IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

OVC cleanup. (#18649)

Browse Source 
* WIP: parameters cleanup

* Removed debug output, fixed CLI

* Fixed python objects conversion

* Finally renamed mmap to share_weights

* Fixed TF conversion from a file or a directory

* Fixed obvious errors in unit tests

* Deleted layouts from OVC. Fixed most of the fails in ovc unit tests (there are still failures)

* Clenaup other references to layouts and fixed --version

* Fixed case when two model files are passed in TF case

* Fixed multiple model parts passing in ovc command line

* Tests fixed, support of unnamed input in cli parser.

* Remove convert_model from runtime.

* Changed silent to verbose.

* Removed transform param.

* Removed example_input, share_weights from ovc cli tool.

* Remove wrong change.

* Test fix.

* Code corrections.

* Returned comment.

* WA to fix process hanging after extension loading.

* Removed not needed code.

* Added comment.

---------

Co-authored-by: Sergey Lyalin <sergey.lyalin@intel.com>
This commit is contained in:
Anastasiia Pnevskaia 2023-07-20 22:11:28 +02:00 committed by GitHub
parent 1ce744a00f
commit 8d5a0b1d53
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
42 changed files with 563 additions and 4278 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
src/bindings/python/src/openvino/runtime/__init__.py
View File

@ -68,13 +68,6 @@ from openvino.runtime.ie_api import tensor_from_file
from openvino.runtime.ie_api import compile_model
# Model Conversion API
try:
from openvino.tools.ovc import convert_model, InputCutInfo, LayoutMap
except ImportError:
pass
# Extend Node class to support binary operators
Node.__add__ = opset12.add
Node.__sub__ = opset12.subtract

8
tests/layer_tests/common/mo_convert_test_class.py
View File

@ -3,7 +3,8 @@
from pathlib import Path
from openvino.runtime import serialize, convert_model
from openvino.runtime import serialize
from openvino.tools.ovc import convert_model
from openvino.tools.mo import convert_model as legacy_convert_model
from openvino.test_utils import compare_functions
@ -16,7 +17,10 @@ class CommonMOConvertTest:
output_dir = kwargs['output_dir']
model_name = kwargs['model_name']
del kwargs['output_dir']
if 'use_legacy_frontend' in kwargs:
del kwargs['model_name']
if 'use_legacy_frontend' in kwargs or 'use_convert_model_from_mo' in kwargs:
if 'use_convert_model_from_mo' in kwargs:
del kwargs['use_convert_model_from_mo']
model = legacy_convert_model(**kwargs)
else:
model = convert_model(**kwargs)

15
tests/layer_tests/common/utils/common_utils.py
View File

@ -7,14 +7,13 @@ import shutil
import subprocess
import sys
from pathlib import Path
import numpy as np
from openvino.tools.mo import mo
logger = logging.getLogger(__name__)
def generate_ir(coverage=False, **kwargs):
from openvino.tools.mo import mo
mo_path = Path(mo.__file__).parent
mo_runner = mo_path.joinpath('main.py').as_posix()
if coverage:
@ -48,15 +47,13 @@ def generate_ir(coverage=False, **kwargs):
def generate_ir_python_api(coverage=False, **kwargs):
from openvino.runtime import convert_model, serialize
from openvino.tools.mo import convert_model as legacy_convert_model
if "use_legacy_frontend" in kwargs and kwargs['use_legacy_frontend']:
ov_model = legacy_convert_model(**kwargs)
else:
ov_model = convert_model(**kwargs)
from openvino.runtime import serialize
from openvino.tools.mo import convert_model
out_dir = kwargs['output_dir'] + os.sep + kwargs['model_name'] + ".xml"
# TODO: Remove usage of legacy params from layer tests and switch to convert_model from tools.ovc
ov_model = convert_model(**kwargs)
serialize(ov_model, out_dir)
return 0, ""

2
tests/layer_tests/mo_python_api_tests/mo_convert_help.py
View File

@ -1,7 +1,7 @@
# Copyright (C) 2018-2023 Intel Corporation
# SPDX-License-Identifier: Apache-2.0
from openvino.runtime import convert_model
from openvino.tools.mo import convert_model
if __name__ == "__main__":
convert_model(help=True)

65
tests/layer_tests/mo_python_api_tests/test_mo_convert_complex_params.py
View File

@ -4,7 +4,9 @@
import numpy as np
import os
import pytest
from openvino.runtime import Model, Layout, PartialShape, Shape, layout_helpers, Type, Dimension, InputCutInfo, LayoutMap
from openvino.runtime import Model, Layout, PartialShape, Shape, layout_helpers, Type, Dimension
from openvino.tools.ovc import InputCutInfo
from openvino.tools.mo import LayoutMap
from common.mo_convert_test_class import CommonMOConvertTest
from common.tf_layer_test_class import save_to_pb
@ -132,11 +134,11 @@ class TestComplexParams(CommonMOConvertTest):
{'params_test': {'input_shape': [PartialShape([2, 3, 4]),
[2, 3, 4],
[Dimension(2), Dimension(3), Dimension(4)]],
'input':['Input1', 'Input2', 'Relu3']},
'input':['Input1', 'Input2', 'Relu3'], 'use_convert_model_from_mo': True},
'params_ref': {'input_shape': "[2,3,4],[2,3,4],[2,3,4]", 'input': 'Input1,Input2,Relu3'}},
{'params_test': {'input_shape': [PartialShape([Dimension(), Dimension(1, 3), Dimension(4, -1), Dimension(-1, 5)]),
[Dimension(), Dimension(1, 3), 4, Dimension(-1, 5)],
[Dimension(), 3, Dimension(4, -1), Dimension(-1, 5)]],
[Dimension(), 3, Dimension(4, -1), Dimension(-1, 5)]], 'use_convert_model_from_mo': True,
'input':['Input1', 'Input2', 'Relu3']},
'params_ref': {'input_shape': "[?,1..3,4..,..5],[?,1..3,4,..5],[?,3,4..,..5]", 'input': 'Input1,Input2,Relu3'}},
{'params_test': {'input': [InputCutInfo("Relu1", Shape([3, 2]), Type(np.int32)),
@ -149,26 +151,28 @@ class TestComplexParams(CommonMOConvertTest):
'params_ref': {'input': "Relu1[3 2]{i32},Relu2[3..10 2..]{i32},Relu3[3 2]{i32}"}},
{'params_test': {'output': ["Sigmoid_0", "Sigmoid_2"]},
'params_ref': {'output': "Sigmoid_0,Sigmoid_2"}},
{'params_test': {'mean_values': {'Input1': [0.5,1.3,0.67], 'Input2':[4.2, 6.7, 3.15], 'Input3':[0.757, 4.6, 7.3]}},
{'params_test': {'mean_values': {'Input1': [0.5,1.3,0.67], 'Input2':[4.2, 6.7, 3.15], 'Input3':[0.757, 4.6, 7.3]},
'use_convert_model_from_mo': True},
'params_ref': {'mean_values': "Input1[0.5,1.3,0.67],Input2[4.2,6.7,3.15],Input3[0.757,4.6,7.3]"}},
{'params_test': {
'mean_values': [[0.5, 1.3, 0.67], [4.2, 6.7, 3.15], [0.757, 4.6, 7.3]]},
'mean_values': [[0.5, 1.3, 0.67], [4.2, 6.7, 3.15], [0.757, 4.6, 7.3]], 'use_convert_model_from_mo': True},
'params_ref': {'mean_values': "[0.5,1.3,0.67],[4.2,6.7,3.15],[0.757,4.6,7.3]"}},
{'params_test': {'scale_values': {'Input1': [0.5,1.3,0.67], 'Input2':[4.2, 6.7, 3.15], 'Input3':[0.757, 4.6, 7.3]}},
{'params_test': {'scale_values': {'Input1': [0.5,1.3,0.67], 'Input2':[4.2, 6.7, 3.15], 'Input3':[0.757, 4.6, 7.3]},
'use_convert_model_from_mo': True},
'params_ref': {'scale_values': "Input1[0.5,1.3,0.67],Input2[4.2,6.7,3.15],Input3[0.757,4.6,7.3]"}},
{'params_test': {
'scale_values': [[0.5, 1.3, 0.67], [4.2, 6.7, 3.15], [0.757, 4.6, 7.3]]},
'scale_values': [[0.5, 1.3, 0.67], [4.2, 6.7, 3.15], [0.757, 4.6, 7.3]], 'use_convert_model_from_mo': True},
'params_ref': {'scale_values': "[0.5,1.3,0.67],[4.2,6.7,3.15],[0.757,4.6,7.3]"}},
{'params_test': {
'source_layout': {'Input1': Layout("nchw"), 'Input2': "nchw", 'Input3': "nc??"}},
'source_layout': {'Input1': Layout("nchw"), 'Input2': "nchw", 'Input3': "nc??"}, 'use_convert_model_from_mo': True},
'params_ref': {'source_layout': "Input1(nchw),Input2(nchw),Input3(nc??)"}},
{'params_test': {
'target_layout': {'Input1': Layout("nhwc"), 'Input2': "nhwc", 'Input3': "n??c"}},
'target_layout': {'Input1': Layout("nhwc"), 'Input2': "nhwc", 'Input3': "n??c"}, 'use_convert_model_from_mo': True},
'params_ref': {'target_layout': "Input1(nhwc),Input2(nhwc),Input3(n??c)"}},
{'params_test': {
'layout': {'Input1': LayoutMap(source_layout=Layout("nchw"), target_layout="nhwc"),
'Input2': LayoutMap(source_layout="nc??", target_layout=Layout("n??c")),
'Input3': LayoutMap(source_layout="abcd", target_layout="acdb")}},
'Input3': LayoutMap(source_layout="abcd", target_layout="acdb")}, 'use_convert_model_from_mo': True},
'params_ref': {'layout': "Input1(nchw->nhwc),Input2(nc??->n??c),Input3(abcd->acdb)"}},
{'params_test': {'input': [PartialShape([2, 3, 4]), [2, 3, 4], [Dimension(2), Dimension(3), Dimension(4)]]},
'params_ref': {'input_shape': "[2,3,4],[2,3,4],[2,3,4]", 'input': 'Input1,Input2,Input3'}},
@ -222,13 +226,14 @@ class TestComplexParams(CommonMOConvertTest):
test_params = params['params_test']
ref_params = params['params_ref']
test_params.update({'input_model': tf_net_path})
test_params.update({'use_convert_model_from_mo': True})
ref_params.update({'input_model': tf_net_path})
self._test(temp_dir, test_params, ref_params)
test_data = [
{'params_test': {'input_shape': PartialShape([2, 3, 4])},
{'params_test': {'input_shape': PartialShape([2, 3, 4]), 'use_convert_model_from_mo': True},
'params_ref': {'input_shape': "[2,3,4]"}},
{'params_test': {'input_shape': [Dimension(), Dimension(1, 3), 4, Dimension(-1, 5)]},
{'params_test': {'input_shape': [Dimension(), Dimension(1, 3), 4, Dimension(-1, 5)], 'use_convert_model_from_mo': True},
'params_ref': {'input_shape': "[?,1..3,4,..5]"}},
{'params_test': {'input': InputCutInfo("Relu", [3, 2], Type(np.int32), [1, 2, 3, 4, 5, 6])},
'params_ref': {'input': "Relu[3 2]{i32}->[1 2 3 4 5 6]"}},
@ -240,17 +245,17 @@ class TestComplexParams(CommonMOConvertTest):
'params_ref': {'input': "Relu[3 2]"}},
{'params_test': {'input': ("Relu")},
'params_ref': {'input': "Relu"}},
{'params_test': {'mean_values': [0.5, 1.3, 0.67]},
{'params_test': {'mean_values': [0.5, 1.3, 0.67], 'use_convert_model_from_mo': True},
'params_ref': {'mean_values': "[0.5,1.3,0.67]"}},
{'params_test': {'scale_values': [0.5, 1.3, 0.67]},
{'params_test': {'scale_values': [0.5, 1.3, 0.67], 'use_convert_model_from_mo': True},
'params_ref': {'scale_values': "[0.5,1.3,0.67]"}},
{'params_test': {'source_layout': Layout("nchw")},
{'params_test': {'source_layout': Layout("nchw"), 'use_convert_model_from_mo': True},
'params_ref': {'source_layout': "nchw"}},
{'params_test': {'target_layout': Layout("nchw")},
{'params_test': {'target_layout': Layout("nchw"), 'use_convert_model_from_mo': True},
'params_ref': {'target_layout': "nchw"}},
{'params_test': {'layout': LayoutMap(source_layout=Layout("nchw"), target_layout="nhwc")},
{'params_test': {'layout': LayoutMap(source_layout=Layout("nchw"), target_layout="nhwc"), 'use_convert_model_from_mo': True},
'params_ref': {'layout': "nchw->nhwc"}},
{'params_test': {'layout': Layout("nchw")},
{'params_test': {'layout': Layout("nchw"), 'use_convert_model_from_mo': True},
'params_ref': {'layout': "nchw"}},
{'params_test': {'input': [3, 2]},
'params_ref': {'input': "Input[3 2]"}},
@ -266,13 +271,13 @@ class TestComplexParams(CommonMOConvertTest):
'params_ref': {'input': "Input[1]{i32}->[10]"}},
{'params_test': {'input': (np.int32, [1, 2, 3])},
'params_ref': {'input': "Input[1,2,3]{i32}"}},
{'params_test': {'input_shape': [Dimension(3, 10), 10, -1]},
{'params_test': {'input_shape': [Dimension(3, 10), 10, -1], 'use_convert_model_from_mo': True},
'params_ref': {'input_shape': '[3..10,10,?]'}},
{'params_test': {'input': [Dimension(3, 10), 10, -1]},
'params_ref': {'input': 'Input[3..10,10,?]'}},
{'params_test': {'input': PartialShape([1, 100, 100, 3]), 'mean_values': [0.5, 1.3, 0.67]},
{'params_test': {'input': PartialShape([1, 100, 100, 3]), 'mean_values': [0.5, 1.3, 0.67], 'use_convert_model_from_mo': True},
'params_ref': {'input': "Input[1,100,100,3]", 'mean_values': "[0.5,1.3,0.67]"}},
{'params_test': {'input': [1, 100, 100, 3], 'scale_values': [0.5, 1.3, 0.67]},
{'params_test': {'input': [1, 100, 100, 3], 'scale_values': [0.5, 1.3, 0.67], 'use_convert_model_from_mo': True},
'params_ref': {'input': "Input[1,100,100,3]", 'scale_values': "[0.5,1.3,0.67]"}},
]
@ -289,24 +294,6 @@ class TestComplexParams(CommonMOConvertTest):
ref_params.update({'input_model': tf_net_path})
self._test(temp_dir, test_params, ref_params)
test_data = [
{
'params_test': {'transform': ('MakeStateful', {'param_res_names': {'Input:0': 'Identity:0'}})},
'params_ref': {'transform': "MakeStateful[param_res_names={\'Input:0\':\'Identity:0\'}]"}}
]
@pytest.mark.parametrize("params", test_data)
@pytest.mark.nightly
def test_mo_convert_transform(self, params, ie_device, precision, ir_version,
temp_dir, use_new_frontend, use_old_api):
tf_net_path = self.create_tf_param_res_model(temp_dir)
test_params = params['params_test']
ref_params = params['params_ref']
test_params.update({'input_model': tf_net_path})
ref_params.update({'input_model': tf_net_path})
self._test(temp_dir, test_params, ref_params)
@pytest.mark.nightly
@pytest.mark.precommit
def test_mo_convert_clearing_transformation_registry(self, ie_device, precision, ir_version,

