Fixed bugs in parameter replacement logic and Flatten attribute repla…

…cement timing

README.md

@@ -71,7 +71,7 @@ Video speed is adjusted approximately 50 times slower than actual speed.
 $ docker run --rm -it \
 -v `pwd`:/workdir \
 -w /workdir \
-ghcr.io/pinto0309/onnx2tf:1.0.27
+ghcr.io/pinto0309/onnx2tf:1.0.28
 
 or
 

onnx2tf/__init__.py

@@ -1,3 +1,3 @@
 from onnx2tf.onnx2tf import convert, main
 
-__version__ = '1.0.27'
+__version__ = '1.0.28'

onnx2tf/ops/Flatten.py

@@ -68,6 +68,14 @@ def make_node(
         'dtype': dtype,
     }
 
+    # Param replacement
+    axis = replace_parameter(
+        value_before_replacement=axis,
+        param_target='attributes',
+        param_name='axis',
+        **kwargs,
+    )
+
     # Generation of TF OP
     cal_shape = None
     if axis == 0:
@@ -85,12 +93,6 @@ def make_node(
         param_name=graph_node.inputs[0].name,
         **kwargs,
     )
-    cal_shape = replace_parameter(
-        value_before_replacement=cal_shape,
-        param_target='attributes',
-        param_name='axis',
-        **kwargs,
-    )
 
     # Pre-process transpose
     input_tensor = pre_process_transpose(

onnx2tf/utils/common_functions.py

@@ -57,7 +57,8 @@ def replace_parameter(
     op_rep_params = kwargs.get('op_rep_params', [])
     for op_rep_param in op_rep_params:
         if op_rep_param['param_target'] == param_target \
-            and op_rep_param['param_name'] == param_name:
+            and op_rep_param['param_name'] == param_name \
+            and 'values' in op_rep_param:
             replace_value = op_rep_param.get('values', value_before_replacement)
             if isinstance(value_before_replacement, np.ndarray):
                 replace_value = np.asarray(