Add documentation, fix UT.

Author: jignparm

README.md

@@ -139,6 +139,9 @@ python -m tf2onnx.convert
     [--outputs GRAPH_OUTPUS]
     [--inputs-as-nchw inputs_provided_as_nchw]
     [--opset OPSET]
+    [--tag TAG]
+    [--signature_def SIGNATURE_DEF]
+    [--concrete_function CONCRETE_FUNCTION]
     [--target TARGET]
     [--custom-ops list-of-custom-ops]
     [--fold_const]
@@ -176,6 +179,20 @@ By default we preserve the image format of inputs (`nchw` or `nhwc`) as given in
 
 By default we use the opset 8 to generate the graph. By specifying ```--opset``` the user can override the default to generate a graph with the desired opset. For example ```--opset 5``` would create a onnx graph that uses only ops available in opset 5. Because older opsets have in most cases fewer ops, some models might not convert on a older opset.
 
+#### --tag
+
+Only valid with parameter `--saved_model`. Specifies the tag in the saved_model to be used. Typical value is 'serve'.
+
+#### --signature_def
+
+Only valid with parameter `--saved_model`. Specifies which signature to use within the specified --tag value. Typical value is 'serving_default'.
+
+#### --concrete_function
+
+(This is experimental, valid only for TF2.x models)
+
+Only valid with parameter `--saved_model`. If a model contains a list of concrete functions, under the function name `__call__` (as can be viewed using the command `saved_model_cli show --all`), this parameter is a 0-based integer specifying which function in that list should be converted. This parameter takes priority over `--signature_def`, which will be ignored.
+
 #### --target
 
 Some models require special handling to run on some runtimes. In particular, the model may use unsupported data types. Workarounds are activated with ```--target TARGET```. Currently supported values are listed on this [wiki](https://github.com/onnx/tensorflow-onnx/wiki/target). If your model will be run on Windows ML, you should specify the appropriate target value.

tests/run_pretrained_models.py

@@ -79,13 +79,18 @@ def get_ones(shape):
     """Get ones."""
     return np.ones(shape).astype(np.float32)
 
+def get_zeros(shape):
+    """Get zeros."""
+    return np.zeros(shape).astype(np.float32)
+
 
 _INPUT_FUNC_MAPPING = {
     "get_beach": get_beach,
     "get_random": get_random,
     "get_random256": get_random256,
     "get_ramp": get_ramp,
-    "get_ones": get_ones
+    "get_ones": get_ones,
+    "get_zeros": get_zeros,
 }
 
 OpsetConstraint = namedtuple("OpsetConstraint", "domain, min_version, max_version, excluded_version")
@@ -100,7 +105,7 @@ class Test(object):
     def __init__(self, url, local, make_input, input_names, output_names,
                  disabled=False, rtol=0.01, atol=1e-6,
                  check_only_shape=False, model_type="frozen", force_input_shape=False,
-                 skip_tensorflow=False, opset_constraints=None, tf_min_version=None):
+                 skip_tensorflow=False, opset_constraints=None, tf_min_version=None, tag=None):
         self.url = url
         self.make_input = make_input
         self.local = local
@@ -114,6 +119,7 @@ def __init__(self, url, local, make_input, input_names, output_names,
         self.tf_runtime = 0
         self.onnx_runtime = 0
         self.model_type = model_type
+        self.tag = tag
         self.force_input_shape = force_input_shape
         self.skip_tensorflow = skip_tensorflow
         self.opset_constraints = opset_constraints
@@ -240,7 +246,7 @@ def run_test(self, name, backend="caffe2", onnx_file=None, opset=None, extra_ops
         if self.model_type in ["checkpoint"]:
             graph_def, input_names, outputs = tf_loader.from_checkpoint(model_path, input_names, outputs)
         elif self.model_type in ["saved_model"]:
-            graph_def, input_names, outputs = tf_loader.from_saved_model(model_path, input_names, outputs)
+            graph_def, input_names, outputs = tf_loader.from_saved_model(model_path, input_names, outputs, self.tag)
         elif self.model_type in ["keras"]:
             graph_def, input_names, outputs = tf_loader.from_keras(model_path, input_names, outputs)
         else:
@@ -436,7 +442,7 @@ def load_tests_from_yaml(path):
 
         kwargs = {}
         for kw in ["rtol", "atol", "disabled", "check_only_shape", "model_type",
-                   "skip_tensorflow", "force_input_shape", "tf_min_version"]:
+                   "skip_tensorflow", "force_input_shape", "tf_min_version", "tag"]:
             if settings.get(kw) is not None:
                 kwargs[kw] = settings[kw]
 

tests/run_pretrained_models.yaml

@@ -21,6 +21,7 @@ regression-checkpoint:
 regression-saved-model:
   model: models/regression/saved_model
   model_type: saved_model
+  tag: serve
   input_get: get_ramp
   inputs:
     "X:0": [1]
@@ -239,9 +240,10 @@ vgg-16:
 
 resnet50_v2_nchw: # NOTE: Tensorflow 1.9.0 fails
   skip_tensorflow: true # tensorflow fails: Default MaxPoolingOp only supports NHWC on device type CPU
-  model_type: saved_model
   url: http://download.tensorflow.org/models/official/20181001_resnet/savedmodels/resnet_v2_fp32_savedmodel_NCHW.tar.gz
   model: resnet_v2_fp32_savedmodel_NCHW/1538687196
+  model_type: saved_model
+  tag: serve
   input_get: get_beach
   inputs:
     "input_tensor:0": [64, 224, 224, 3]
@@ -250,9 +252,10 @@ resnet50_v2_nchw: # NOTE: Tensorflow 1.9.0 fails
     - softmax_tensor:0
 
 resnet50_v2_nhwc:
-  model_type: saved_model
   url: http://download.tensorflow.org/models/official/20181001_resnet/savedmodels/resnet_v2_fp32_savedmodel_NHWC.tar.gz
   model: resnet_v2_fp32_savedmodel_NHWC/1538687283
+  model_type: saved_model
+  tag: serve
   input_get: get_beach
   inputs:
     "input_tensor:0": [64, 224, 224, 3]

tf2onnx/tf_loader.py

@@ -184,6 +184,9 @@ def _from_saved_model_v1(sess, model_path, input_names, output_names, tag, signa
     if tag is None:
         tag = [tf.saved_model.tag_constants.SERVING]
 
+    if not isinstance(tag, list):
+        tag = [tag]
+
     imported = tf.saved_model.loader.load(sess, tag, model_path)
     for k in imported.signature_def.keys():
         if k.startswith("_"):