51
tests/layer_tests/mo_python_api_tests/test_mo_convert_pytorch.py
View File

@ -10,8 +10,8 @@ import openvino.runtime as ov
import pytest
import torch
import unittest
from openvino.runtime import PartialShape, Dimension, Model, Type, InputCutInfo
from openvino.runtime import PartialShape, Dimension, Model, Type
from openvino.tools.ovc import InputCutInfo
from common.mo_convert_test_class import CommonMOConvertTest
@ -159,7 +159,7 @@ def create_pytorch_nn_module_case2(tmp_dir):
sample_input2 = torch.zeros(1, 3, 10, 10)
sample_input = sample_input1, sample_input2
return pt_model, ref_model, {'input_shape': ["[?,3,?,?]", PartialShape([-1, 3, -1, -1])],
return pt_model, ref_model, {'input': [PartialShape("[?,3,?,?]"), PartialShape([-1, 3, -1, -1])],
'example_input': sample_input}
@ -171,7 +171,7 @@ def create_pytorch_nn_module_with_scalar_input(tmp_dir):
sample_input2 = torch.zeros(1, 3, 10, 10)
sample_input = sample_input1, sample_input2
return pt_model, ref_model, {'input_shape': ["[]", PartialShape([-1, 3, -1, -1])],
return pt_model, ref_model, {'input': ["[]", PartialShape([-1, 3, -1, -1])],
'example_input': sample_input}
@ -183,7 +183,7 @@ def create_pytorch_nn_module_case3(tmp_dir):
sample_input2 = torch.zeros(1, 3, 10, 10)
sample_input = tuple([sample_input1, sample_input2])
return pt_model, ref_model, {'input_shape': "[?,3,?,?],[?,3,?,?]",
return pt_model, ref_model, {'input': "[?,3,?,?],[?,3,?,?]",
'example_input': sample_input}
@ -194,7 +194,7 @@ def create_pytorch_nn_module_case4(tmp_dir):
ref_model = make_ref_pt_model_one_input(PartialShape([1, 3, 20, 20]))
return pt_model, ref_model, {'example_input': sample_input, "input_shape": [1, 3, 20, 20]}
return pt_model, ref_model, {'example_input': sample_input, "input": [1, 3, 20, 20]}
def create_pytorch_nn_module_case5(tmp_dir):
@ -247,7 +247,7 @@ def create_pytorch_nn_module_sample_input_int32(tmp_dir):
def create_pytorch_nn_module_sample_input_int32_two_inputs(tmp_dir):
pt_model = make_pt_model_two_inputs()
inp_shapes = ["[?,3,?,?]", PartialShape([-1, 3, -1, -1])]
inp_shapes = [PartialShape("[?,3,?,?]"), PartialShape([-1, 3, -1, -1])]
sample_input1 = torch.zeros(1, 3, 10, 10, dtype=torch.int32)
sample_input2 = torch.zeros(1, 3, 10, 10, dtype=torch.int32)
@ -255,8 +255,7 @@ def create_pytorch_nn_module_sample_input_int32_two_inputs(tmp_dir):
ref_model = make_ref_pt_model_two_inputs(
[PartialShape([-1, 3, -1, -1]), inp_shapes[1]], dtype=np.int32)
return pt_model, ref_model, {'input_shape': inp_shapes,
'input': [np.int32, np.int32],
return pt_model, ref_model, {'input': [(np.int32, inp_shapes[0]), (np.int32, inp_shapes[1])],
'example_input': sample_input}
@ -293,7 +292,7 @@ def create_pytorch_nn_module_layout_list(tmp_dir):
ref_model.inputs[1].node.layout = Layout('nhwc')
return pt_model, ref_model, {
'input_shape': [shape, shape], 'layout': ['nchw', Layout('nhwc')],
'input_shape': [shape, shape], 'layout': ['nchw', Layout('nhwc')], 'use_convert_model_from_mo': True
}
@ -308,7 +307,7 @@ def create_pytorch_nn_module_layout_list_case2(tmp_dir):
ref_model.inputs[1].node.layout = Layout('nhwc')
return pt_model, ref_model, {
'input_shape': [shape, shape], 'layout': ('nchw', Layout('nhwc'))}
'input_shape': [shape, shape], 'layout': ('nchw', Layout('nhwc')), 'use_convert_model_from_mo': True}
def create_pytorch_nn_module_mean_list(tmp_dir):
@ -330,7 +329,8 @@ def create_pytorch_nn_module_mean_list(tmp_dir):
ref_model = Model([sigm], parameter_list, "test")
return pt_model, ref_model, {
'input_shape': [shape, shape], 'mean_values': [[0, 0, 0], [0, 0, 0]], 'compress_to_fp16': False}
'input_shape': [shape, shape], 'mean_values': [[0, 0, 0], [0, 0, 0]], 'compress_to_fp16': False,
'use_convert_model_from_mo': True}
def create_pytorch_nn_module_mean_list_default_no_compression(tmp_dir):
@ -352,7 +352,7 @@ def create_pytorch_nn_module_mean_list_default_no_compression(tmp_dir):
parameter_list = [param1, param2]
ref_model = Model([sigm], parameter_list, "test")
return pt_model, ref_model, {'input_shape': [shape, shape], 'mean_values': [[0, 0, 0], [0, 0, 0]]}
return pt_model, ref_model, {'input_shape': [shape, shape], 'mean_values': [[0, 0, 0], [0, 0, 0]], 'use_convert_model_from_mo': True}
def create_pytorch_nn_module_mean_list_compression_enabled(tmp_dir):
@ -375,7 +375,7 @@ def create_pytorch_nn_module_mean_list_compression_enabled(tmp_dir):
return pt_model, ref_model, {
'input_shape': [shape, shape], 'mean_values': [[0, 0, 0], [0, 0, 0]],
'compress_to_fp16': False}
'compress_to_fp16': False, 'use_convert_model_from_mo': True}
def create_pytorch_nn_module_scale_list(tmp_dir):
@ -396,7 +396,8 @@ def create_pytorch_nn_module_scale_list(tmp_dir):
parameter_list = [param1, param2]
ref_model = Model([sigm], parameter_list, "test")
return pt_model, ref_model, {'input_shape': [shape, shape], 'scale_values': [[1, 1, 1], [1, 1, 1]], 'compress_to_fp16': False}
return pt_model, ref_model, {'input_shape': [shape, shape], 'scale_values': [[1, 1, 1], [1, 1, 1]], 'compress_to_fp16': False,
'use_convert_model_from_mo': True}
def create_pytorch_nn_module_scale_list_default_no_compression(tmp_dir):
@ -418,7 +419,7 @@ def create_pytorch_nn_module_scale_list_default_no_compression(tmp_dir):
parameter_list = [param1, param2]
ref_model = Model([sigm], parameter_list, "test")
return pt_model, ref_model, {'input_shape': [shape, shape], 'scale_values': [[1, 1, 1], [1, 1, 1]]}
return pt_model, ref_model, {'input_shape': [shape, shape], 'scale_values': [[1, 1, 1], [1, 1, 1]], 'use_convert_model_from_mo': True}
def create_pytorch_nn_module_scale_list_compression_enabled(tmp_dir):
@ -444,14 +445,14 @@ def create_pytorch_nn_module_scale_list_compression_enabled(tmp_dir):
ref_model = Model([sigm], parameter_list, "test")
return pt_model, ref_model, {'input_shape': [shape, shape], 'scale_values': [[1, 1, 1], [1, 1, 1]],
'compress_to_fp16': True}
'compress_to_fp16': True, 'use_convert_model_from_mo': True}
def create_pytorch_nn_module_shapes_list_static(tmp_dir):
pt_model = make_pt_model_two_inputs()
ref_model = make_ref_pt_model_two_inputs([1, 3, 20, 20])
return pt_model, ref_model, {'input_shape': [[1, 3, 20, 20], [1, 3, 20, 20]]}
return pt_model, ref_model, {'input': [[1, 3, 20, 20], [1, 3, 20, 20]]}
def create_pytorch_nn_module_shapes_list_static_via_input(tmp_dir):
@ -476,7 +477,7 @@ def create_pytorch_nn_module_shapes_list_dynamic(tmp_dir):
parameter_list = [param1, param2]
ref_model = Model([sigm], parameter_list, "test")
return pt_model, ref_model, {'input_shape': inp_shapes}
return pt_model, ref_model, {'input': inp_shapes}
def create_pytorch_nn_module_shapes_list_dynamic_via_input(tmp_dir):
@ -501,7 +502,7 @@ def create_pytorch_nn_module_shapes_list_dynamic_single_input(tmp_dir):
pt_model = make_pt_model_one_input()
inp_shapes = [[Dimension(-1), 3, 20, Dimension(20, -1)]]
ref_model = make_ref_pt_model_one_input(inp_shapes[0])
return pt_model, ref_model, {'input_shape': inp_shapes}
return pt_model, ref_model, {'input': inp_shapes}
def create_pytorch_nn_module_shapes_list_dynamic_single_input_via_input(tmp_dir):
@ -515,7 +516,7 @@ def create_pytorch_nn_module_shapes_list_static_single_input(tmp_dir):
pt_model = make_pt_model_one_input()
inp_shapes = [[1, 3, 20, 20]]
ref_model = make_ref_pt_model_one_input(inp_shapes[0])
return pt_model, ref_model, {'input_shape': inp_shapes}
return pt_model, ref_model, {'input': inp_shapes}
def create_pytorch_nn_module_shapes_list_static_single_input_via_input(tmp_dir):
@ -677,7 +678,7 @@ def create_pytorch_module_with_optional_inputs_case3(tmp_dir):
(1, 3, 10, 10)), "z": torch.ones((1, 3, 10, 10))}
ref_model = make_ref_pt_model_with_optional_inputs(
[3, 3, 3, 3], z_exist=True)
return net, ref_model, {"example_input": example_input, "input_shape": [[3, 3, 3, 3], [3, 3, 3, 3]]}
return net, ref_model, {"example_input": example_input, "input": [[3, 3, 3, 3], [3, 3, 3, 3]]}
def create_pytorch_module_with_optional_inputs_case4(tmp_dir):
@ -691,7 +692,7 @@ def create_pytorch_module_with_optional_inputs_case5(tmp_dir):
net = make_pt_model_with_optional_input()
ref_model = make_ref_pt_model_with_optional_inputs(
[1, 3, -1, -1], z_exist=True)
return net, ref_model, {"input": ["x", "z"], "input_shape": [[1, 3, -1, -1], [1, 3, -1, -1]]}
return net, ref_model, {"input": [("x",[1, 3, -1, -1]), ("z", [1, 3, -1, -1])]}
def create_pytorch_module_with_compressed_int8_constant(tmp_dir):
@ -956,11 +957,11 @@ def create_pt_model_with_custom_op():
class ConvertRaises(unittest.TestCase):
def test_example_inputs(self):
from openvino.runtime import convert_model
from openvino.tools.ovc import convert_model
pytorch_model = create_pt_model_with_custom_op()
# Check that mo raises error message of wrong argument.
with self.assertRaisesRegex(AssertionError, ".*argument is not recognized.*"):
with self.assertRaisesRegex(TypeError, ".*got an unexpected keyword argument 'example_inputs'.*"):
convert_model(pytorch_model, example_inputs=(torch.tensor(1),))
def test_failed_extension(self):

25
tests/layer_tests/mo_python_api_tests/test_mo_convert_tf.py
View File

@ -139,7 +139,7 @@ def create_tf_module(tmp_dir):
model_ref = Model([sigm], parameter_list, "test")
net = Net()
return net, model_ref, {'input_shape': [PartialShape([1, 2, 3]), PartialShape([1, 2, 3])]}
return net, model_ref, {'input': [PartialShape([1, 2, 3]), PartialShape([1, 2, 3])]}
def create_tf_module_layout_list(tmp_dir):
@ -166,7 +166,8 @@ def create_tf_module_layout_list(tmp_dir):
model_ref.inputs[1].node.layout = Layout('NHC')
net = Net()
return net, model_ref, {'input_shape': [PartialShape([1, 2, 3]), PartialShape([1, 2, 3])], 'layout': ["NCH", "NHC"]}
return net, model_ref, {'input_shape': [PartialShape([1, 2, 3]), PartialShape([1, 2, 3])], 'layout': ["NCH", "NHC"],
'use_convert_model_from_mo': True}
def create_tf_module_dynamic(tmp_dir):
@ -192,7 +193,7 @@ def create_tf_module_dynamic(tmp_dir):
model_ref = Model([sigm], parameter_list, "test")
net = Net()
return net, model_ref, {'input_shape': input_shapes}
return net, model_ref, {'input': input_shapes}
def create_keras_layer(tmp_dir):
@ -216,7 +217,7 @@ def create_keras_layer(tmp_dir):
model_ref = Model([sigm], parameter_list, "test")
net = LayerModel()
return net, model_ref, {'input_shape': [PartialShape([1, 2, 3]), PartialShape([1, 2, 3])]}
return net, model_ref, {'input': [PartialShape([1, 2, 3]), PartialShape([1, 2, 3])]}
def create_keras_layer_dynamic(tmp_dir):
@ -242,7 +243,7 @@ def create_keras_layer_dynamic(tmp_dir):
model_ref = Model([sigm], parameter_list, "test")
net = LayerModel()
return net, model_ref, {'input_shape': input_shapes}
return net, model_ref, {'input': input_shapes}
def create_tf_checkpoint(tmp_dir):
@ -518,7 +519,7 @@ def create_keras_layer_with_example_input_2(tmp_dir):
def create_keras_layer_with_input_shapes_case1(tmp_dir):
model, model_ref = create_keras_layer_input_list()
return model, model_ref, {'input_shape': [[1, 2, 3], [1, 2, 3]]}
return model, model_ref, {'input': [[1, 2, 3], [1, 2, 3]]}
def create_keras_layer_with_input_shapes_case2(tmp_dir):
@ -528,7 +529,7 @@ def create_keras_layer_with_input_shapes_case2(tmp_dir):
def create_keras_layer_with_input_shapes_case3(tmp_dir):
model, model_ref = create_keras_layer_input_dict_one_inp()
return model, model_ref, {'input': ['args'], 'input_shape': [1, 2, 3]}
return model, model_ref, {'input': [('args', [1, 2, 3])]}
def create_keras_layer_with_input_shapes_case4(tmp_dir):
@ -669,7 +670,7 @@ class TestMoConvertTF(CommonMOConvertTest):
temp_dir):
fw_model, graph_ref, mo_params = create_model(temp_dir)
test_params = {'input_model': fw_model, 'use_new_frontend': True}
test_params = {'input_model': fw_model}
if mo_params is not None:
test_params.update(mo_params)
self._test_by_ref_graph(temp_dir, test_params, graph_ref, compare_tensor_names=False)
@ -679,10 +680,10 @@ class TestMoConvertTF(CommonMOConvertTest):
def test_unnamed_saved_model_dir(self, ie_device, precision, ir_version, temp_dir):
saved_model_dir, graph_ref = create_tf_saved_model_dir(temp_dir)
test_params = {'input_model': saved_model_dir, 'use_new_frontend': True}
test_params = {'input_model': saved_model_dir}
self._test_by_ref_graph(temp_dir, test_params, graph_ref, compare_tensor_names=False)
test_params = {'input_model': saved_model_dir, 'use_new_frontend': False}
test_params = {'input_model': saved_model_dir}
self._test_by_ref_graph(temp_dir, test_params, graph_ref, compare_tensor_names=False)
def test_zero_copy(self, ie_device, precision, ir_version, temp_dir):
@ -741,7 +742,7 @@ class TestMoConvertTF(CommonMOConvertTest):
import tensorflow as tf
tf.compat.v1.reset_default_graph()
from openvino.tools.mo import convert_model
from openvino.tools.ovc import convert_model
from openvino.runtime import compile_model
import gc
@ -795,7 +796,7 @@ class TFConvertTest(unittest.TestCase):
@pytest.mark.precommit
def test_tf_function_no_signature(self):
import tensorflow as tf
from openvino.runtime import convert_model
from openvino.tools.ovc import convert_model
@tf.function()
def function(x1, x2):

