Merge pull request #194 from CaptainDario/feature-TF_String_support

TensorFlow String support for model output

android/build.gradle

@@ -55,13 +55,13 @@ android {
     }
 
     defaultConfig {
-        minSdkVersion 26
+        minSdkVersion 19
     }
 }
 
 
 dependencies {
-    def tflite_version = "2.12.0"
+    def tflite_version = "2.11.0"
 
     implementation("org.tensorflow:tensorflow-lite:${tflite_version}")
     implementation("org.tensorflow:tensorflow-lite-gpu:${tflite_version}")

lib/src/util/byte_conversion_utils.dart

@@ -14,6 +14,8 @@
  * limitations under the License.
  */
 
+import 'dart:convert';
+import 'dart:ffi';
 import 'dart:typed_data';
 
 import 'package:tflite_flutter/tflite_flutter.dart';
@@ -150,6 +152,34 @@ class ByteConversionUtils {
     );
   }
 
+  /// Decodes a TensorFlow string to a List<String>
+  static List<String> decodeTFStrings(Uint8List bytes) {
+    /// The decoded string
+    List<String> decodedStrings = [];
+
+    /// get the first 32bit int representing num of strings
+    int numStrings = ByteData.view(bytes.sublist(0, sizeOf<Int32>()).buffer)
+        .getInt32(0, Endian.little);
+
+    /// parse subsequent string position and sizes
+    for (int s = 0; s < numStrings; s++) {
+      // get current str index
+      int startIdx = ByteData.view(bytes
+              .sublist((1 + s) * sizeOf<Int32>(), (2 + s) * sizeOf<Int32>())
+              .buffer)
+          .getInt32(0, Endian.little);
+      // get next str index, or in last case the ending byte position
+      int endIdx = ByteData.view(bytes
+              .sublist((2 + s) * sizeOf<Int32>(), (3 + s) * sizeOf<Int32>())
+              .buffer)
+          .getInt32(0, Endian.little);
+
+      decodedStrings.add(utf8.decode(bytes.sublist(startIdx, endIdx)));
+    }
+
+    return decodedStrings;
+  }
+
   static Object convertBytesToObject(
       Uint8List bytes, TensorType tensorType, List<int> shape) {
     // stores flattened data
@@ -191,6 +221,9 @@ class ByteConversionUtils {
         list.add(ByteData.view(bytes.buffer).getInt64(i));
       }
       return list.reshape<int>(shape);
+    } else if (tensorType.value == TfLiteType.kTfLiteString) {
+      list.add(decodeTFStrings(bytes));
+      return list;
     }
     throw UnsupportedError("$tensorType is not Supported.");
   }