From 870a6c061a5ee28a724b3e0d56a58a774c5a995e Mon Sep 17 00:00:00 2001
From: Yegor Kruglov <yegor.kruglov@intel.com>
Date: Sun, 27 Dec 2020 22:15:17 +0300
Subject: [PATCH] Fix of ascii decode problem in TF Const with string value
 (#3672)

* added try/except to tf_tensor_content

* code refactoring

* added unittest

* update test
---
 .../mo/front/tf/extractors/utils.py           | 12 +++++++++++-
 .../mo/front/tf/extractors/utils_test.py      | 19 +++++++++++++++++++
 2 files changed, 30 insertions(+), 1 deletion(-)

diff --git a/model-optimizer/mo/front/tf/extractors/utils.py b/model-optimizer/mo/front/tf/extractors/utils.py
index f17d5ed3f5e..1b74a71ccb4 100644
--- a/model-optimizer/mo/front/tf/extractors/utils.py
+++ b/model-optimizer/mo/front/tf/extractors/utils.py
@@ -72,7 +72,17 @@ def tf_tensor_content(tf_dtype, shape, pb_tensor):
         value = np.array(np.frombuffer(pb_tensor.tensor_content, type_helper[0]))
     else:
         # load typed value
-        value = np.array(type_helper[1](pb_tensor), dtype=type_helper[0])
+        if type_helper[0] != np.str:
+            value = np.array(type_helper[1](pb_tensor), dtype=type_helper[0])
+        else:
+            try:
+                value = np.array(type_helper[1](pb_tensor), dtype=type_helper[0])
+            except UnicodeDecodeError:
+                log.error(
+                    'Failed to parse a tensor with Unicode characters. Note that Inference Engine does not support '
+                    'string literals, so the string constant should be eliminated from the graph.',
+                    extra={'is_warning': True})
+                value = np.array(type_helper[1](pb_tensor))
 
     if len(shape) == 0 or shape.prod() == 0:
         if len(value) == 1:
diff --git a/model-optimizer/mo/front/tf/extractors/utils_test.py b/model-optimizer/mo/front/tf/extractors/utils_test.py
index f10466b029f..15306206866 100644
--- a/model-optimizer/mo/front/tf/extractors/utils_test.py
+++ b/model-optimizer/mo/front/tf/extractors/utils_test.py
@@ -14,10 +14,12 @@
  limitations under the License.
 """
 
+import logging as log
 import unittest
 
 import numpy as np
 
+from mo.front.common.partial_infer.utils import int64_array
 from mo.front.tf.extractors.utils import collect_tf_attrs, tf_tensor_content
 from mo.utils.unittest.extractors import PB
 
@@ -193,3 +195,20 @@ class TensorContentParsing(unittest.TestCase):
                [[6, 6], [6, 6], [6, 6], [6, 6], [6, 6]]]
         res = tf_tensor_content(tf_dtype, shape, pb_tensor)
         self.assertTrue(np.all(res == ref))
+
+    def test_str_decode(self):
+        pb_tensor = PB({
+            'dtype': 7,
+            'string_val': [b"\037\000\036\000\002\000\303\237\035\000\002"]
+        })
+        tf_dtype = pb_tensor.dtype
+        shape = int64_array([1])
+        warning_message = 'ERROR:root:Failed to parse a tensor with Unicode characters. Note that Inference Engine ' \
+                          'does not support string literals, so the string constant should be eliminated from the ' \
+                          'graph.'
+        ref_val = np.array([b'\x1f\x00\x1e\x00\x02\x00\xc3\x9f\x1d\x00\x02'])
+        with self.assertLogs(log.getLogger(), level="ERROR") as cm:
+            result = tf_tensor_content(tf_dtype, shape, pb_tensor)
+            self.assertEqual([warning_message], cm.output)
+            self.assertEqual(ref_val, result)
+