12
tests/layer_tests/mo_python_api_tests/test_mo_help.py
View File

@ -5,8 +5,9 @@
import os
import sys
import unittest
from openvino.tools.mo import mo
from openvino.tools.ovc import ovc
from openvino.tools.ovc.cli_parser import get_mo_convert_params
from openvino.tools.mo.utils.cli_parser import get_mo_convert_params as legacy_mo_params
from pathlib import Path
from common.utils.common_utils import shell
@ -16,8 +17,8 @@ class TestSubprocessMoConvert(unittest.TestCase):
def test_mo_convert(self):
mo_convert_params = get_mo_convert_params()
# Test cli tool help
mo_path = Path(mo.__file__).parent
# Test ovc tool help
mo_path = Path(ovc.__file__).parent
mo_runner = mo_path.joinpath('main.py').as_posix()
params = [sys.executable, mo_runner, "--help"]
_, mo_output, _ = shell(params)
@ -29,11 +30,12 @@ class TestSubprocessMoConvert(unittest.TestCase):
for param_name in group:
assert param_name in mo_output
# Test Python API help
# Test Python API help, applicable for convert_model from tools.mo only
mo_help_file = os.path.join(os.path.dirname(__file__), "mo_convert_help.py")
params = [sys.executable, mo_help_file]
_, mo_output, _ = shell(params)
for group in mo_convert_params:
legacy_params = legacy_mo_params()
for group in legacy_params:
for param_name in group:
assert param_name in mo_output

6
tests/layer_tests/mo_python_api_tests/test_ovc_cli_tool.py
View File

@ -27,7 +27,9 @@ def generate_ir_ovc(coverage=False, **kwargs):
else:
params = [sys.executable, ovc_runner]
for key, value in kwargs.items():
if key == "batch":
if key == "input_model":
params.append((str(value)))
elif key == "batch":
params.extend(("-b", str(value)))
elif key == "k":
params.extend(("-k", str(value)))
@ -81,7 +83,7 @@ class TestOVCTool(CommonMOConvertTest):
core = Core()
# tests for MO cli tool
exit_code, stderr = generate_ir_ovc(coverage=False, **{"input_model": model_path, "output_dir": temp_dir})
exit_code, stderr = generate_ir_ovc(coverage=False, **{"input_model": model_path, "output_model": temp_dir + os.sep + "model"})
assert not exit_code
ov_model = core.read_model(os.path.join(temp_dir, "model.xml"))

2
tests/layer_tests/pytorch_tests/pytorch_layer_test_class.py
View File

@ -148,7 +148,7 @@ class PytorchLayerTest:
def convert_via_mo(self, model, example_input, trace_model, dynamic_shapes, ov_inputs):
import torch
from openvino.runtime import convert_model
from openvino.tools.ovc import convert_model
kwargs = {"example_input": example_input if len(
example_input) > 1 else example_input[0], "compress_to_fp16": False}
with torch.no_grad():

3
tools/mo/openvino/tools/mo/__init__.py
View File

@ -1,5 +1,4 @@
# Copyright (C) 2018-2023 Intel Corporation
# SPDX-License-Identifier: Apache-2.0
from openvino.tools.mo.convert import convert_model
from openvino.tools.ovc import InputCutInfo, LayoutMap # pylint: disable=no-name-in-module,import-error
from openvino.tools.mo.convert import convert_model, LayoutMap, InputCutInfo

5
tools/mo/openvino/tools/mo/convert.py
View File

@ -2,14 +2,17 @@
# SPDX-License-Identifier: Apache-2.0
import os
import pathlib
from collections import namedtuple
from typing import Any
from openvino.runtime import PartialShape, Shape, Layout, Model
from openvino.tools.mo.convert_impl import _convert
from openvino.tools.ovc import InputCutInfo, LayoutMap # pylint: disable=no-name-in-module,import-error
from openvino.tools.mo.utils.cli_parser import get_all_cli_parser # pylint: disable=no-name-in-module,import-error
from openvino.tools.mo.utils.logger import get_logger_state, restore_logger_state # pylint: disable=no-name-in-module,import-error
LayoutMap = namedtuple("LayoutMap", ["source_layout", "target_layout"], defaults=[None, None])
InputCutInfo = namedtuple("InputInfo", ["name", "shape", "type", "value"], defaults=[None, None, None, None])
def convert_model(
input_model: [str, pathlib.Path, Any] = None,

28
tools/mo/openvino/tools/mo/utils/cli_parser.py
View File

@ -147,13 +147,13 @@ def single_input_to_input_cut_info(input: [str, tuple, list, PartialShape, Type,
if isinstance(input, str):
# Parse params from string
node_name, shape, value, data_type = parse_input_value(input)
return openvino.runtime.InputCutInfo(node_name,
return openvino.tools.mo.InputCutInfo(node_name,
PartialShape(shape) if shape is not None else None,
data_type,
value)
if isinstance(input, openvino.runtime.InputCutInfo):
if isinstance(input, openvino.tools.mo.InputCutInfo):
# Wrap input.shape to PartialShape if possible and wrap to InputCutInfo
return openvino.runtime.InputCutInfo(input.name,
return openvino.tools.mo.InputCutInfo(input.name,
PartialShape(input.shape) if input.shape is not None else None,
input.type,
input.value)
@ -183,18 +183,18 @@ def single_input_to_input_cut_info(input: [str, tuple, list, PartialShape, Type,
else:
raise Exception("Incorrect input parameters provided. Expected tuple with input name, "
"input type or input shape. Got unknown object: {}".format(val))
return openvino.runtime.InputCutInfo(name,
return openvino.tools.mo.InputCutInfo(name,
PartialShape(shape) if shape is not None else None,
inp_type,
None)
# Case when only type is set
if isinstance(input, (type, Type)):
return openvino.runtime.InputCutInfo(None, None, input, None)
return openvino.tools.mo.InputCutInfo(None, None, input, None)
# We don't expect here single unnamed value. If list of int is set it is considered as shape.
# Setting of value is expected only using InputCutInfo or string analog.
raise Exception("Unexpected object provided for input. Expected openvino.runtime.InputCutInfo "
raise Exception("Unexpected object provided for input. Expected openvino.toos.mo.InputCutInfo "
"or tuple or str. Got {}".format(type(input)))
@ -213,12 +213,12 @@ def input_to_input_cut_info(input: [str, tuple, list]):
# Parse string with parameters for single input
node_name, shape, value, data_type = parse_input_value(input_value)
inputs.append(openvino.runtime.InputCutInfo(node_name,
inputs.append(openvino.tools.mo.InputCutInfo(node_name,
PartialShape(shape) if shape is not None else None,
data_type,
value))
return inputs
if isinstance(input, openvino.runtime.InputCutInfo):
if isinstance(input, openvino.tools.mo.InputCutInfo):
# Wrap to list and return
return [input]
if isinstance(input, tuple):
@ -269,11 +269,11 @@ def input_shape_to_input_cut_info(input_shape: [str, Shape, PartialShape, list,
shape = PartialShape(shape)
assert inputs[idx].shape is None, "Shape was set in both \"input\" and in \"input_shape\" parameter." \
"Please use either \"input\" or \"input_shape\" for shape setting."
inputs[idx] = openvino.runtime.InputCutInfo(inputs[idx].name, shape, inputs[idx].type, inputs[idx].value)
inputs[idx] = openvino.tools.mo.InputCutInfo(inputs[idx].name, shape, inputs[idx].type, inputs[idx].value)
else:
for shape in input_shape:
inputs.append(openvino.runtime.InputCutInfo(None, PartialShape(shape), None, None))
inputs.append(openvino.tools.mo.InputCutInfo(None, PartialShape(shape), None, None))
return
raise Exception("Unexpected object provided for input_shape. Expected PartialShape, Shape, tuple, list or str. "
@ -375,7 +375,7 @@ def source_target_layout_to_str(value):
def layoutmap_to_str(value):
if isinstance(value, str):
return value
if isinstance(value, openvino.runtime.LayoutMap):
if isinstance(value, openvino.tools.mo.LayoutMap):
assert value.source_layout is not None, "Incorrect layout map. 'source_layout' should be set."
source_layout = layout_to_str(value.source_layout)
if value.target_layout is not None:
@ -400,7 +400,7 @@ def layout_param_to_str(value):
raise Exception("Incorrect operation name type. Expected string, got {}".format(type(op_name)))
values_str.append(op_name + "(" + layoutmap_to_str(layout) + ")")
return ",".join(values_str)
if isinstance(value, openvino.runtime.LayoutMap):
if isinstance(value, openvino.tools.mo.LayoutMap):
return layoutmap_to_str(value)
if isinstance(value, list) or isinstance(value, tuple):
values_str = []
@ -490,7 +490,7 @@ ParamDescription = namedtuple("ParamData",
def get_mo_convert_params():
mo_convert_docs = openvino.runtime.convert_model.__doc__
mo_convert_docs = openvino.tools.mo.convert_model.__doc__
mo_convert_params = {}
group = "Optional parameters:"
mo_convert_params[group] = {}
@ -784,7 +784,7 @@ def writable_dir(path: str):
def add_args_by_description(args_group, params_description):
signature = inspect.signature(openvino.runtime.convert_model)
signature = inspect.signature(openvino.tools.mo.convert_model)
filepath_args = get_params_with_paths_list()
cli_tool_specific_descriptions = get_convert_model_help_specifics()
for param_name, param_description in params_description.items():

4
tools/mo/unit_tests/mo/utils/args_to_string_test.py
View File

@ -2,9 +2,9 @@
# SPDX-License-Identifier: Apache-2.0
import numpy as np
from openvino.runtime import Layout, PartialShape, Dimension, Shape, Type
from openvino.runtime import Layout, Dimension
from openvino.runtime import InputCutInfo, LayoutMap
from openvino.tools.mo import LayoutMap
from openvino.tools.mo.utils.cli_parser import mean_scale_value_to_str, \
transform_param_to_str, str_list_to_str, source_target_layout_to_str, layout_param_to_str
from unit_tests.mo.unit_test_with_mocked_telemetry import UnitTestWithMockedTelemetry

3
tools/mo/unit_tests/mo/utils/cli_parser_test.py
View File

@ -20,7 +20,8 @@ from openvino.tools.mo.utils.cli_parser import get_placeholder_shapes, get_tuple
from openvino.tools.mo.convert_impl import pack_params_to_args_namespace
from openvino.tools.mo.utils.error import Error
from unit_tests.mo.unit_test_with_mocked_telemetry import UnitTestWithMockedTelemetry
from openvino.runtime import PartialShape, Dimension, Layout, InputCutInfo, LayoutMap
from openvino.runtime import PartialShape, Dimension, Layout
from openvino.tools.mo import LayoutMap, InputCutInfo
class TestingMeanScaleGetter(UnitTestWithMockedTelemetry):

11
tools/ovc/openvino/tools/ovc/__init__.py
View File

@ -1,13 +1,4 @@
# Copyright (C) 2018-2023 Intel Corporation
# SPDX-License-Identifier: Apache-2.0
from openvino.tools.ovc.convert import convert_model, InputCutInfo, LayoutMap
# pylint: disable=no-name-in-module,import-error,no-member
try:
import openvino.runtime
openvino.runtime.convert_model = convert_model
openvino.runtime.InputCutInfo = InputCutInfo
openvino.runtime.LayoutMap = LayoutMap
except:
pass
from openvino.tools.ovc.convert import convert_model, InputCutInfo

