Merge pull request #524 from onnx/gs/opt-relu6

optimize relu6

Author: nbcsm

README.md

@@ -6,9 +6,9 @@ tf2onnx - convert TensorFlow models to ONNX models.
 # Supported ONNX version
 tensorflow-onnx will use the ONNX version installed on your system and installs the latest ONNX version if none is found.
 
-By default we use opset 7 for the resulting ONNX graph since most runtimes will support opset 7. Opset 7 was introduced in onnx-1.2.
+We support opset 6 to 10. By default we use opset 7 for the resulting ONNX graph since most runtimes will support opset 7.
 
-Newer releases of ONNX support higher opsets. For example, to create an ONNX graph for opset 8 use in the command line ```--opset 8```.
+If you want the graph to be generated with a newer opset, use ```--opset``` in the command line, for example ```--opset 10```.
 
 # Status
 We support many TensorFlow models. Support for Fully Connected and Convolutional networks is mature. Dynamic LSTM/GRU/Attention networks should work but the code for this is evolving. 
@@ -41,7 +41,7 @@ For pytorch/caffe2, follow the instructions here:
 We tested with pytorch/caffe2 and onnxruntime and unit tests are passing for those.
 
 ## Supported Tensorflow and Python Versions
-We tested with tensorflow 1.5-1.13 and anaconda **3.5,3.6**.
+We are testing with tensorflow 1.5-1.13 and anaconda **3.5,3.6,3.7**.
 
 # Installation
 ## From pypi
@@ -55,7 +55,7 @@ python setup.py install
 or 
 python setup.py develop
 ```
-tensorflow-onnx requires onnx-1.2.2 or better and will install/upgrade onnx if needed.
+tensorflow-onnx requires onnx-1.5 or better and will install/upgrade onnx if needed.
 
 To create a distribution:
 ```
@@ -69,10 +69,10 @@ names with ```--inputs INPUTS``` and ```--outputs OUTPUTS```.
 
 ```
 python -m tf2onnx.convert 
-    --input SOURCE_GRAPHDEF_PB
-    --graphdef SOURCE_GRAPHDEF_PB
-    --checkpoint SOURCE_CHECKPOINT
-    --saved-model SOURCE_SAVED_MODEL
+    [--input SOURCE_GRAPHDEF_PB]
+    [--graphdef SOURCE_GRAPHDEF_PB]
+    [--checkpoint SOURCE_CHECKPOINT]
+    [--saved-model SOURCE_SAVED_MODEL]
     [--output TARGET_ONNX_MODEL]
     [--inputs GRAPH_INPUTS]
     [--outputs GRAPH_OUTPUS]

tests/test_backend.py

@@ -900,10 +900,9 @@ def test_tanh(self):
         _ = tf.identity(x_, name=_TFOUTPUT)
         self._run_test_case([_OUTPUT], {_INPUT: x_val}, rtol=1e-05)
 
-    @check_onnxruntime_incompatibility("Max")
     def test_relu6(self):
-        x_val = np.array([0.5, 1.0, -0.5, -1.0], dtype=np.float32).reshape((2, 2))
-        x = tf.placeholder(tf.float32, [2, 2], name=_TFINPUT)
+        x_val = np.array([0.5, 1.0, -0.5, -1.0, 6, 7], dtype=np.float32).reshape((2, 3))
+        x = tf.placeholder(tf.float32, x_val.shape, name=_TFINPUT)
         x_ = tf.nn.relu6(x)
         _ = tf.identity(x_, name=_TFOUTPUT)
         self._run_test_case([_OUTPUT], {_INPUT: x_val})

tests/test_graph.py

@@ -249,8 +249,8 @@ def test_relu6(self):
             _ = tf.identity(x_, name="output")
             g = process_tf_graph(sess.graph, opset=self.config.opset)
             self.assertEqual(
-                'digraph { input1 [op_type=Placeholder shape="[2, 3]"] Relu6 [op_type=Relu] Relu6__2 [op_type=Clip] '
-                'output [op_type=Identity] input1:0 -> Relu6 Relu6:0 -> Relu6__2 Relu6__2:0 -> output }',
+                'digraph { input1 [op_type=Placeholder shape="[2, 3]"] Relu6 [op_type=Clip] output [op_type=Identity] '
+                'input1:0 -> Relu6 Relu6:0 -> output }',
                 onnx_to_graphviz(g))
 
     def test_conv2d(self):

tf2onnx/onnx_opset/math.py

@@ -135,10 +135,10 @@ class Relu6:
     @classmethod
     def version_4(cls, ctx, node, **kwargs):
         # relu6 = min(max(features, 0), 6)
-        node.type = "Relu"
-        clip_name = utils.make_name(node.name)
-        clip_node = ctx.insert_new_node_on_output("Clip", node.output[0], name=clip_name, min=0.0, max=6.0)
-        ctx.copy_shape(node.output[0], clip_node.output[0])
+        # relu6 = min(max(features, 0), 6)
+        node.type = "Clip"
+        node.set_attr("min", 0.0)
+        node.set_attr("max", 6.0)
 
 
 @tf_op("Rsqrt")