[TENSORFLOW] Tensorflow 2.x support

    * Frontend supports concreate function along with graphdef.
    * New test cases added to validate TF2.x functions.
    * E2E testcases will use TFHub inputs.

python/tvm/relay/frontend/tensorflow.py

@@ -41,6 +41,7 @@
 from .common import infer_channels as _infer_channels
 from .common import infer_value as _infer_value
 
+
 __all__ = ["from_tensorflow"]
 
 
@@ -2462,6 +2463,7 @@ def _impl(inputs, attr, params, mod):
     "Round": AttrCvt("round"),
     "Rsqrt": _rsqrt(),
     "Select": _where(),
+    "SelectV2": _where(),
     "Selu": _selu(),
     "Shape": _shape(),
     "Sigmoid": AttrCvt("sigmoid"),
@@ -3656,13 +3658,58 @@ def from_tensorflow(self, graph, layout="NHWC", shape=None, outputs=None):
         return func, self._params
 
 
+# Ref. taken from Tensorflow JS
+def _build_signature_def(frozen_graph, input_nodes, output_nodes):
+    try:
+        from tensorflow.core.protobuf import meta_graph_pb2
+    except ImportError as e:
+        raise ImportError("Unable to import tensorflow which is required {}".format(e))
+    signature = meta_graph_pb2.SignatureDef()
+    for input_tensor in input_nodes:
+        op_name = input_tensor.name.split(":")[0]
+        # The graph freezing may turn the original inputs into constants, or remove
+        # them from the graph, so we need to ignore those.
+        try:
+            op = frozen_graph.get_operation_by_name(op_name)
+            if op.type != "Const":
+                signature.inputs[input_tensor.name].name = input_tensor.name
+                signature.inputs[input_tensor.name].dtype = input_tensor.dtype.as_datatype_enum
+                signature.inputs[input_tensor.name].tensor_shape.CopyFrom(
+                    input_tensor.shape.as_proto()
+                )
+        except KeyError:
+            # The original input was removed when the graph was frozen.
+            continue
+    for output_tensor in output_nodes:
+        if hasattr(output_tensor, "name"):
+            signature.outputs[output_tensor.name].name = output_tensor.name
+            signature.outputs[output_tensor.name].dtype = output_tensor.dtype.as_datatype_enum
+            signature.outputs[output_tensor.name].tensor_shape.CopyFrom(
+                output_tensor.shape.as_proto()
+            )
+        else:  # just the tensor name string array
+            signature.outputs[output_tensor].name = output_tensor
+    return signature
+
+
+def _run_grappler(config, graph_def, graph, signature_def):
+    try:
+        from tensorflow.python.grappler import tf_optimizer
+        from tensorflow.python.training.saver import export_meta_graph
+    except ImportError as e:
+        raise ImportError("Unable to import tensorflow which is required {}".format(e))
+    meta_graph = export_meta_graph(graph_def=graph_def, graph=graph)
+    meta_graph.signature_def["not_used_key"].CopyFrom(signature_def)
+    return tf_optimizer.OptimizeGraph(config, meta_graph)
+
+
 def from_tensorflow(graph, layout="NHWC", shape=None, outputs=None):
     """Load tensorflow graph which is a python tensorflow graph object into relay.
     The companion parameters will be handled automatically.
 
     Parameters
     ----------
-    graph : GraphDef object
+    graph : GraphDef object or concrete function
         Tensorflow GraphDef
 
     layout : target layout to be used (Optional)
@@ -3682,6 +3729,30 @@ def from_tensorflow(graph, layout="NHWC", shape=None, outputs=None):
     params : dict of str to tvm.nd.NDArray
         Dict of converted parameters stored in tvm.nd.NDArray format
     """
+    from tensorflow.python.eager.function import ConcreteFunction
+
+    if isinstance(graph, ConcreteFunction):
+        try:
+            from tensorflow.python.framework import convert_to_constants
+            from tensorflow.core.protobuf import config_pb2
+        except ImportError as e:
+            raise ImportError("Unable to import tensorflow which is required {}".format(e))
+        concrete_func = graph
+        graph = convert_to_constants.convert_variables_to_constants_v2(concrete_func).graph
+        signature = _build_signature_def(graph, concrete_func.inputs, concrete_func.outputs)
+        graph_def = graph.as_graph_def()
+
+        # Some optimization
+        config = config_pb2.ConfigProto()
+        rewriter_config = config.graph_options.rewrite_options
+        rewriter_config.optimizers[:] = [
+            "debug_stripper",
+            "arithmetic",
+            "dependency",
+            "arithmetic",
+            "dependency",
+        ]
+        graph = _run_grappler(config, graph_def, graph, signature)
 
     g = GraphProto()
     mod, params = g.from_tensorflow(graph, layout, shape, outputs)