882
tools/ovc/openvino/tools/ovc/cli_parser.py
View File

File diff suppressed because it is too large Load Diff

271
tools/ovc/openvino/tools/ovc/convert.py
View File

@ -6,105 +6,42 @@ import pathlib
from collections import namedtuple
from typing import Any
from openvino.runtime import PartialShape, Shape, Layout, Model # pylint: disable=no-name-in-module,import-error
from openvino.runtime import PartialShape, Shape, Model # pylint: disable=no-name-in-module,import-error
from openvino.tools.ovc.convert_impl import _convert
from openvino.tools.ovc.logger import get_logger_state, restore_logger_state
from openvino.tools.ovc.cli_parser import get_all_cli_parser
#TODO: Why names InputCutInfo and InputInfo are different
InputCutInfo = namedtuple("InputInfo", ["name", "shape", "type", "value"], defaults=[None, None, None, None])
LayoutMap = namedtuple("LayoutMap", ["source_layout", "target_layout"], defaults=[None, None])
def convert_model(
input_model: [str, pathlib.Path, Any] = None,
# Optional parameters
help: bool = False,
framework: [str] = None,
input_model: [str, pathlib.Path, Any, list], # TODO: Instead of list just accept arbitrary number of positional arguments
# Framework-agnostic parameters
input: [str, list, tuple, InputCutInfo] = None,
output: [str, list] = None,
input_shape: [str, PartialShape, Shape, list] = None,
example_input: Any = None,
batch: int = None,
mean_values: [str, dict, list] = (),
scale_values: [str, dict, list] = (),
scale: [str, float] = None,
reverse_input_channels: bool = False,
source_layout: [str, Layout, dict] = (),
target_layout: [str, Layout, dict] = (),
layout: [str, Layout, LayoutMap, list, dict] = (),
compress_to_fp16: bool = False,
extensions: [str, pathlib.Path, list, Any] = None,
transform: [str, list, tuple] = "",
transformations_config: [str, pathlib.Path] = None,
silent: bool = True,
log_level: str = 'ERROR',
version: bool = None,
progress: bool = False,
stream_output: bool = False,
verbose: bool = False,
share_weights: bool = True,
# PaddlePaddle-specific parameters:
example_output: Any = None,
example_output: Any = None, # TODO: Consider removing
# TensorFlow*-specific parameters
input_model_is_text: bool = None,
input_checkpoint: [str, pathlib.Path] = None,
input_meta_graph: [str, pathlib.Path] = None,
saved_model_dir: [str, pathlib.Path] = None,
saved_model_tags: [str, list] = None,
tensorflow_custom_operations_config_update: [str, pathlib.Path] = None,
tensorflow_object_detection_api_pipeline_config: [str, pathlib.Path] = None,
tensorboard_logdir: [str, pathlib.Path] = None,
tensorflow_custom_layer_libraries: [str, pathlib.Path] = None,
# MXNet-specific parameters:
input_symbol: [str, pathlib.Path] = None,
nd_prefix_name: str = None,
pretrained_model_name: str = None,
save_params_from_nd: bool = None,
legacy_mxnet_model: bool = None,
enable_ssd_gluoncv: bool = False,
# Caffe*-specific parameters:
input_proto: [str, pathlib.Path] = None,
caffe_parser_path: [str, pathlib.Path] = None,
k: [str, pathlib.Path] = None,
disable_omitting_optional: bool = False,
enable_flattening_nested_params: bool = False,
# Kaldi-specific parameters:
counts: [str, pathlib.Path] = None,
remove_output_softmax: bool = False,
remove_memory: bool = False,
**args
saved_model_tags: [str, list] = None, # TODO: Consider removing
) -> Model:
"""
Converts the model from original framework to OpenVino Model.
Args:
:param help:
Print available parameters.
:param framework:
Name of the framework used to train the input model.
Framework-agnostic parameters:
:param input_model:
Model object in original framework (PyTorch, Tensorflow) or path to model file.
Tensorflow*: a file with a pre-trained model (binary or text .pb file after freezing).
Caffe*: a model proto file with model weights
Supported formats of input model:
PaddlePaddle
paddle.hapi.model.Model
paddle.fluid.dygraph.layers.Layer
paddle.fluid.executor.Executor
PyTorch
torch.nn.Module
torch.jit.ScriptModule
@ -123,6 +60,11 @@ def convert_model(
tf.Module
tf.train.checkpoint
PaddlePaddle
paddle.hapi.model.Model
paddle.fluid.dygraph.layers.Layer
paddle.fluid.executor.Executor
:param input:
Input can be set by passing a list of InputCutInfo objects or by a list
of tuples. Each tuple can contain optionally input name, input
@ -149,94 +91,11 @@ def convert_model(
The name of the output operation of the model or list of names. For TensorFlow*,
do not add :0 to this name.The order of outputs in converted model is the
same as order of specified operation names.
:param input_shape:
Input shape(s) that should be fed to an input node(s) of the model. Input
shapes can be defined by passing a list of objects of type PartialShape,
Shape, [Dimension, ...] or [int, ...] or by a string of the following
format. Shape is defined as a comma-separated list of integer numbers
enclosed in parentheses or square brackets, for example [1,3,227,227]
or (1,227,227,3), where the order of dimensions depends on the framework
input layout of the model. For example, [N,C,H,W] is used for ONNX* models
and [N,H,W,C] for TensorFlow* models. The shape can contain undefined
dimensions (? or -1) and should fit the dimensions defined in the input
operation of the graph. Boundaries of undefined dimension can be specified
with ellipsis, for example [1,1..10,128,128]. One boundary can be
undefined, for example [1,..100] or [1,3,1..,1..]. If there are multiple
inputs in the model, "input_shape" should contain definition of shape
for each input separated by a comma, for example: [1,3,227,227],[2,4]
for a model with two inputs with 4D and 2D shapes. Alternatively, specify
shapes with the "input" option.
:param example_input:
Sample of model input in original framework.
For PyTorch it can be torch.Tensor.
For Tensorflow it can be tf.Tensor or numpy.ndarray.
For PaddlePaddle it can be Paddle Variable.
:param batch:
Set batch size. It applies to 1D or higher dimension inputs.
The default dimension index for the batch is zero.
Use a label 'n' in "layout" or "source_layout" option to set the batch dimension.
For example, "x(hwnc)" defines the third dimension to be the batch.
:param mean_values:
Mean values to be used for the input image per channel. Mean values can
be set by passing a dictionary, where key is input name and value is mean
value. For example mean_values={'data':[255,255,255],'info':[255,255,255]}.
Or mean values can be set by a string of the following format. Values to
be provided in the (R,G,B) or [R,G,B] format. Can be defined for desired
input of the model, for example: mean_values="data[255,255,255],info[255,255,255]".
The exact meaning and order of channels depend on how the original model
was trained.
:param scale_values:
Scale values to be used for the input image per channel. Scale values
can be set by passing a dictionary, where key is input name and value is
scale value. For example scale_values={'data':[255,255,255],'info':[255,255,255]}.
Or scale values can be set by a string of the following format. Values
are provided in the (R,G,B) or [R,G,B] format. Can be defined for desired
input of the model, for example: scale_values="data[255,255,255],info[255,255,255]".
The exact meaning and order of channels depend on how the original model
was trained. If both "mean_values" and "scale_values" are specified,
the mean is subtracted first and then scale is applied regardless of
the order of options in command line.
:param scale:
All input values coming from original network inputs will be divided
by this value. When a list of inputs is overridden by the "input" parameter,
this scale is not applied for any input that does not match with the original
input of the model. If both "mean_values" and "scale" are specified,
the mean is subtracted first and then scale is applied regardless of
the order of options in command line.
:param reverse_input_channels:
Switch the input channels order from RGB to BGR (or vice versa). Applied
to original inputs of the model if and only if a number of channels equals
3. When "mean_values"/"scale_values" are also specified, reversing
of channels will be applied to user's input data first, so that numbers
in "mean_values" and "scale_values" go in the order of channels used
in the original model. In other words, if both options are specified,
then the data flow in the model looks as following: Parameter -> ReverseInputChannels
-> Mean apply-> Scale apply -> the original body of the model.
:param source_layout:
Layout of the input or output of the model in the framework. Layout can
be set by passing a dictionary, where key is input name and value is LayoutMap
object. Or layout can be set by string of the following format. Layout
can be specified in the short form, e.g. nhwc, or in complex form, e.g.
"[n,h,w,c]". Example for many names: "in_name1([n,h,w,c]),in_name2(nc),out_name1(n),out_name2(nc)".
Layout can be partially defined, "?" can be used to specify undefined
layout for one dimension, "..." can be used to specify undefined layout
for multiple dimensions, for example "?c??", "nc...", "n...c", etc.
:param target_layout:
Same as "source_layout", but specifies target layout that will be in
the model after processing by ModelOptimizer.
:param layout:
Combination of "source_layout" and "target_layout". Can't be used
with either of them. If model has one input it is sufficient to specify
layout of this input, for example "layout" nhwc. To specify layouts
of many tensors, names must be provided, for example: layout="name1(nchw),name2(nc)".
It is possible to instruct ModelOptimizer to change layout, for example:
layout="name1(nhwc->nchw),name2(cn->nc)".
Also "*" in long layout form can be used to fuse dimensions, for example "[n,c,...]->[n*c,...]".
:param compress_to_fp16:
If the original model has FP32 weights or biases, they are compressed
to FP16. All intermediate data is kept in original precision. Option
can be specified alone as "compress_to_fp16", or explicit True/False
values can be set, for example: "compress_to_fp16=False", or "compress_to_fp16=True"
:param extensions:
Paths to libraries (.so or .dll) with extensions, comma-separated
list of paths, objects derived from BaseExtension class or lists of
@ -244,124 +103,28 @@ def convert_model(
a directory or a comma-separated list of directories with extensions
are supported. To disable all extensions including those that are placed
at the default location, pass an empty string.
:param transform:
Apply additional transformations. 'transform' can be set by a list
of tuples, where the first element is transform name and the second element
is transform parameters. For example: [('LowLatency2', {{'use_const_initializer':
False}}), ...] transform="transformation_name1[args],transformation_name2..."
where [args] is key=value pairs separated by semicolon. Examples:
transform="LowLatency2" or
transform="Pruning" or
transform="LowLatency2[use_const_initializer=False]" or
transform="MakeStateful[param_res_names=
{'input_name_1':'output_name_1','input_name_2':'output_name_2'}]"
Available transformations: "LowLatency2", "MakeStateful", "Pruning"
:param transformations_config:
Use the configuration file with transformations description or pass
object derived from BaseExtension class. Transformations file can
be specified as relative path from the current directory, as absolute
path or as relative path from the mo root directory.
:param silent:
Prevent any output messages except those that correspond to log level
equals ERROR, that can be set with the following option: "log_level".
By default, log level is already ERROR.
:param log_level:
Logger level of logging massages from MO.
Expected one of ['CRITICAL', 'ERROR', 'WARN', 'WARNING', 'INFO', 'DEBUG', 'NOTSET'].
:param version:
Version of Model Conversion API
:param progress:
Enable model conversion progress display.
:param stream_output:
Switch model conversion progress display to a multiline mode.
:param verbose:
Print detailed information about conversion.
:param share_weights:
Map memory of weights instead reading files or share memory from input model.
Currently, mapping feature is provided only for ONNX models
that do not require fallback to the legacy ONNX frontend for the conversion.
Reuse weights allocated in the original model. If input model is in file,
then mmap is used to allocate weights directly from file. If input model is
runtime object, then original memory regions allocated in the original model
are reused for weights in the converted model.
PaddlePaddle-specific parameters:
:param example_output:
Sample of model output in original framework. For PaddlePaddle it can be Paddle Variable.
TensorFlow*-specific parameters:
:param input_model_is_text:
TensorFlow*: treat the input model file as a text protobuf format. If
not specified, the convert_model() treats it as a binary file by default.
:param input_checkpoint:
TensorFlow*: variables file to load.
:param input_meta_graph:
Tensorflow*: a file with a meta-graph of the model before freezing
:param saved_model_dir:
TensorFlow*: directory with a model in SavedModel format of TensorFlow
1.x or 2.x version.
:param saved_model_tags:
Group of tag(s) of the MetaGraphDef to load, in string format, separated
by ','. For tag-set contains multiple tags, all tags must be passed in.
:param tensorflow_custom_operations_config_update:
TensorFlow*: update the configuration file with node name patterns
with input/output nodes information.
:param tensorflow_object_detection_api_pipeline_config:
TensorFlow*: path to the pipeline configuration file used to generate
model created with help of Object Detection API.
:param tensorboard_logdir:
TensorFlow*: dump the input graph to a given directory that should be
used with TensorBoard.
:param tensorflow_custom_layer_libraries:
TensorFlow*: comma separated list of shared libraries with TensorFlow*
custom operations implementation.
MXNet-specific parameters:
:param input_symbol:
Symbol file (for example, model-symbol.json) that contains a topology
structure and layer attributes
:param nd_prefix_name:
Prefix name for args.nd and argx.nd files.
:param pretrained_model_name:
Name of a pretrained MXNet model without extension and epoch number.
This model will be merged with args.nd and argx.nd files
:param save_params_from_nd:
Enable saving built parameters file from .nd files
:param legacy_mxnet_model:
Enable MXNet loader to make a model compatible with the latest MXNet
version. Use only if your model was trained with MXNet version lower
than 1.0.0
:param enable_ssd_gluoncv:
Enable pattern matchers replacers for converting gluoncv ssd topologies.
Caffe*-specific parameters:
:param input_proto:
Deploy-ready prototxt file that contains a topology structure and
layer attributes
:param caffe_parser_path:
Path to Python Caffe* parser generated from caffe.proto
:param k:
Path to CustomLayersMapping.xml to register custom layers
:param disable_omitting_optional:
Disable omitting optional attributes to be used for custom layers.
Use this option if you want to transfer all attributes of a custom layer
to IR. Default behavior is to transfer the attributes with default values
and the attributes defined by the user to IR.
:param enable_flattening_nested_params:
Enable flattening optional params to be used for custom layers. Use
this option if you want to transfer attributes of a custom layer to IR
with flattened nested parameters. Default behavior is to transfer
the attributes without flattening nested parameters.
Kaldi-specific parameters:
:param counts:
Path to the counts file
:param remove_output_softmax:
Removes the SoftMax layer that is the output layer
:param remove_memory:
Removes the Memory layer and use additional inputs outputs instead
Returns:
openvino.runtime.Model
"""
params = locals()
logger_state = get_logger_state()
del params['args']
params.update(args)
cli_parser = get_all_cli_parser()
ov_model, _ = _convert(cli_parser, params, True)
restore_logger_state(logger_state)

452
tools/ovc/openvino/tools/ovc/convert_impl.py
View File

@ -15,40 +15,31 @@ try:
except ImportError:
import openvino.tools.ovc.telemetry_stub as tm
from openvino.tools.ovc.moc_frontend.check_config import legacy_transformations_config_used, \
tensorflow_custom_operations_config_update_used, new_extensions_used
from openvino.tools.ovc.moc_frontend.check_config import new_extensions_used
from openvino.tools.ovc.moc_frontend.pipeline import moc_pipeline
from openvino.tools.ovc.moc_frontend.moc_emit_ir import moc_emit_ir
from openvino.tools.ovc.convert_data_type import destination_type_to_np_data_type
from openvino.tools.ovc.cli_parser import check_available_transforms, \
get_advanced_cli_options, get_available_front_ends, get_caffe_cli_options, \
get_common_cli_options, get_kaldi_cli_options, get_layout_values, get_freeze_placeholder_values, \
get_mean_scale_dictionary, get_mxnet_cli_options, get_onnx_cli_options, \
get_placeholder_shapes, get_tf_cli_options, parse_transform, parse_tuple_pairs, \
get_model_name_from_args, depersonalize, get_mo_convert_params, input_to_input_cut_info, \
input_shape_to_input_cut_info, freeze_placeholder_to_input_cut_info
from openvino.tools.ovc.cli_parser import get_available_front_ends, \
get_common_cli_options, get_model_name_from_args, depersonalize, get_mo_convert_params, \
input_to_input_cut_info, freeze_placeholder_to_input_cut_info
from openvino.tools.ovc.error import Error, FrameworkError, legacy_path_error
from openvino.tools.ovc.get_ov_update_message import get_ov_update_message, get_ov_api20_message, \
get_tf_fe_message, get_try_legacy_fe_message, get_compression_message
from openvino.tools.ovc.error import Error, FrameworkError
from openvino.tools.ovc.get_ov_update_message import get_ov_update_message, get_compression_message
from openvino.tools.ovc.version import VersionChecker
from openvino.tools.ovc.utils import deduce_legacy_frontend_by_namespace, refer_to_faq_msg, check_values_equal
from openvino.tools.ovc.logger import init_logger, progress_printer
from openvino.tools.ovc.utils import check_values_equal
from openvino.tools.ovc.logger import init_logger
from openvino.tools.ovc.telemetry_utils import send_params_info, send_conversion_result, \
get_tid
from openvino.tools.ovc.moc_frontend.check_config import legacy_extensions_used
from openvino.tools.ovc.moc_frontend.check_config import default_path as extensions_default_path
from openvino.tools.ovc.moc_frontend.pytorch_frontend_utils import get_pytorch_decoder, extract_input_info_from_example
from openvino.tools.ovc.moc_frontend.paddle_frontend_utils import paddle_frontend_converter
from openvino.tools.ovc.moc_frontend.shape_utils import parse_input_shapes
# pylint: disable=no-name-in-module,import-error
from openvino.frontend import FrontEndManager, OpConversionFailure, ProgressReporterExtension, TelemetryExtension
from openvino.frontend import FrontEndManager, OpConversionFailure, TelemetryExtension
from openvino.runtime import get_version as get_rt_version
from openvino.runtime import Type, PartialShape
try:
from openvino.frontend.tensorflow.utils import type_supported_by_tf_fe, create_tf_graph_iterator, \
from openvino.frontend.tensorflow.utils import create_tf_graph_iterator, type_supported_by_tf_fe, \
extract_model_graph # pylint: disable=no-name-in-module,import-error
tf_frontend_with_python_bindings_installed = True
@ -63,31 +54,13 @@ def replace_ext(name: str, old: str, new: str):
return base + new
def print_argv(argv: argparse.Namespace, is_caffe: bool, is_tf: bool, is_mxnet: bool, is_kaldi: bool, is_onnx: bool,
model_name: str):
def print_argv(argv: argparse.Namespace, model_name: str):
print('Model Conversion arguments:')
props = OrderedDict()
props['common_args'] = get_common_cli_options(model_name)
props['advanced_args'] = get_advanced_cli_options()
if is_caffe:
props['caffe_args'] = get_caffe_cli_options()
if is_tf:
props['tf_args'] = get_tf_cli_options()
if is_mxnet:
props['mxnet_args'] = get_mxnet_cli_options()
if is_kaldi:
props['kaldi_args'] = get_kaldi_cli_options()
if is_onnx:
props['onnx_args'] = get_onnx_cli_options()
framework_specifics_map = {
'common_args': 'Common parameters:',
'advanced_args': 'Advanced parameters:',
'caffe_args': 'Caffe specific parameters:',
'tf_args': 'TensorFlow specific parameters:',
'mxnet_args': 'MXNet specific parameters:',
'kaldi_args': 'Kaldi specific parameters:',
'onnx_args': 'ONNX specific parameters:',
'common_args': 'Common parameters:'
}
lines = []
@ -97,233 +70,50 @@ def print_argv(argv: argparse.Namespace, is_caffe: bool, is_tf: bool, is_mxnet:
if isinstance(desc, list):
lines.append('\t{}: \t{}'.format(desc[0], desc[1](getattr(argv, op, 'NONE'))))
else:
if op == 'k':
default_path = os.path.join(os.path.dirname(sys.argv[0]),
'openvino/tools/mo/front/caffe/CustomLayersMapping.xml')
if getattr(argv, op, 'NONE') == default_path:
lines.append('\t{}: \t{}'.format(desc, 'Default'))
continue
lines.append('\t{}: \t{}'.format(desc, getattr(argv, op, 'NONE')))
print('\n'.join(lines), flush=True)
def legacy_framework_check(is_caffe, is_mxnet, is_kaldi):
if is_caffe:
legacy_path_error("The provided model is from Caffe framework. This is legacy functionality. ")
if is_mxnet:
legacy_path_error("The provided model is from MxNet framework. This is legacy functionality. ")
if is_kaldi:
legacy_path_error("The provided model is from Kaldi framework. This is legacy functionality. ")
def check_legacy_args(non_default_params, python_api_used):
ignored_cli_options = ["output_dir", "model_name"]
legacy_groups = ['Kaldi-specific parameters:', 'Caffe*-specific parameters:', 'MXNet-specific parameters:']
tf_legacy_args = ['tensorflow_custom_operations_config_update', 'tensorflow_object_detection_api_pipeline_config',
'tensorboard_logdir', 'tensorflow_custom_layer_libraries', 'saved_model_tags']
mo_convert_params = get_mo_convert_params()
for key, value in non_default_params.items():
if key in ignored_cli_options:
if python_api_used:
print("The provided option \"{}\" is applicable in command line tool only. The option will be ignored.".format(key))
for group in legacy_groups:
if key in mo_convert_params[group]:
legacy_path_error("The provided option \"{}\" refers to legacy functionality. ".format(key))
if key in tf_legacy_args:
legacy_path_error("The provided option \"{}\" refers to legacy functionality. ".format(key))
def arguments_post_parsing(argv: argparse.Namespace):
use_legacy_frontend = argv.use_legacy_frontend
use_new_frontend = argv.use_new_frontend
if argv.extensions is None:
argv.extensions = [extensions_default_path()]
if use_new_frontend and use_legacy_frontend:
raise Error('Options "use_new_frontend" and "use_legacy_frontend" must not be used simultaneously.')
if use_legacy_frontend:
legacy_path_error('Option "use_legacy_frontend" was used, but legacy frontends are not available. ')
moc_front_end, available_moc_front_ends = get_moc_frontends(argv)
if not moc_front_end and use_new_frontend:
raise Error('Option "use_new_frontend" is specified but the Model Conversion API is unable to find new frontend. '
'Please ensure that your environment contains new frontend for the input model format or '
'try to install openvino-dev and convert the model using convert_model() from openvino.tools.mo.')
is_tf, is_caffe, is_mxnet, is_kaldi, is_onnx = \
deduce_legacy_frontend_by_namespace(argv) if not moc_front_end else [False, False, False, False, False]
legacy_framework_check(is_caffe, is_mxnet, is_kaldi)
is_legacy_frontend = any([is_tf, is_caffe, is_mxnet, is_kaldi, is_onnx])
# handle a default case, i.e. use_new_frontend and use_legacy_frontend are not specified, when no frontend is found
if not is_legacy_frontend and not moc_front_end:
legacy_frameworks = ['tf', 'caffe', 'mxnet', 'kaldi', 'onnx']
frameworks = list(set(legacy_frameworks + available_moc_front_ends))
if not argv.framework:
raise Error('Framework name can not be deduced from the given options: {}={}. '
'Please use "framework" with one from the list: {}.',
'"input_model="', argv.input_model, frameworks)
elif argv.framework not in frameworks:
if argv.framework == 'ir':
raise Error('OpenVINO IR is passed as input_model in convert_model/mo, the IR doesn\'t need '
'conversion, please use it in runtime for inference with read_model/compile_model.')
raise Error('Framework {} is not a valid target. Please use "framework" with one from the list: {}. ' +
refer_to_faq_msg(15), argv.framework, frameworks)
if is_tf and not argv.input_model and not argv.saved_model_dir and not argv.input_meta_graph:
raise Error('Path to input model or saved model dir is required: use "input_model", "saved_model_dir" or '
'"input_meta_graph"')
elif is_onnx and not argv.input_model:
raise Error('Path to input model is required: use "input_model".')
# TODO: This function looks similar to another one. Check for code duplicates.
log.debug("Model Conversion API started")
log.debug('Output model name would be {}{{.xml, .bin}}'.format(argv.output_model))
log.debug('Output model name would be {}{{.xml, .bin}}'.format(argv.model_name))
if argv.verbose:
print_argv(argv, argv.output_model)
if not argv.silent:
print_argv(argv, is_caffe, is_tf, is_mxnet, is_kaldi, is_onnx, argv.model_name)
argv.data_type = 'FP32' # if compression was enabled will be restored back to 'FP16' after apply_offline_transformations
# This is just to check that transform key is valid and transformations are available
check_available_transforms(parse_transform(argv.transform))
if argv.scale and argv.scale_values:
raise Error(
'Both "scale" and "scale_values" are defined. Specify either scale factor or scale values per input ' +
'channels. ' + refer_to_faq_msg(19))
if argv.scale and argv.scale < 1.0:
log.error("The scale value is less than 1.0. This is most probably an issue because the scale value specifies "
"floating point value which all input values will be *divided*.", extra={'is_warning': True})
if argv.input_model and (is_tf and argv.saved_model_dir):
raise Error('Both "input_model" and "saved_model_dir" are defined. '
'Specify either input model or saved model directory.')
if is_tf:
if argv.saved_model_tags is not None:
if ' ' in argv.saved_model_tags:
raise Error('Incorrect saved model tag was provided. Specify "saved_model_tags" with no spaces in it')
argv.saved_model_tags = argv.saved_model_tags.split(',')
if hasattr(argv, 'is_python_api_used') and argv.is_python_api_used:
python_api_params_parsing(argv)
else:
argv.inputs_list, argv.placeholder_shapes, argv.placeholder_data_types = get_placeholder_shapes(
argv.input, argv.input_shape, argv.batch)
argv.freeze_placeholder_with_value, argv.input = get_freeze_placeholder_values(
argv.input,
argv.freeze_placeholder_with_value)
argv.unnamed_freeze_placeholder_with_value = {}
params_parsing(argv)
argv.output = argv.output.split(',') if argv.output else None
argv.layout_values = get_layout_values(argv.layout, argv.source_layout, argv.target_layout)
mean_values = parse_tuple_pairs(argv.mean_values)
scale_values = parse_tuple_pairs(argv.scale_values)
mean_scale = get_mean_scale_dictionary(mean_values, scale_values, argv.input)
argv.mean_scale_values = mean_scale
log.debug("Placeholder shapes : {}".format(argv.placeholder_shapes))
return argv
def check_fallback(argv: argparse.Namespace):
fallback_reasons = {}
# Some frontend such as PDPD does not have legacy path so it has no reasons to fallback
if not any(deduce_legacy_frontend_by_namespace(argv)):
return fallback_reasons
if argv.use_new_frontend:
return fallback_reasons
fallback_reasons['extensions'] = legacy_extensions_used
fallback_reasons['transformations_config'] = legacy_transformations_config_used
fallback_reasons['tensorflow_custom_operations_config_update'] = tensorflow_custom_operations_config_update_used
reasons = [reason for reason, is_applicable in fallback_reasons.items() if is_applicable(argv)]
return reasons
def update_fallback_with_conversion_error(use_new_frontend: bool, is_tf: bool, ex_msg: str, fallback_reasons: list):
import re
if not is_tf:
# this sort of fallback is only used by TensorFlow Frontend
return False
if use_new_frontend:
# this option forces to use new TensorFlow Frontend
# so it is not possible for the fallback
return False
# for TensorFlow FE we have a set of operations that should lead to the fallback to the legacy
conversion_error_re = r"^(\[TensorFlow\ Frontend\]\ Internal\ error\,\ no\ translator\ found\ for\ operation\(s\)\:\ )((\w+)(\,\ \w+)*)$"
conversion_error_match = re.findall(conversion_error_re, ex_msg, re.MULTILINE)
all_fallback_operations = [
# corresponds to TF1 While operation
"TensorArrayScatterV3", "TensorArrayV3", "TensorArraySizeV3", "TensorArrayGatherV3",
"LoopCond", "Enter", "NextIteration", "Exit",
# corresponds to operations with complex tensors
"FFT", "FFT2D", "FFT3D", "IFFT", "IFFT2D", "IFFT3D",
"RFFT", "RFFT2D", "RFFT3D", "IRFFT", "IRFFT2D", "IRFFT3D",
"Complex", "ComplexAbs", "Real", "Imag",
]
if len(conversion_error_match) < 1 or len(conversion_error_match[0]) != 4:
# no match for the fallback by unsupported operation
return False
unsupported_operations = conversion_error_match[0][1].replace(" ", "").split(",")
fallback_operations = [operation for operation in unsupported_operations if operation in all_fallback_operations]
if len(fallback_operations) == 0:
return False
fallback_reasons.append("Fallback to the legacy TF FE due to operation(s): " + ', '.join(fallback_operations))
return True
def get_default_frontends():
# Set which frontend to use by default, values should be 'new' or 'legacy'
default_frontends = {
'onnx': 'new',
'tf': 'new'
}
return default_frontends
def get_moc_frontends(argv: argparse.Namespace):
fem = argv.feManager
# Read user flags:
use_legacy_frontend = argv.use_legacy_frontend
use_new_frontend = argv.use_new_frontend
if not fem or use_legacy_frontend:
if not fem:
return None, []
available_moc_front_ends = get_available_front_ends(fem)
if not argv.framework and argv.input_model:
if argv.framework:
moc_front_end = fem.load_by_framework(argv.framework) # WA to prevent process hanging. Need to remove when 115994 fixed.
moc_front_end = fem.load_by_framework(argv.framework)
return moc_front_end, available_moc_front_ends
if argv.input_model:
if isinstance(argv.input_model, (tuple, list)) and len(argv.input_model) == 2:
moc_front_end = fem.load_by_model([argv.input_model[0], argv.input_model[1]]) # WA to prevent process hanging. Need to remove when 115994 fixed.
moc_front_end = fem.load_by_model([argv.input_model[0], argv.input_model[1]]) # TODO: Pass all input model parts
else:
moc_front_end = fem.load_by_model(argv.input_model) # WA to prevent process hanging. Need to remove when 115994 fixed.
moc_front_end = fem.load_by_model(argv.input_model)
if not moc_front_end:
return None, available_moc_front_ends
argv.framework = moc_front_end.get_name()
elif argv.framework in available_moc_front_ends:
moc_front_end = fem.load_by_framework(argv.framework)
else:
return None, []
default_frontends = get_default_frontends()
# Disable MOC frontend if default is set to legacy and no user override
if default_frontends.get(moc_front_end.get_name()) == 'legacy' and not use_new_frontend:
return None, available_moc_front_ends
# This check as a workaround to skip IR frontend
if not moc_front_end.get_name() in available_moc_front_ends:
return None, available_moc_front_ends
@ -332,59 +122,32 @@ def get_moc_frontends(argv: argparse.Namespace):
def prepare_ir(argv: argparse.Namespace):
# TODO: remove this workaround once new TensorFlow frontend supports non-frozen formats: checkpoint, MetaGraph, and SavedModel
# Now it converts all TensorFlow formats to the frozen .pb format in case new TensorFlow frontend
is_tf, _, _, _, _ = deduce_legacy_frontend_by_namespace(argv)
argv = arguments_post_parsing(argv)
t = tm.Telemetry()
graph = None
fallback_reasons = []
if isinstance(argv.input_model, (tuple, list)) and len(argv.input_model) == 1:
argv.input_model = argv.input_model[0]
moc_front_end, available_moc_front_ends = get_moc_frontends(argv)
if moc_front_end:
fallback_reasons = check_fallback(argv)
if len(fallback_reasons) == 0:
if is_tf and tf_frontend_with_python_bindings_installed and \
type_supported_by_tf_fe(argv.input_model):
# TODO: Should be moved to the same place where paddle and pytorch handle their objects
if argv.framework == 'tf' and argv.is_python_object and type_supported_by_tf_fe(argv.input_model):
argv.input_model = create_tf_graph_iterator(argv.input_model,
argv.placeholder_shapes,
argv.placeholder_data_types,
getattr(argv, "example_input", None))
try:
t.send_event("mo", "conversion_method", moc_front_end.get_name() + "_frontend")
moc_front_end.add_extension(TelemetryExtension("mo", t.send_event, t.send_error, t.send_stack_trace))
moc_front_end.add_extension(ProgressReporterExtension(progress_printer(argv)))
if legacy_transformations_config_used(argv):
raise Error('Legacy extensions are not supported for the new frontend')
if legacy_extensions_used(argv):
raise Error('Legacy transformations configuration is not supported for the new frontend')
if tensorflow_custom_operations_config_update_used(argv) and is_tf:
raise Error('TensorFlow custom operation config is not supported for the new frontend')
if new_extensions_used(argv):
for extension in argv.extensions:
moc_front_end.add_extension(extension)
ngraph_function = moc_pipeline(argv, moc_front_end)
return graph, ngraph_function
except OpConversionFailure as ex:
# in some set of operations (TF1 While), we have to fallback to the Legacy TensorFlow Frontend
# this is the second attempt for the fallback
if not update_fallback_with_conversion_error(argv.use_new_frontend, is_tf, str(ex), fallback_reasons):
# re-throw exception for all frontends except TensorFlow FE
# and in case unexpected conversion failures
raise
ov_model = moc_pipeline(argv, moc_front_end)
return ov_model
if len(fallback_reasons) > 0:
reasons_message = ", ".join(fallback_reasons)
t.send_event("mo", "fallback_reason", reasons_message)
log.warning("The IR preparation cannot be executed with new frontend. "
f"The detailed reason why fallback to legacy is needed: not supported {reasons_message} were used. " +
refer_to_faq_msg(105))
assert not hasattr(argv, 'is_fallback'), '`is_fallback` argument must not exist.'
argv.is_fallback = True
if not argv.input_model:
raise Error('No input model is provided')
t.send_event("mo", "conversion_method", "mo_legacy")
legacy_path_error("The provided model cannot be converted with new frontend, as fallback to legacy is needed. ")
return None, None
raise Error('Cannot recognize input model.')
def check_model_object(argv):
@ -405,6 +168,10 @@ def check_model_object(argv):
pass
import io
# FIXME: Consuming any io.BytesIO object as an ONNX model is too dengerous and
# can conflict with others in the future (not future proof).
# TODO: Refer to https://onnx.ai/onnx/intro/python.html to find examples with
# real ONNX python objects. ONNX model has onnx.onnx_ml_pb2.ModelProto type.
if isinstance(model, io.BytesIO):
return 'onnx'
@ -419,24 +186,18 @@ def check_model_object(argv):
def driver(argv: argparse.Namespace, non_default_params: dict):
init_logger(argv.log_level.upper(), argv.silent)
if not hasattr(argv, 'log_level'):
argv.log_level = 'ERROR'
init_logger(argv.log_level.upper(), argv.verbose)
# Log dictionary with non-default cli parameters where complex classes are excluded.
log.debug(str(non_default_params))
start_time = datetime.datetime.now()
graph, ngraph_function = prepare_ir(argv)
legacy_path = False
if graph is not None:
legacy_path_error("")
else:
res_ngraph_function = moc_emit_ir(ngraph_function, argv)
ov_model = moc_emit_ir(prepare_ir(argv), argv)
if res_ngraph_function is None:
return res_ngraph_function
if not argv.silent:
if argv.verbose:
elapsed_time = datetime.datetime.now() - start_time
print('[ SUCCESS ] Total execution time: {:.2f} seconds. '.format(elapsed_time.total_seconds()))
try:
@ -448,7 +209,7 @@ def driver(argv: argparse.Namespace, non_default_params: dict):
except ImportError:
pass
return res_ngraph_function, legacy_path
return ov_model
def args_dict_to_list(cli_parser, **kwargs):
@ -536,19 +297,17 @@ def show_mo_convert_help():
print()
def input_model_is_object(argv):
# Input model can be set as object only for "input_model" parameter.
# "saved_model_dir" or meta specific options are only used to store paths to the input model.
if 'input_model' not in argv:
def input_model_is_object(input_model):
if input_model == ():
return False
if isinstance(argv['input_model'], (str, Path)):
return False
if argv['input_model'] is None:
if isinstance(input_model, (str, Path)):
return False
if isinstance(input_model, (tuple, list)):
return all(input_model_is_object(part) for part in input_model)
return True
def python_api_params_parsing(argv: argparse.Namespace):
def params_parsing(argv: argparse.Namespace):
"""
Parses params passed to convert_model and wraps resulting values into dictionaries or lists.
After working of this method following values are set in argv:
@ -583,15 +342,12 @@ def python_api_params_parsing(argv: argparse.Namespace):
argv.inputs_list = input_names_list
argv.input = ','.join(input_names_list)
# Parse input_shape param and update InputCutInfo list
input_shape_to_input_cut_info(argv.input_shape, inputs)
# Parse freeze_placeholder_with_value.
# values for freezing can be set both by named and unnamed approach if
# 'input' was used without names and 'freeze_placeholder_with_value' was used with names.
# So named and unnamed values are stored separately.
argv.freeze_placeholder_with_value, argv.unnamed_freeze_placeholder_with_value = \
freeze_placeholder_to_input_cut_info(argv.freeze_placeholder_with_value, inputs)
freeze_placeholder_to_input_cut_info(inputs)
if len(input_names_list) > 0:
# Named inputs case
@ -638,7 +394,7 @@ def python_api_params_parsing(argv: argparse.Namespace):
def pack_params_to_args_namespace(args: dict, cli_parser: argparse.ArgumentParser):
if len(args) > 0:
args_string = params_to_string(**args)
argv, _ = cli_parser.parse_known_args(args_dict_to_list(cli_parser, **args_string))
argv, _ = cli_parser.parse_known_args(args_dict_to_list(cli_parser, **args_string)) # FIXME: input_model_args can be a list
# get list of all available params for convert_model()
all_params = {}
@ -659,36 +415,12 @@ def pack_params_to_args_namespace(args: dict, cli_parser: argparse.ArgumentParse
return argv
def update_args_for_saved_model_dir(args: dict):
"""
If directory is set in 'input_model' argument, the directory is considered as TF saved model.
In this case this method updates args and moves saved model directory to 'saved_model_dir' param.
:param args: dictionary with arguments from user
"""
if 'saved_model_dir' in args and args['saved_model_dir'] is not None and \
'input_model' in args and args['input_model'] is not None:
raise Error("Both \"input_model\" and \"saved_model_dir\" are defined. "
"Please specify either \"input_model\" or \"saved_model_dir\" directory.")
if 'input_model' in args and isinstance(args['input_model'], (str, Path)) and os.path.isdir(args['input_model']):
args['saved_model_dir'] = args['input_model']
args['input_model'] = None
def silent_is_false(argv: argparse.Namespace):
return argv is not None and hasattr(argv, 'silent') and argv.silent is False
def framework_is_tf(args, argv):
if input_model_is_object(args) and check_model_object(args) == "tf":
return True
if argv is not None:
is_tf, _, _, _, _ = deduce_legacy_frontend_by_namespace(argv)
return is_tf
return False
def is_verbose(argv: argparse.Namespace):
return argv is not None and hasattr(argv, 'verbose') and argv.verbose
def _convert(cli_parser: argparse.ArgumentParser, args, python_api_used):
# FIXME: It doesn't work when -h is passed
if 'help' in args and args['help']:
show_mo_convert_help()
return None, None
@ -701,9 +433,16 @@ def _convert(cli_parser: argparse.ArgumentParser, args, python_api_used):
# before arg parser deliver log_level requested by user
init_logger('ERROR', False)
argv = None
# Minimize modifications among other places in case if multiple pieces are passed as input_model
if python_api_used:
if 'input_model' not in args:
args['input_model'] = ()
if isinstance(args['input_model'], (tuple, list)) and len(args['input_model']) == 1:
args['input_model'] = args['input_model'][0]
try:
model_framework = None
inp_model_is_object = input_model_is_object(args)
inp_model_is_object = input_model_is_object(args['input_model']) if python_api_used else False
if inp_model_is_object:
model_framework = check_model_object(args)
if model_framework == "pytorch":
@ -714,12 +453,13 @@ def _convert(cli_parser: argparse.ArgumentParser, args, python_api_used):
raise AssertionError(
"'example_inputs' argument is not recognized, maybe you meant to provide 'example_input'?")
decoder = get_pytorch_decoder(args['input_model'], parse_input_shapes(args), example_inputs, args)
get_pytorch_decoder(args['input_model'], example_inputs, args)
if model_framework == "paddle":
example_inputs = None
if 'example_input' in args and args['example_input'] is not None:
example_inputs = args['example_input']
#TODO: Check what example_outputs is and remove if not needed
example_outputs = None
if 'example_output' in args and args['example_output'] is not None:
example_outputs = args['example_output']
@ -727,45 +467,27 @@ def _convert(cli_parser: argparse.ArgumentParser, args, python_api_used):
example_outputs)
pdmodel = paddle_runtime_converter.convert_paddle_to_pdmodel()
args['input_model'] = pdmodel
args['framework'] = model_framework
update_args_for_saved_model_dir(args)
argv = pack_params_to_args_namespace(args, cli_parser)
argv.framework = model_framework
argv.is_python_object = inp_model_is_object
argv.feManager = FrontEndManager()
frameworks = list(set(['tf', 'caffe', 'mxnet', 'kaldi', 'onnx'] + (get_available_front_ends(argv.feManager)
if argv.feManager else [])))
framework = argv.framework if hasattr(argv, 'framework') and argv.framework is not None else framework
if framework is not None:
assert framework in frameworks, "error: argument \"framework\": invalid choice: '{}'. " \
"Expected one of {}.".format(framework, frameworks)
setattr(argv, 'framework', framework)
# send telemetry with params info
send_params_info(argv, cli_parser)
non_default_params = get_non_default_params(argv, cli_parser)
check_legacy_args(non_default_params, python_api_used)
argv.is_python_api_used = python_api_used
if inp_model_is_object:
argv.model_name = "model"
if not hasattr(argv, "model_name") or argv.model_name is None:
argv.model_name = get_model_name_from_args(argv)
argv.output_model = "model" # TODO: Consider removing
if not hasattr(argv, "output_model") or argv.output_model is None:
argv.output_model = get_model_name_from_args(argv)
if model_framework is not None:
if argv.framework is not None:
if argv.framework != model_framework:
raise Error("Provided model does not correspond to provided framework. The provided "
"framework is {}, the model type is {} which is expected to be {} framework.".format(
argv.framework,
type(argv.input_model),
model_framework))
else:
argv.framework = model_framework
ov_model, legacy_path = driver(argv, {"conversion_parameters": non_default_params})
ov_model = driver(argv, {"conversion_parameters": non_default_params})
if inp_model_is_object and model_framework == "paddle":
if paddle_runtime_converter:
@ -773,30 +495,22 @@ def _convert(cli_parser: argparse.ArgumentParser, args, python_api_used):
# add MO meta data to model
ov_model.set_rt_info(get_rt_version(), "Runtime_version")
ov_model.set_rt_info(str(legacy_path), "legacy_frontend")
for key, value in non_default_params.items():
ov_model.set_rt_info(str(value), ["conversion_parameters", str(key)])
if silent_is_false(argv) or not python_api_used:
if is_verbose(argv) or not python_api_used:
if 'compress_to_fp16' in argv and argv.compress_to_fp16:
print(get_compression_message())
ov_update_message = get_ov_update_message()
ov_api20_message = get_ov_api20_message()
if ov_update_message is not None:
print(ov_update_message)
if ov_api20_message is not None and ov_model is not None:
print(ov_api20_message)
is_fallback = getattr(argv, 'is_fallback', False)
if not argv.use_legacy_frontend and framework_is_tf(args, argv) and not is_fallback:
# now TF FE is default frontend for TensorFlow models conversion
print(get_tf_fe_message())
send_conversion_result('success')
return ov_model, argv
except Exception as e:
if silent_is_false(argv) or not python_api_used:
if is_verbose(argv) or not python_api_used:
if isinstance(e, (FileNotFoundError, NotADirectoryError)):
log.error('File {} was not found'.format(str(e).split('No such file or directory:')[1]))
log.debug(traceback.format_exc())
@ -810,17 +524,17 @@ def _convert(cli_parser: argparse.ArgumentParser, args, python_api_used):
log.error("-------------------------------------------------")
log.error("----------------- INTERNAL ERROR ----------------")
log.error("Unexpected exception happened.")
log.error("Please contact Model Conversion API developers and forward the following information:")
log.error("Please verify parameters and environment.")
log.error("If you think this is a bug, please create new ticket here: ")
log.error("https://github.com/openvinotoolkit/openvino/issues.")
log.error("-------------- DETAILED INFORMATION -------------")
log.error(str(e))
log.error(traceback.format_exc())
log.error("---------------- END OF BUG REPORT --------------")
log.error("----------------- END OF REPORT -----------------")
log.error("-------------------------------------------------")
is_fallback = getattr(argv, 'is_fallback', False) if argv is not None else False
if not argv.use_legacy_frontend and framework_is_tf(args, argv) and not is_fallback:
print(get_try_legacy_fe_message())
send_conversion_result('fail')
if python_api_used:
raise e.with_traceback(None)
raise e
else:
return None, argv

5
tools/ovc/openvino/tools/ovc/error.py
View File

@ -46,8 +46,3 @@ def classify_error_type(e):
if m:
return m.group(0)
return "undefined"
def legacy_path_error(functionality_description):
raise Exception("{}Please try to install openvino-dev and use convert_model() "
"from openvino.tools.mo.".format(functionality_description))

23
tools/ovc/openvino/tools/ovc/get_ov_update_message.py
View File

@ -16,24 +16,6 @@ def get_ov_update_message():
return msg_fmt.format(link) if current_date >= expected_update_date else None
def get_ov_api20_message():
link = "https://docs.openvino.ai/2023.0/openvino_2_0_transition_guide.html"
message = '[ INFO ] The model was converted to IR v11, the latest model format that corresponds to the source DL framework ' \
'input/output format. While IR v11 is backwards compatible with OpenVINO Inference Engine API v1.0, ' \
'please use API v2.0 (as of 2022.1) to take advantage of the latest improvements in IR v11.\n' \
'Find more information about API v2.0 and IR v11 at {}'.format(link)
return message
def get_tf_fe_message():
link = "https://docs.openvino.ai/2023.0/openvino_docs_MO_DG_TensorFlow_Frontend.html"
message = '[ INFO ] IR generated by new TensorFlow Frontend is compatible only with API v2.0. Please make sure to use API v2.0.\n' \
'Find more information about new TensorFlow Frontend at {}'.format(link)
return message
def get_compression_message():
link = "https://docs.openvino.ai/2023.0/openvino_docs_MO_DG_FP16_Compression.html"
message = '[ INFO ] Generated IR will be compressed to FP16. ' \
@ -41,8 +23,3 @@ def get_compression_message():
'by removing argument "compress_to_fp16" or set it to false "compress_to_fp16=False".\n' \
'Find more information about compression to FP16 at {}'.format(link)
return message
def get_try_legacy_fe_message():
message = '[ INFO ] You can also try to install openvino-dev and use convert_model from openvino.tools.mo.\n'
return message

