2021-08-10 onnxruntime shape [] != None#
None is the undefined shape, [] is an empty shape. And when shapes do not fit the results, the outputs can be suprising. The following example shows what onnxruntime produces for the same graph except input and output shapes when defined as None and [].
<<<
import numpy
from onnx import helper, TensorProto
from onnxruntime import InferenceSession
def model(shape):
X = helper.make_tensor_value_info('X', TensorProto.FLOAT, shape)
Z = helper.make_tensor_value_info('Z', TensorProto.INT64, shape)
node_def = helper.make_node('Shape', ['X'], ['Z'], name='Zt')
graph_def = helper.make_graph([node_def], 'test-model', [X], [Z])
model_def = helper.make_model(
graph_def, producer_name='mlprodict', ir_version=7, producer_version='0.1',
opset_imports=[helper.make_operatorsetid('', 13)])
sess = InferenceSession(model_def.SerializeToString())
rnd = numpy.random.randn(12).astype(numpy.float32)
print("shape=%r results=%r" % (shape, sess.run(None, {"X": rnd})))
model(None)
model([])
>>>
shape=None results=[array([12], dtype=int64)]
shape=[] results=[array([], dtype=int64)]