make tf meta file temporary (#621)

* make tf meta file temporary

mmdnn/conversion/pytorch/pytorch_emitter.py

@@ -254,7 +254,7 @@ def emit_Pool(self, IR_node):
                 ceil_mode = self.is_ceil_mode(IR_node.get_attr('pads'))
 
                 # input_node = self._defuse_padding(IR_node, exstr)
-                code = "{:<15} = F.{}({}, kernel_size={}, stride={}, padding={}, ceil_mode={})".format(
+                code = "{:<15} = F.{}({}, kernel_size={}, stride={}, padding={}, ceil_mode={}, count_include_pad=False)".format(
                     IR_node.variable_name,
                     pool_name,
                     self.parent_variable_name(IR_node),

mmdnn/conversion/tensorflow/tensorflow_frozenparser.py

@@ -8,6 +8,9 @@
 from mmdnn.conversion.common.utils import *
 from mmdnn.conversion.common.DataStructure.parser import Parser
 from distutils.version import LooseVersion
+import tempfile
+import os
+import shutil
 
 
 class TensorflowParser2(Parser):
@@ -120,11 +123,14 @@ def __init__(self, frozen_file, inputshape, in_nodes, dest_nodes):
                 output_node_names = dest_nodes,
                 placeholder_type_enum = dtypes.float32.as_datatype_enum)
         # Save it to an output file
-        frozen_model_file = './frozen.pb'
+        tempdir = tempfile.mkdtemp()
+        frozen_model_file = os.path.join(tempdir, 'frozen.pb')
         with gfile.GFile(frozen_model_file, "wb") as f:
             f.write(original_gdef.SerializeToString())
         with open(frozen_model_file, 'rb') as f:
             serialized = f.read()
+        shutil.rmtree(tempdir)
+
         tensorflow.reset_default_graph()
         model = tensorflow.GraphDef()
         model.ParseFromString(serialized)
@@ -149,14 +155,15 @@ def __init__(self, frozen_file, inputshape, in_nodes, dest_nodes):
                     x = tensorflow.placeholder(dtype)
 
                 input_map[in_nodes[i] + ':0'] = x
-           
+
             tensorflow.import_graph_def(model, name='', input_map=input_map)
 
         with tensorflow.Session(graph = g) as sess:
 
-            meta_graph_def = tensorflow.train.export_meta_graph(filename='./my-model.meta')
+            tempdir = tempfile.mkdtemp()
+            meta_graph_def = tensorflow.train.export_meta_graph(filename=os.path.join(tempdir, 'my-model.meta'))
             model = meta_graph_def.graph_def
-
+            shutil.rmtree((tempdir))
 
         self.tf_graph = TensorflowGraph(model)
         self.tf_graph.build()
@@ -346,7 +353,7 @@ def _add_constant_node(self, source_node):
             parent_node = self.tf_graph.get_node(s)
             if parent_node.type == 'Const':
                 self._rename_Const(parent_node)
-    
+
     def _rename_Const(self, source_node):
         IR_node = self._convert_identity_operation(source_node, end_idx=0, new_op='Constant') # Constant
         value = source_node.get_attr('value')
@@ -369,13 +376,13 @@ def gen_IR(self):
                 continue
 
             node_type = current_node.type
-            
+
             if hasattr(self, "rename_" + node_type):
- 
+
                 func = getattr(self, "rename_" + node_type)
                 func(current_node)
             else:
-                
+
                 self.rename_UNKNOWN(current_node)
 
 
@@ -800,9 +807,9 @@ def rename_Gather(self, source_node):
         assign_IRnode_values(IR_node, kwargs)
 
         return IR_node
-    
+
     def rename_GatherV2(self, source_node):
-        
+
         IR_node = self.rename_Gather(source_node)
 
         kwargs = {}
@@ -1013,7 +1020,7 @@ def rename_Rank(self, source_node):
 
     def rename_Transpose(self, source_node):
         IR_node = self._convert_identity_operation(source_node, end_idx=1, new_op = 'Transpose')
-        
+
         input_node_perm = self.get_parent(source_node.name, [1])
         # input_node_perm = self.check_const(self.get_parent(source_node.name, [1], True))
         tensor_content = input_node_perm.get_attr('value')
@@ -1142,6 +1149,6 @@ def rename_Tanh(self, source_node):
         kwargs['shape'] = self.tensor_shape_to_list(input_node.get_attr('_output_shapes'))[0]
 
         assign_IRnode_values(IR_node, kwargs)
-    
+
     def rename_Log(self, source_node):
         IR_node = self._convert_identity_operation(source_node, new_op = 'Log')

mmdnn/conversion/tensorflow/tensorflow_parser.py

@@ -14,6 +14,9 @@
 from mmdnn.conversion.common.DataStructure.parser import Parser
 from tensorflow.tools.graph_transforms import TransformGraph
 from mmdnn.conversion.rewriter.utils import *
+import tempfile
+import os
+import shutil
 
 
 class TensorflowParser(Parser):
@@ -308,9 +311,10 @@ def __init__(self, meta_file, checkpoint_file, dest_nodes, inputShape = None, in
             tensorflow.import_graph_def(transformed_graph_def, name='', input_map=input_map)
 
         with tensorflow.Session(graph = g) as sess:
-
-            meta_graph_def = tensorflow.train.export_meta_graph(filename='./my-model.meta')
+            tempdir = tempfile.mkdtemp()
+            meta_graph_def = tensorflow.train.export_meta_graph(filename=os.path.join(tempdir, 'my-model.meta'))
             model = meta_graph_def.graph_def
+            shutil.rmtree(tempdir)
 
         self.tf_graph = TensorflowGraph(model)
         self.tf_graph.build()

tests/conversion_imagenet.py

@@ -691,12 +691,6 @@ def onnx_emit(original_framework, architecture_name, architecture_path, weight_p
 
             predict = tf_rep.run(input_data)[0]
 
-            return predict
-
-        except ImportError:
-            print('Please install Onnx! Or Onnx is not supported in your platform.', file=sys.stderr)
-
-        finally:
             del prepare
             del model_converted
             del tf_rep
@@ -705,6 +699,10 @@ def onnx_emit(original_framework, architecture_name, architecture_path, weight_p
             os.remove(converted_file + '.py')
             os.remove(converted_file + '.npy')
 
+            return predict
+
+        except ImportError:
+            print('Please install Onnx! Or Onnx is not supported in your platform.', file=sys.stderr)
 
 
     # In case of odd number add the extra padding at the end for SAME_UPPER(eg. pads:[0, 2, 2, 0, 0, 3, 3, 0]) and at the beginning for SAME_LOWER(eg. pads:[0, 3, 3, 0, 0, 2, 2, 0])