92
tools/ovc/openvino/tools/ovc/help.py
View File

@ -2,19 +2,18 @@
# SPDX-License-Identifier: Apache-2.0
def get_convert_model_help_specifics():
from openvino.tools.ovc.cli_parser import CanonicalizeTransformationPathCheckExistenceAction, \
CanonicalizePathCheckExistenceAction, CanonicalizeExtensionsPathCheckExistenceAction, \
CanonicalizePathCheckExistenceIfNeededAction, readable_file_or_dir, readable_dirs_or_files_or_empty, \
check_positive
from openvino.tools.ovc.cli_parser import CanonicalizePathCheckExistenceAction, CanonicalizeExtensionsPathCheckExistenceAction, \
readable_file_or_dir, readable_dirs_or_files_or_empty
from openvino.tools.ovc.version import VersionChecker
return {
'input_model':
{'description':
'Tensorflow*: a file with a pre-trained model '
'(binary or text .pb file after freezing). '
'Caffe*: a model proto file with model weights.', 'action': CanonicalizePathCheckExistenceAction,
'Input model file(s) from TensorFlow, ONNX, PaddlePaddle. '
'Use openvino.convert_model in Python to convert models from Pytorch.'
'',
'action': CanonicalizePathCheckExistenceAction,
'type': readable_file_or_dir,
'aliases': {'-w', '-m'}},
'aliases': {}},
'input_shape':
{'description':
'Input shape(s) that should be fed to an input node(s) '
@ -38,10 +37,11 @@ def get_convert_model_help_specifics():
'Alternatively, specify shapes with the --input option.'},
'input':
{'description':
'Quoted list of comma-separated input nodes names with '
'shapes, data types, and values for freezing. The order '
'of inputs in converted model is the same as order of '
'specified operation names. The shape and value are '
'Information for model input required for model conversion. '
'This is a comma separated list with optional '
'input names, shapes, data types, and values for freezing. '
'The order of inputs in converted model will match the order of '
'specified inputs. The shape and value are '
'specified as comma-separated lists. The data type of '
'input node is specified in braces and can have one of '
'the values: f64 (float64), f32 (float32), f16 '
@ -62,38 +62,6 @@ def get_convert_model_help_specifics():
'\"node_name2\" with the value [20,15] of the int32 type '
'and shape [2]: \n '
'\"0:node_name1[3,4],node_name2:1[2]{i32}->[20,15]\".'},
'mean_values':
{'description':
'Mean values to be used for the input image per '
'channel. Values to be provided in the (R,G,B) or '
'[R,G,B] format. Can be defined for desired input of '
'the model, for example: "--mean_values '
'data[255,255,255],info[255,255,255]". The exact '
'meaning and order of channels depend on how the '
'original model was trained.'},
'scale_values':
{'description':
'Scale values to be used for the input image per '
'channel. Values are provided in the (R,G,B) or [R,G,B] '
'format. Can be defined for desired input of the model, '
'for example: "--scale_values '
'data[255,255,255],info[255,255,255]". The exact '
'meaning and order of channels depend on how the '
'original model was trained. If both --mean_values and '
'--scale_values are specified, the mean is subtracted '
'first and then scale is applied regardless of the '
'order of options in command line.'},
'source_layout':
{'description':
'Layout of the input or output of the model in the '
'framework. Layout can be specified in the short form, '
'e.g. nhwc, or in complex form, e.g. \"[n,h,w,c]\". '
'Example for many names: \"in_name1([n,h,w,c]),in_name2('
'nc),out_name1(n),out_name2(nc)\". Layout can be '
'partially defined, \"?\" can be used to specify '
'undefined layout for one dimension, \"...\" can be used '
'to specify undefined layout for multiple dimensions, '
'for example \"?c??\", \"nc...\", \"n...c\", etc.'},
'transform':
{'description':
'Apply additional transformations. Usage: \"--transform '
@ -115,47 +83,19 @@ def get_convert_model_help_specifics():
'those that are placed at the default location, pass an empty string.',
'action': CanonicalizeExtensionsPathCheckExistenceAction,
'type': readable_dirs_or_files_or_empty},
'transformations_config':
{'description':
'Use the configuration file with transformations '
'description. Transformations file can be specified as '
'relative path from the current directory, as absolute '
'path or as arelative path from the mo root directory.',
'action': CanonicalizeTransformationPathCheckExistenceAction},
'counts':
{'action': CanonicalizePathCheckExistenceIfNeededAction},
'version':
{'action': 'version',
'version': 'Version of Model Optimizer is: {}'.format(VersionChecker().get_ie_version())},
'scale':
{'type': float,
'aliases': {'-s'}},
'batch':
{'type': check_positive,
'aliases': {'-b'}},
'input_proto':
{'aliases': {'-d'}},
'log_level':
{'choices': ['CRITICAL', 'ERROR', 'WARN', 'WARNING', 'INFO', 'DEBUG', 'NOTSET']}
#FIXME: Why the following is not accessible from arg parser?
'version': 'OpenVINO Model Converter (ovc) {}'.format(VersionChecker().get_ie_version())},
}
# TODO: remove this when internal converting of params to string is removed
# TODO: remove this when internal converting of params to string is removed <-- DO IT
def get_to_string_methods_for_params():
from openvino.tools.ovc.cli_parser import path_to_str_or_object, str_list_to_str, \
mean_scale_value_to_str, source_target_layout_to_str, layout_param_to_str, transform_param_to_str, \
extensions_to_str_or_extensions_class, batch_to_int, transformations_config_to_str
from openvino.tools.ovc.cli_parser import path_to_str_or_object, str_list_to_str, extensions_to_str_or_extensions_class
return {
'input_model': path_to_str_or_object,
'output': str_list_to_str,
'mean_values': mean_scale_value_to_str,
'scale_values': mean_scale_value_to_str,
'source_layout': source_target_layout_to_str,
'target_layout': source_target_layout_to_str,
'layout': layout_param_to_str,
'transform': transform_param_to_str,
'extensions': extensions_to_str_or_extensions_class,
'batch': batch_to_int,
'transformations_config': transformations_config_to_str,
'saved_model_tags': str_list_to_str
}

73
tools/ovc/openvino/tools/ovc/logger.py
View File

@ -64,10 +64,10 @@ class TagFilter(log.Filter):
return True # if regex wasn't set print all logs
def init_logger(lvl: str, silent: bool):
def init_logger(lvl: str, verbose: bool):
global handler_num
log_exp = os.environ.get('MO_LOG_PATTERN')
if silent:
if not verbose:
lvl = 'ERROR'
fmt = LvlFormatter(lvl=lvl)
handler = log.StreamHandler()
@ -89,72 +89,3 @@ def restore_logger_state(state: tuple):
logger.setLevel(level)
logger.filters = filters
logger.handlers = handlers
def progress_bar(function: callable):
"""
Decorator for model conversion pipeline progress display
Works in combination with function: mo.utils.class_registration.apply_transform
"""
def wrapper(*args, **kwargs):
for arg in ['graph', 'curr_transform_num', 'num_transforms']:
msg = 'Progress bar decorator is enabled for Model Conversion API transformation applying cycle only. ' \
'Argument `{}` {}'
assert arg in kwargs, msg.format(arg, 'is missing')
assert kwargs[arg] is not None, msg.format(arg, 'should not be None')
if 'progress' in kwargs['graph'].graph['cmd_params'] and kwargs['graph'].graph['cmd_params'].progress:
bar_len = 20
total_replacers_count = kwargs['num_transforms']
def progress(i):
return int((i + 1) / total_replacers_count * bar_len)
def percent(i):
return (i + 1) / total_replacers_count * 100
end = '' if not kwargs['graph'].graph['cmd_params'].stream_output else '\n'
curr_i = kwargs['curr_transform_num']
print('\rProgress: [{:{}}]{:>7.2f}% done'.format('.' * progress(curr_i), bar_len, percent(curr_i)), end=end)
sys.stdout.flush()
function(*args, **kwargs)
return wrapper
def progress_printer(argv: Namespace):
"""
A higher-order factory function returning a configurable callback displaying a progress bar
Depending on the configuration stored in 'argv' the progress bar can be one-line, multi-line, or silent.
"""
def _progress_bar(progress, total, completed, endline):
bar_len = 20
def dots():
return '.' * int(progress * bar_len)
print('\rProgress: [{:{}}]{:>7.2f}% done'.format(dots(), bar_len, progress*100), end=endline)
sys.stdout.flush()
def no_progress_bar(progress, total, completed):
""" A 'dummy' progressbar which doesn't print anything """
pass
def oneline_progress_bar(progress, total, completed):
""" A callback that always prints the progress in the same line (mimics real GUI progress bar)"""
_progress_bar(progress, total, completed, '')
def newline_progress_bar(progress, total, completed):
""" A callback that prints an updated progress bar in separate lines """
_progress_bar(progress, total, completed, '\n')
if "progress" in argv and argv.progress:
if "stream_output" in argv and argv.stream_output:
return newline_progress_bar
else:
return oneline_progress_bar
else:
return no_progress_bar

5
tools/ovc/openvino/tools/ovc/main.py
View File

@ -21,13 +21,12 @@ def main():
if ngraph_function is None:
return 1
output_dir = argv.output_dir if argv.output_dir != '.' else os.getcwd()
model_path_no_ext = os.path.normpath(os.path.join(output_dir, argv.model_name))
output_dir = os.getcwd()
model_path_no_ext = os.path.normpath(os.path.join(output_dir, argv.output_model))
model_path = model_path_no_ext + '.xml'
serialize(ngraph_function, model_path.encode('utf-8'), model_path.replace('.xml', '.bin').encode('utf-8'))
print('[ SUCCESS ] Generated IR version {} model.'.format(get_ir_version()))
print('[ SUCCESS ] XML file: {}'.format(model_path))
print('[ SUCCESS ] BIN file: {}'.format(model_path.replace('.xml', '.bin')))
return 0

8
tools/ovc/openvino/tools/ovc/moc_frontend/moc_emit_ir.py
View File

@ -4,7 +4,6 @@
import argparse
from openvino.runtime import Model # pylint: disable=no-name-in-module,import-error
from openvino.tools.ovc.cli_parser import parse_transform
from openvino.tools.ovc.moc_frontend.preprocessing import apply_preprocessing
@ -21,15 +20,14 @@ def moc_emit_ir(ngraph_function: Model, argv: argparse.Namespace):
from openvino._offline_transformations import compress_quantize_weights_transformation # pylint: disable=no-name-in-module,import-error
compress_quantize_weights_transformation(ngraph_function)
if argv.framework == "onnx":
if argv.framework == "onnx": # TODO: Consider removing
# set OldApi map in IR to be executed via OV API 1.x and for parity with legacy MO
params_with_custom_types = [] if argv.placeholder_data_types is None \
else list(argv.placeholder_data_types.keys())
apply_moc_legacy_transformations(ngraph_function, params_with_custom_types)
apply_user_transformations(ngraph_function, parse_transform(argv.transform))
if argv.compress_to_fp16:
# TODO: Move compression to save_model at the level of main function where serialize is called
if not argv.is_python_api_used and argv.compress_to_fp16:
from openvino.tools.ovc.moc_frontend.offline_transformations import compress_model
compress_model(ngraph_function)

2
tools/ovc/openvino/tools/ovc/moc_frontend/offline_transformations.py
View File

@ -4,7 +4,6 @@
import argparse
from typing import List
from openvino.tools.ovc.cli_parser import parse_transform
from openvino.tools.ovc.error import Error
from openvino.runtime import Model # pylint: disable=no-name-in-module,import-error
@ -117,7 +116,6 @@ def apply_offline_transformations(func: Model, argv: argparse.Namespace):
params_with_custom_types = create_params_with_custom_types(argv.packed_user_shapes)
apply_moc_legacy_transformations(func, params_with_custom_types)
apply_user_transformations(func, parse_transform(argv.transform))
if "compress_to_fp16" in argv and argv.compress_to_fp16:
compress_model(func)

100
tools/ovc/openvino/tools/ovc/moc_frontend/pipeline.py
View File

@ -45,19 +45,18 @@ def moc_pipeline(argv: argparse.Namespace, moc_front_end: FrontEnd):
:param: moc_front_end: Loaded Frontend for converting input model
:return: converted nGraph function ready for serialization
"""
input_checkpoint = getattr(argv, 'input_checkpoint', None)
share_weights = getattr(argv, 'share_weights', True)
if argv.input_model and input_checkpoint:
share_weights = getattr(argv, 'share_weights', True) #FIXME: Should be controlled by default value
if isinstance(argv.input_model, (tuple, list)) and len(argv.input_model) == 2:
# frozen format with v1 checkpoints
input_model = moc_front_end.load([argv.input_model, argv.input_checkpoint], share_weights)
elif argv.input_model:
input_model = moc_front_end.load(argv.input_model, share_weights)
elif argv.saved_model_dir:
if argv.saved_model_tags:
input_model = moc_front_end.load([argv.saved_model_dir, argv.saved_model_tags], share_weights)
assert not hasattr(argv, 'saved_model_tags') or not argv.saved_model_tags
input_model = moc_front_end.load([part for part in argv.input_model], share_weights)
elif hasattr(argv, 'saved_model_tags') and argv.saved_model_tags:
input_model = moc_front_end.load([argv.input_model, argv.saved_model_tags], share_weights)
else:
input_model = moc_front_end.load(argv.saved_model_dir, share_weights)
elif argv.input_meta_graph:
input_model = moc_front_end.load(argv.input_model, share_weights)
'''elif argv.input_meta_graph: # TODO: Cover this case
input_model = moc_front_end.load(argv.input_meta_graph, share_weights)
if argv.output:
# Simulate original behavior with freezing model
@ -65,7 +64,7 @@ def moc_pipeline(argv: argparse.Namespace, moc_front_end: FrontEnd):
# need to simulate similar behavior with natively supported model
outputs = fe_output_user_data_repack(input_model, argv.output, moc_front_end.get_name())
input_model.override_all_outputs([x['node'] for x in outputs])
'''
argv.placeholder_shapes, argv.placeholder_data_types, argv.freeze_placeholder_with_value = convert_params_lists_to_dicts(
input_model, argv.placeholder_shapes, argv.placeholder_data_types,
argv.freeze_placeholder_with_value, argv.unnamed_freeze_placeholder_with_value)
@ -236,83 +235,6 @@ def moc_pipeline(argv: argparse.Namespace, moc_front_end: FrontEnd):
layout_values = update_layout_to_dict(model_inputs, argv.layout_values,
lambda input_place: input_place.get_names())
deferred_batch_names = []
# set batch size for inputs with a static rank
# for all other inputs, set it after shape deduction is performed during model conversion
if argv.batch is not None and argv.batch > 0:
log.debug('Setting batch size to {}'.format(argv.batch))
frozen_input_names = list(freeze_placeholder.keys()) if freeze_placeholder else []
for place in model_inputs:
input_partial_shape = input_model.get_partial_shape(place)
input_names = place.get_names()
joined_name = ' '.join(place.get_names())
assert len(input_names) > 0, "One input place has no names"
# if this input is frozen, there is no need to set the batch
is_frozen_input = len([name for name in input_names if name in frozen_input_names]) > 0
if is_frozen_input:
# skip the frozen input
continue
if not input_partial_shape.rank.is_static:
# found input with dynamic rank, so have to repeat the batch setting after the model conversion
deferred_batch_names += input_names
continue
batch_dim, is_default_index = get_dimension_index_by_label(input_partial_shape,
place.get_names(), layout_values, 'N', 0)
if batch_dim is None:
# skip because no batch dimension exists in the input
continue
if is_default_index:
# if the batch index is chosen by default, we need to ensure that its size equals -1, 0 or 1
validate_batch_in_shape(shape_to_array(input_partial_shape), joined_name)
assert batch_dim < input_partial_shape.rank.get_length(), \
"Incorrect layout is specified for {}:" \
" index of the batch dimension is out of range.".format(input_names[0])
new_partial_shape = copy(input_partial_shape)
new_partial_shape[batch_dim] = argv.batch
log.debug('Input: {}, Old shape: {}, New shape: {}'.format(
joined_name, input_partial_shape, new_partial_shape))
input_model.set_partial_shape(place, new_partial_shape)
ov_model = moc_front_end.convert(input_model)
if argv.batch is not None and argv.batch > 0 and len(deferred_batch_names) > 0:
# Frontend convert method can include reverse infer functionality that can deduce undefined input shapes
# so try to repeat batch setting again
reshape_dict = {}
log.debug('Deferred batch setting to size {}'.format(argv.batch))
is_batch_clarified = False
for model_input in ov_model.inputs:
input_name = model_input.any_name
input_partial_shape = model_input.get_partial_shape()
if input_name in deferred_batch_names and input_partial_shape.rank.is_static:
# update input shape with the specified batch for input that originally has dynamic rank
batch_dim, is_default_index = get_dimension_index_by_label(input_partial_shape,
model_input.get_names(),
layout_values, 'N', 0)
if batch_dim is None:
continue
if is_default_index:
# if the batch index is chosen by default, we need to ensure that its size equals -1, 0 or 1
validate_batch_in_shape(shape_to_array(input_partial_shape), input_name)
assert batch_dim < input_partial_shape.rank.get_length(), \
"Incorrect layout is specified for {}: " \
"index of the batch dimension is out of range.".format(input_name)
input_partial_shape[batch_dim] = argv.batch
is_batch_clarified = True
reshape_dict.update({input_name: input_partial_shape})
if is_batch_clarified:
# call reshape only if batch dimension for one of the input is clarified
ov_model.reshape(reshape_dict)
return ov_model

