mo/front/tf/extractors/utils_test.py

@@ -211,3 +211,30 @@ class TensorContentParsing(unittest.TestCase):
         with self.assertLogs(log.getLogger(), level="ERROR") as cm:
             result = tf_tensor_content(pb_tensor.dtype, shape, pb_tensor)
             self.assertEqual([warning_message, warning_message], cm.output)
+
+    def test_empty_value(self):
+        pb_tensor = PB({
+            'dtype': 1,
+            'float_val': []
+        })
+
+        shape = int64_array([1, 1, 0])
+        tf_dtype = pb_tensor.dtype
+        ref = np.array([[[]]], dtype=np.float32)
+        res = tf_tensor_content(tf_dtype, shape, pb_tensor)
+
+        self.assertEqual(res.shape, ref.shape)
+        self.assertTrue(np.all(res == ref))
+
+    def test_scalar_value(self):
+        pb_tensor = PB({
+            'dtype': 3,
+            'int_val': 4
+        })
+
+        shape = int64_array([])
+        tf_dtype = pb_tensor.dtype
+        ref = np.array(4, dtype=np.int32)
+        res = tf_tensor_content(tf_dtype, shape, pb_tensor)
+
+        self.assertEqual(ref, res)