--rigorous_optimization_for_myriad added

Author: PINTO0309

Dockerfile

@@ -7,7 +7,7 @@ ARG CPVER=cp38
 ARG OPENVINOVER=2021.4.582
 ARG OPENVINOROOTDIR=/opt/intel/openvino_2021
 ARG TENSORRTVER=cuda11.3-trt8.0.1.6-ga-20210626
-ARG APPVER=v1.13.0
+ARG APPVER=v1.13.1
 ARG wkdir=/home/user
 
 # dash -> bash

README.md

@@ -308,6 +308,7 @@ usage: tflite2tensorflow
   [--vpu_number_of_shaves VPU_NUMBER_OF_SHAVES]
   [--vpu_number_of_cmx_slices VPU_NUMBER_OF_CMX_SLICES]
   [--optimizing_for_openvino_and_myriad]
+  [--rigorous_optimization_for_myriad]
   [--replace_swish_and_hardswish]
   [--optimizing_hardswish_for_edgetpu]
   [--replace_prelu_and_minmax]
@@ -393,6 +394,8 @@ optional arguments:
                         vpu number of cmx slices. Default: 4
   --optimizing_for_openvino_and_myriad
                         Optimizing graph for openvino/myriad
+  --rigorous_optimization_for_myriad
+                        Replace operations that are not supported by myriad with operations that are as feasible as possible.
   --replace_swish_and_hardswish
                         Replace swish and hard-swish with each other
   --optimizing_hardswish_for_edgetpu
@@ -422,6 +425,16 @@ $ tflite2tensorflow \
 ```
 or
 ```
+$ tflite2tensorflow \
+  --model_path segm_full_v679.tflite \
+  --flatc_path ../flatc \
+  --schema_path ../schema.fbs \
+  --output_pb \
+  --optimizing_for_openvino_and_myriad \
+  --rigorous_optimization_for_myriad
+```
+or
+```
 $ tflite2tensorflow \
   --model_path segm_full_v679.tflite \
   --flatc_path ../flatc \

setup.py

@@ -11,7 +11,7 @@
     setup(
         name="tflite2tensorflow",
         scripts=scripts,
-        version="1.13.0",
+        version="1.13.1",
         description="Generate saved_model, tfjs, tf-trt, EdgeTPU, CoreML, quantized tflite, ONNX, OpenVINO, Myriad Inference Engine blob and .pb from .tflite.",
         long_description=long_description,
         long_description_content_type="text/markdown",

tflite2tensorflow/tflite2tensorflow.py

@@ -419,7 +419,8 @@ def make_graph(
     replace_swish_and_hardswish,
     replace_prelu_and_minmax,
     optimizing_for_edgetpu_flg,
-    optimizing_for_openvino_and_myriad):
+    optimizing_for_openvino_and_myriad,
+    rigorous_optimization_for_myriad):
 
     import tensorflow.compat.v1 as tf
     tf.get_logger().setLevel('INFO')
@@ -3240,10 +3241,22 @@ def pad_v2(x, paddings, constant_values):
                 input_tensor1 = backward_quantization(input_detail, input_tensor1)
 
             output_detail = interpreter._get_tensor_details(op['outputs'][0])
-            output_tensor = tf.math.abs(
-                input_tensor1,
-                name=get_op_name(output_detail['name'])
-            )
+            if not optimizing_for_openvino_and_myriad:
+                output_tensor = tf.math.abs(
+                    input_tensor1,
+                    name=get_op_name(output_detail['name'])
+                )
+            elif optimizing_for_openvino_and_myriad and not rigorous_optimization_for_myriad:
+                output_tensor = tf.math.abs(
+                    input_tensor1,
+                    name=get_op_name(output_detail['name'])
+                )
+            elif optimizing_for_openvino_and_myriad and rigorous_optimization_for_myriad:
+                output_tensor = tf.math.sqrt(
+                    tf.math.square(input_tensor1),
+                    name=get_op_name(output_detail['name'])
+                )
+
             tensors[output_detail['index']] = output_tensor
 
         elif op_type == 'UNIQUE':
@@ -5384,6 +5397,7 @@ def main():
     parser.add_argument('--vpu_number_of_shaves', type=int, default=4, help='vpu number of shaves. Default: 4')
     parser.add_argument('--vpu_number_of_cmx_slices', type=int, default=4, help='vpu number of cmx slices. Default: 4')
     parser.add_argument('--optimizing_for_openvino_and_myriad', action='store_true', help='Optimizing graph for openvino/myriad')
+    parser.add_argument('--rigorous_optimization_for_myriad', action='store_true', help='Replace operations that are not supported by myriad with operations that are as feasible as possible.')
     parser.add_argument('--replace_swish_and_hardswish', action='store_true', help='Replace swish and hard-swish with each other')
     parser.add_argument('--optimizing_hardswish_for_edgetpu', action='store_true', help='Optimizing hardswish for edgetpu')
     parser.add_argument('--replace_prelu_and_minmax', action='store_true', help='Replace prelu and minimum/maximum with each other')
@@ -5427,6 +5441,7 @@ def main():
     vpu_number_of_shaves = args.vpu_number_of_shaves
     vpu_number_of_cmx_slices = args.vpu_number_of_cmx_slices
     optimizing_for_openvino_and_myriad = args.optimizing_for_openvino_and_myriad
+    rigorous_optimization_for_myriad = args.rigorous_optimization_for_myriad
     replace_swish_and_hardswish = args.replace_swish_and_hardswish
     optimizing_hardswish_for_edgetpu = args.optimizing_hardswish_for_edgetpu
     replace_prelu_and_minmax = args.replace_prelu_and_minmax
@@ -5530,6 +5545,9 @@ def main():
         print(f'{Color.RED}ERROR:{Color.RESET} optimizing_for_openvino_and_myriad and optimizing_hardswish_for_edgetpu cannot be True at the same time.')
         sys.exit(-1)
 
+    if not optimizing_for_openvino_and_myriad and rigorous_optimization_for_myriad:
+        optimizing_for_openvino_and_myriad = True
+
     if tfv1_flg:
         from tensorflow.lite.python.interpreter import Interpreter as tflite_interpreter
 
@@ -5563,7 +5581,8 @@ def main():
             replace_swish_and_hardswish,
             replace_prelu_and_minmax,
             optimizing_for_edgetpu_flg,
-            optimizing_for_openvino_and_myriad)
+            optimizing_for_openvino_and_myriad,
+            rigorous_optimization_for_myriad)
         print('outputs:')
         if not TFLite_Detection_PostProcess_flg:
             for output in output_details: