Alexander Shchepetovandalexander.shchepetov
b0c508d4ff
Co-authored-by: alexander.shchepetov <ashchepe@nnlvdp-cflr51.inn.intel.com>
308 lines
11 KiB
Python
308 lines
11 KiB
Python
# Copyright (C) 2018-2021 Intel Corporation
|
|
# SPDX-License-Identifier: Apache-2.0
|
|
|
|
import pytest
|
|
|
|
from common.onnx_layer_test_class import Caffe2OnnxLayerTest
|
|
|
|
|
|
test_data_3D = [
|
|
dict(input_shape=[1, 50, 50], output_shapes=[[1, 50, 25], [1, 50, 25]], axis=2),
|
|
dict(input_shape=[2, 50, 50], output_shapes=[[2, 20, 50], [2, 15, 50], [2, 15, 50]], axis=1),
|
|
dict(input_shape=[4, 50, 50], output_shapes=[[1, 50, 50], [1, 50, 50], [1, 50, 50], [1, 50, 50]], axis=0)]
|
|
|
|
test_data_4D = [
|
|
dict(input_shape=[1, 32, 800, 800], output_shapes=[[1, 16, 800, 800], [1, 16, 800, 800]], axis=1),
|
|
dict(input_shape=[4, 32, 80, 80], output_shapes=[[4, 8, 80, 80], [4, 8, 80, 80], [4, 8, 80, 80], [4, 8, 80, 80]],
|
|
axis=1),
|
|
dict(input_shape=[2, 21, 80, 80], output_shapes=[[2, 7, 80, 80], [2, 7, 80, 80], [2, 7, 80, 80]], axis=1),
|
|
dict(input_shape=[3, 21, 80, 80], output_shapes=[[3, 14, 80, 80], [3, 5, 80, 80], [3, 2, 80, 80]], axis=1),
|
|
dict(input_shape=[3, 21, 80, 80], output_shapes=[[1, 21, 80, 80], [1, 21, 80, 80], [1, 21, 80, 80]], axis=0),
|
|
dict(input_shape=[3, 21, 80, 80], output_shapes=[[3, 21, 20, 80], [3, 21, 35, 80], [3, 21, 25, 80]], axis=2),
|
|
dict(input_shape=[3, 21, 80, 80], output_shapes=[[3, 21, 80, 40], [3, 21, 80, 10], [3, 21, 80, 30]], axis=3)]
|
|
|
|
test_data_5D = [
|
|
dict(input_shape=[1, 50, 50, 80, 60],
|
|
output_shapes=[[1, 50, 10, 80, 60],
|
|
[1, 50, 10, 80, 60],
|
|
[1, 50, 10, 80, 60],
|
|
[1, 50, 10, 80, 60],
|
|
[1, 50, 10, 80, 60]], axis=2),
|
|
dict(input_shape=[1, 50, 50, 80, 60], output_shapes=[[1, 25, 50, 80, 60], [1, 25, 50, 80, 60]], axis=1)]
|
|
|
|
|
|
class TestSplitConcat(Caffe2OnnxLayerTest):
|
|
# TODO Add test with default values (axis=0)
|
|
def create_split_concat_net(self, input_shape, output_shapes, axis, ir_version):
|
|
"""
|
|
ONNX net IR net
|
|
|
|
Input->Split->Concat->Output => Input->Split->Concat
|
|
|
|
"""
|
|
|
|
#
|
|
# Create ONNX model
|
|
#
|
|
|
|
import onnx
|
|
from onnx import helper
|
|
from onnx import TensorProto
|
|
|
|
input = helper.make_tensor_value_info('input', TensorProto.FLOAT, input_shape)
|
|
outputs, split = [], []
|
|
for id, output_shape in enumerate(output_shapes):
|
|
helper.make_tensor_value_info('output_{}'.format(id), TensorProto.FLOAT, output_shape)
|
|
outputs.append('output_{}'.format(id))
|
|
split.append(output_shape[axis])
|
|
|
|
# Output for concat
|
|
output_concat = helper.make_tensor_value_info('output_concat', TensorProto.FLOAT, input_shape)
|
|
|
|
node_split_def = onnx.helper.make_node(
|
|
'Split',
|
|
inputs=['input'],
|
|
outputs=outputs,
|
|
axis=axis,
|
|
split=split
|
|
)
|
|
|
|
node_concat_def = onnx.helper.make_node(
|
|
'Concat',
|
|
inputs=outputs,
|
|
outputs=['output_concat'],
|
|
axis=axis
|
|
)
|
|
|
|
# Create the graph (GraphProto)
|
|
graph_def = helper.make_graph(
|
|
[node_split_def, node_concat_def],
|
|
'test_split_model',
|
|
[input],
|
|
[output_concat],
|
|
)
|
|
|
|
# Create the model (ModelProto)
|
|
onnx_net = helper.make_model(graph_def, producer_name='test_split_model')
|
|
|
|
#
|
|
# Create reference IR net
|
|
# Please, spesify 'type': 'Input' for inpit node
|
|
# Moreover, do not forget to validate ALL layer attributes!!!
|
|
#
|
|
|
|
ref_net = None
|
|
|
|
return onnx_net, ref_net
|
|
|
|
# TODO Add test with default values (axis=0)
|
|
def create_split_concat_net_const(self, input_shape, output_shapes, axis, ir_version):
|
|
"""
|
|
ONNX net IR net
|
|
|
|
Input(const)->Split->Concat--->Concat->Output => Input--->Concat
|
|
Input-' Const-'
|
|
|
|
"""
|
|
|
|
#
|
|
# Create ONNX model
|
|
#
|
|
|
|
import onnx
|
|
from onnx import helper
|
|
from onnx import TensorProto
|
|
import numpy as np
|
|
|
|
concat_axis = 0
|
|
concat_output_shape = input_shape.copy()
|
|
concat_output_shape[concat_axis] *= 2
|
|
|
|
const_number = np.prod(input_shape)
|
|
constant = np.random.randint(-127, 127, const_number).astype(np.float)
|
|
|
|
input = helper.make_tensor_value_info('input', TensorProto.FLOAT, input_shape)
|
|
outputs, split = [], []
|
|
for id, output_shape in enumerate(output_shapes):
|
|
helper.make_tensor_value_info('output_{}'.format(id), TensorProto.FLOAT, output_shape)
|
|
outputs.append('output_{}'.format(id))
|
|
split.append(output_shape[axis])
|
|
|
|
# Output for concat
|
|
output_concat = helper.make_tensor_value_info('output_dyn_concat', TensorProto.FLOAT, concat_output_shape)
|
|
|
|
node_const_def = onnx.helper.make_node(
|
|
'Constant',
|
|
inputs=[],
|
|
outputs=['const1'],
|
|
value=helper.make_tensor(
|
|
name='const_tensor',
|
|
data_type=TensorProto.FLOAT,
|
|
dims=input_shape,
|
|
vals=constant,
|
|
),
|
|
)
|
|
|
|
node_split_def = onnx.helper.make_node(
|
|
'Split',
|
|
inputs=['const1'],
|
|
outputs=outputs,
|
|
axis=axis,
|
|
split=split
|
|
)
|
|
|
|
node_concat_def = onnx.helper.make_node(
|
|
'Concat',
|
|
inputs=outputs,
|
|
outputs=['output_concat'],
|
|
axis=axis
|
|
)
|
|
|
|
node_dyn_concat_def = onnx.helper.make_node(
|
|
'Concat',
|
|
inputs=['input', 'output_concat'],
|
|
outputs=['output_dyn_concat'],
|
|
axis=concat_axis
|
|
)
|
|
|
|
# Create the graph (GraphProto)
|
|
graph_def = helper.make_graph(
|
|
[node_const_def, node_split_def, node_concat_def, node_dyn_concat_def],
|
|
'test_split_model',
|
|
[input],
|
|
[output_concat],
|
|
)
|
|
|
|
# Create the model (ModelProto)
|
|
onnx_net = helper.make_model(graph_def, producer_name='test_split_model')
|
|
|
|
#
|
|
# Create reference IR net
|
|
# Please, spesify 'type': 'Input' for inpit node
|
|
# Moreover, do not forget to validate ALL layer attributes!!!
|
|
#
|
|
|
|
ref_net = None
|
|
|
|
return onnx_net, ref_net
|
|
|
|
@pytest.mark.parametrize("params", test_data_3D)
|
|
@pytest.mark.nightly
|
|
def test_split_3D(self, params, ie_device, precision, ir_version, temp_dir):
|
|
self._test(*self.create_split_concat_net(**params, ir_version=ir_version), ie_device, precision, ir_version,
|
|
temp_dir=temp_dir)
|
|
|
|
@pytest.mark.parametrize("params", test_data_4D)
|
|
@pytest.mark.nightly
|
|
def test_split_4D(self, params, ie_device, precision, ir_version, temp_dir):
|
|
self._test(*self.create_split_concat_net(**params, ir_version=ir_version), ie_device, precision, ir_version,
|
|
temp_dir=temp_dir)
|
|
|
|
@pytest.mark.parametrize("params", test_data_5D)
|
|
@pytest.mark.nightly
|
|
def test_split_5D(self, params, ie_device, precision, ir_version, temp_dir):
|
|
self._test(*self.create_split_concat_net(**params, ir_version=ir_version), ie_device, precision, ir_version,
|
|
temp_dir=temp_dir)
|
|
|
|
@pytest.mark.parametrize("params", test_data_3D)
|
|
@pytest.mark.nightly
|
|
def test_split_3D_const(self, params, ie_device, precision, ir_version, temp_dir):
|
|
self._test(
|
|
*self.create_split_concat_net_const(**params, ir_version=ir_version), ie_device, precision, ir_version,
|
|
temp_dir=temp_dir)
|
|
|
|
@pytest.mark.parametrize("params", test_data_4D)
|
|
@pytest.mark.nightly
|
|
def test_split_4D_const(self, params, ie_device, precision, ir_version, temp_dir):
|
|
self._test(
|
|
*self.create_split_concat_net_const(**params, ir_version=ir_version), ie_device, precision, ir_version,
|
|
temp_dir=temp_dir)
|
|
|
|
@pytest.mark.parametrize("params", test_data_5D)
|
|
@pytest.mark.nightly
|
|
def test_split_5D_const(self, params, ie_device, precision, ir_version, temp_dir):
|
|
self._test(
|
|
*self.create_split_concat_net_const(**params, ir_version=ir_version), ie_device, precision, ir_version,
|
|
temp_dir=temp_dir)
|
|
|
|
|
|
class TestSplit(Caffe2OnnxLayerTest):
|
|
# TODO Add test with default values (axis=0)
|
|
def create_split_net(self, input_shape, output_shapes, axis, ir_version):
|
|
"""
|
|
ONNX net IR net
|
|
|
|
Input->Split->Output => Input->Split
|
|
|
|
"""
|
|
|
|
#
|
|
# Create ONNX model
|
|
#
|
|
|
|
import onnx
|
|
from onnx import helper
|
|
from onnx import TensorProto
|
|
|
|
input = helper.make_tensor_value_info('input', TensorProto.FLOAT, input_shape)
|
|
outputs, split = [], []
|
|
for id, output_shape in enumerate(output_shapes):
|
|
out = helper.make_tensor_value_info('output_{}'.format(id), TensorProto.FLOAT, output_shape)
|
|
outputs.append((out, 'output_{}'.format(id)))
|
|
split.append(output_shape[axis])
|
|
|
|
node_split_def = onnx.helper.make_node(
|
|
'Split',
|
|
inputs=['input'],
|
|
outputs=['node_{}'.format(x[1]) for x in outputs],
|
|
axis=axis,
|
|
split=split
|
|
)
|
|
nodes = [node_split_def]
|
|
|
|
for x in outputs:
|
|
nodes.append(onnx.helper.make_node(
|
|
'Elu',
|
|
inputs=['node_{}'.format(x[1])],
|
|
outputs=[x[1]]
|
|
))
|
|
|
|
# Create the graph (GraphProto)
|
|
graph_def = helper.make_graph(
|
|
nodes,
|
|
'test_split_model',
|
|
[input],
|
|
[x[0] for x in outputs],
|
|
)
|
|
|
|
# Create the model (ModelProto)
|
|
onnx_net = helper.make_model(graph_def, producer_name='test_split_model')
|
|
|
|
#
|
|
# Create reference IR net
|
|
# Please, spesify 'type': 'Input' for inpit node
|
|
# Moreover, do not forget to validate ALL layer attributes!!!
|
|
#
|
|
|
|
ref_net = None
|
|
|
|
return onnx_net, ref_net
|
|
|
|
@pytest.mark.parametrize("params", test_data_3D)
|
|
@pytest.mark.nightly
|
|
def test_split_3D(self, params, ie_device, precision, ir_version, temp_dir):
|
|
self._test(*self.create_split_net(**params, ir_version=ir_version), ie_device, precision, ir_version,
|
|
temp_dir=temp_dir)
|
|
|
|
@pytest.mark.parametrize("params", test_data_4D)
|
|
@pytest.mark.nightly
|
|
def test_split_4D(self, params, ie_device, precision, ir_version, temp_dir):
|
|
self._test(*self.create_split_net(**params, ir_version=ir_version), ie_device, precision, ir_version,
|
|
temp_dir=temp_dir)
|
|
|
|
@pytest.mark.parametrize("params", test_data_5D)
|
|
@pytest.mark.nightly
|
|
def test_split_5D(self, params, ie_device, precision, ir_version, temp_dir):
|
|
self._test(*self.create_split_net(**params, ir_version=ir_version), ie_device, precision, ir_version,
|
|
temp_dir=temp_dir)
|