225
tools/ovc/openvino/tools/ovc/moc_frontend/preprocessing.py
View File

@ -14,52 +14,6 @@ from openvino.tools.ovc.error import Error
from openvino.tools.ovc.utils import refer_to_faq_msg
def update_mean_scale_to_dict(input_nodes: list, mean_scale_val, scale):
"""
Internal function. Updates mean/scale values from array to dictionary
:param: input_nodes Inputs of model
:param: mean_scale_val Parsed 'mean_scale_val' object from command line arguments
:param: scale Global scale factor for all inputs from scale command line arguments
"""
if not isinstance(mean_scale_val, dict):
if len(mean_scale_val) != len(input_nodes):
raise Error('Numbers of inputs and mean/scale values do not match. ' + refer_to_faq_msg(61))
data = copy(mean_scale_val)
mean_scale_val = {}
for idx, node in enumerate(input_nodes):
names_list = list(node.get_tensor().get_names())
names_list.sort()
if not names_list:
continue
node_name = names_list[0]
mean_scale_val.update(
{
node_name: {
'mean': data[idx][0],
'scale': data[idx][1]
}
}
)
if scale:
for node in input_nodes:
names_list = list(node.get_tensor().get_names())
names_list.sort()
if not names_list:
continue
node_name = names_list[0]
old_val = mean_scale_val[node_name] if node_name in mean_scale_val else None
mean_scale_val.update(
{
node_name: {
'mean': old_val['mean'] if old_val and 'mean' in old_val else None,
'scale': scale
}
}
)
return mean_scale_val
def check_keys_valid(ov_function: Model, dict_to_validate: dict, search_outputs: bool):
"""
Internal function: checks if keys from cmd line arguments correspond to ov_function's inputs/outputs
@ -183,147 +137,6 @@ def find_channels_dimension(shape: PartialShape, num_channels: int, name: str, l
}
return layout_values
def guess_source_layouts_by_mean_scale(ov_function: Model, layout_values, mean_scale_values: dict):
"""
Internal function. Try to guess source layout for input by its shape and/or framework
:param: ov_function Original model
:param: layout_values Existing source/target layout items specified by user
:param: mean_scale_values Dictionary with mean/scale values defined for each argument
:return: updated layout items with guessed layouts
"""
for ms_name, mean_scale in mean_scale_values.items():
num_channels_mean = len(mean_scale['mean']) if mean_scale['mean'] is not None else 0
num_channels_scale = len(mean_scale['scale']) if hasattr(mean_scale['scale'], '__len__') else 0
if num_channels_mean > 1 and \
num_channels_scale > 1 and \
num_channels_mean is not num_channels_scale:
raise Error('Mean/Scale values for {} have different sizes: {} {}'
.format(ms_name, num_channels_mean, num_channels_scale))
need_guess_channels = num_channels_mean > 1 or num_channels_scale > 1
if not need_guess_channels: # Mean/scale is complex and needs 'channels' specified in layout
continue
num_channels = num_channels_mean if num_channels_mean > 1 else num_channels_scale
for i in range(0, len(ov_function.inputs)):
ov_input = ov_function.input(i)
if not ov_function.get_parameters()[i].layout.empty:
continue
if ms_name not in ov_input.get_tensor().get_names():
continue
layout_item = None
for name in ov_input.get_tensor().get_names():
if name in layout_values:
layout_item = layout_values[name]
break
if layout_item is not None:
# User specified some layout, skip guessing
continue
# Guess layout is applicable only when number of channels is '3'
if num_channels != 3:
raise Error('Can\'t determine channels dimension for {}. '
'When number of mean/scale values is {} (not 3), '
'please specify layout for input manually'.format(ms_name, num_channels))
layout_values = find_channels_dimension(shape=ov_input.get_partial_shape(),
num_channels=num_channels,
name=ms_name,
layout_values=layout_values)
return layout_values
def check_suitable_for_reverse(layout: Layout, ov_input):
"""
Internal function. Checks if input with layout is suitable for reversing channels
:param: layout Existing source/target layout items specified by user
:param: ov_input Model's input
:return: True if reverse channels can be applied to input
"""
if not layout_helpers.has_channels(layout):
return False
if ov_input.get_partial_shape().rank.is_dynamic:
return False
c_idx = layout_helpers.channels_idx(layout)
rank = ov_input.get_partial_shape().rank.get_length()
if c_idx < 0:
c_idx += rank
if c_idx >= rank:
raise Error('Layout {} for input {} is inconsistent with shape {}'.format(
layout, ov_input.get_tensor().get_any_name(), ov_input.get_partial_shape()))
c_num = ov_input.get_partial_shape()[c_idx]
return c_num.is_dynamic or c_num.get_length() == 3
def guess_source_layouts_for_reverse_channels(ov_function: Model, layout_values):
"""
Internal function. Try to guess source layout for input by finding dimension with size=3 (RGB/BGR)
Additionally checks existing layouts and detects suitable inputs for reversing of input channels
:param: ov_function Original model
:param: layout_values Existing source/target layout items specified by user
:return: array with suitable parameters for reversing of input channels
"""
all_params = []
suitable_params = []
for i in range(0, len(ov_function.inputs)):
ov_input = ov_function.input(i)
param_info = [ov_input.get_tensor().get_any_name(), ov_input.get_partial_shape()]
all_params.append(param_info)
if not ov_function.get_parameters()[i].layout.empty:
if check_suitable_for_reverse(ov_function.get_parameters()[i].layout, ov_input):
suitable_params.append(param_info)
continue
layout_item = None
first_name = ov_input.get_tensor().get_any_name()
for name in ov_input.get_tensor().get_names():
if name in layout_values:
layout_item = layout_values[name]
break
if layout_item is not None:
# RIC transformation is applied before changing layout so only source_layout
# should be checked (even is target_layout is also provided)
if layout_item.get('source_layout'):
if check_suitable_for_reverse(Layout(layout_item['source_layout']), ov_input):
suitable_params.append(param_info)
continue
try:
layout_values = find_channels_dimension(shape=ov_input.get_partial_shape(),
num_channels=3,
name=first_name,
layout_values=layout_values)
except Error as e:
log.debug('Reverse input channels guess did not succeed {}'.format(e))
else:
layout = layout_values[first_name].get('source_layout')
if layout and check_suitable_for_reverse(Layout(layout), ov_input):
suitable_params.append(param_info)
if not len(suitable_params):
raise Error('Network has {} inputs overall, but none of them are suitable for input channels reversing.\n'
'Suitable for input channel reversing inputs are 4-dimensional with 3 channels (in case of dynamic '
'dimensions C channel must be provided in a layout for this input)\nAll inputs: {}'.format(
len(all_params), all_params))
elif len(suitable_params) < len(all_params):
log.error('Network has {} inputs overall, but only {} of them are suitable for input channels reversing.\n'
'Suitable for input channel reversing inputs are 4-dimensional with 3 channels (in case of dynamic '
'dimensions C channel must be provided in a layout for this input)\nAll inputs: {}\n'
'Suitable inputs {}'.format(len(all_params), len(suitable_params), all_params, suitable_params),
extra={'is_warning': True})
return suitable_params
def update_tensor_names_to_first_in_sorted_list(values_dict: dict, ov_function: Model):
if not isinstance(values_dict, dict):
return values_dict
@ -372,38 +185,14 @@ def apply_preprocessing(ov_function: Model, argv: argparse.Namespace):
"""
prep = PrePostProcessor(ov_function)
if 'mean_scale_values' in argv and argv.mean_scale_values:
mean_scale_values = argv.mean_scale_values
else:
mean_scale_values = {}
# mean_scale_values stores mean/scale values from command line with names which were set by user.
# For models with single input scale or mean may be unnamed, so name is set by first tensor name from
# names list. This may lead to different naming of preprocessing params for a single node and lead to error.
# To make naming for mean/scale values unified, names provided by user are renamed here
# by the first tensor name from sorted names list.
mean_scale_values = update_tensor_names_to_first_in_sorted_list(mean_scale_values, ov_function)
mean_scale_values = update_mean_scale_to_dict(input_nodes=ov_function.inputs,
mean_scale_val=mean_scale_values,
scale=argv.scale)
# On return, mean_scale_values is a dictionary with input names as key and mean/scale pair as value
# {'inputName': {'mean': [1., 2., 3.], 'scale': [2.]}}
layout_values = {}
if 'layout_values' in argv and argv.layout_values:
layout_values = update_layout_to_dict(ov_function.inputs, argv.layout_values,
lambda ov_input: ov_input.get_tensor().get_names())
check_keys_valid(ov_function=ov_function, dict_to_validate=mean_scale_values, search_outputs=False)
check_keys_valid(ov_function=ov_function, dict_to_validate=layout_values, search_outputs=True)
layout_values = update_layout_is_input_flag(ov_function, layout_values)
layout_values = guess_source_layouts_by_mean_scale(ov_function, layout_values, mean_scale_values)
need_reverse = 'reverse_input_channels' in argv and argv.reverse_input_channels
suitable_params_ric = []
if need_reverse:
suitable_params_ric = guess_source_layouts_for_reverse_channels(ov_function=ov_function,
layout_values=layout_values)
for node_name, layout_value in layout_values.items():
if layout_value.get('source_layout'):
@ -417,20 +206,6 @@ def apply_preprocessing(ov_function: Model, argv: argparse.Namespace):
else:
prep.output(node_name).tensor().set_layout(Layout(layout_value['target_layout']))
# Apply reverse_input_channels
if need_reverse:
for name, _ in suitable_params_ric:
prep.input(name).preprocess().reverse_channels()
log.debug('reverse_input_channels pre-processing applied to {}'.format(name))
for node_name, node_mean_scale_values in mean_scale_values.items():
# Apply mean first, then scale
if node_mean_scale_values['mean'] is not None:
prep.input(node_name).preprocess().mean(node_mean_scale_values['mean'])
if node_mean_scale_values['scale'] is not None:
prep.input(node_name).preprocess().scale(node_mean_scale_values['scale'])
log.debug('Mean/Scale pre-processing applied to {}'.format(node_name))
# Apply pre-processing builder to a function
ov_function = prep.build()

16
tools/ovc/openvino/tools/ovc/moc_frontend/pytorch_frontend_utils.py
View File

@ -8,21 +8,20 @@ import numpy as np
from openvino.runtime import Tensor, Type, PartialShape
from openvino.runtime.utils.types import get_element_type_str
from openvino.tools.ovc.cli_parser import input_to_input_cut_info, input_shape_to_input_cut_info
from openvino.tools.ovc.cli_parser import input_to_input_cut_info
from openvino.tools.ovc.error import Error
from openvino.tools.ovc.moc_frontend.shape_utils import get_static_shape
def get_pytorch_decoder(model, input_shape, example_inputs, args):
def get_pytorch_decoder(model, example_inputs, args):
try:
from openvino.frontend.pytorch.decoder import TorchScriptPythonDecoder
except Exception as e:
log.error("PyTorch frontend loading failed")
raise e
inputs = prepare_torch_inputs(example_inputs, input_shape, args.get("input"), allow_none=True)
inputs = prepare_torch_inputs(example_inputs, args.get("input"), allow_none=True)
decoder = TorchScriptPythonDecoder(model, example_input=inputs)
args['input_model'] = decoder
args["framework"] = "pytorch"
args["example_input"] = inputs
return args
@ -169,7 +168,7 @@ def get_torch_dtype(dtype):
raise Error(f"Unexpected data type for input. Supported torch.dtype, numpy.dtype, ov.Type and str. Got {type(dtype)}")
def prepare_torch_inputs(example_inputs, input_shape, input_info=None, allow_none=False):
def prepare_torch_inputs(example_inputs, input_info=None, allow_none=False):
import torch
inputs = None
if example_inputs is not None:
@ -190,16 +189,15 @@ def prepare_torch_inputs(example_inputs, input_shape, input_info=None, allow_non
inputs[name] = to_torch_tensor(tensor)
else:
inputs = to_torch_tensor(inputs)
elif input_info is not None or input_shape is not None:
elif input_info is not None:
input_info = input_to_input_cut_info(input_info) or []
input_shape_to_input_cut_info(input_shape, input_info)
inputs = []
inputs_with_names = {}
for inp in input_info:
shape = inp.shape
if shape is None:
if not allow_none:
raise Error("Please provide input_shape or example_input for all inputs converting PyTorch model.")
raise Error("Please provide shape in `input` or `example_input` for all inputs converting PyTorch model.")
inputs = None
break
dtype = get_torch_dtype(inp.type)
@ -214,5 +212,5 @@ def prepare_torch_inputs(example_inputs, input_shape, input_info=None, allow_non
inputs = inputs_with_names
else:
if not allow_none:
raise Error("Please provide input_shape or example_input for converting PyTorch model.")
raise Error("Please provide shapes `input` or `example_input` for converting PyTorch model.")
return inputs

38
tools/ovc/openvino/tools/ovc/moc_frontend/shape_utils.py
View File

@ -4,7 +4,6 @@
import numpy as np
from openvino.runtime import PartialShape, Dimension # pylint: disable=no-name-in-module,import-error
from openvino.tools.ovc.error import Error
from openvino.tools.ovc.cli_parser import get_placeholder_shapes, split_shapes
def get_static_shape(shape: [PartialShape, list, tuple], dynamic_value=None):
@ -63,40 +62,3 @@ def get_dynamic_dims(shape: [PartialShape, list, tuple]):
dynamic_dims.append(idx)
return dynamic_dims
def parse_input_shapes(argv):
input_shapes = None
if 'input_shape' in argv and argv['input_shape'] is not None:
shapes = argv['input_shape']
if isinstance(shapes, str):
shapes = ["[{}]".format(x) for x in split_shapes(shapes)]
if isinstance(shapes, list) or isinstance(shapes, tuple):
input_shapes = []
is_single_shape = False
for shape in shapes:
if isinstance(shape, str):
_, shape_tuple, _ = get_placeholder_shapes(argv_input=None, argv_input_shape=shape)
input_shapes.append(shape_tuple)
if is_single_shape:
raise Error("Incorrect format of shape.")
elif isinstance(shape, int) or isinstance(shape, np.int64) or isinstance(shape, Dimension):
is_single_shape = True
input_shapes.append(shape)
else:
input_shapes.append(shape)
if is_single_shape:
return [input_shapes]
else:
return input_shapes
elif isinstance(shapes, PartialShape):
return [shapes]
else:
try:
import torch
if isinstance(shapes, torch.Size): # pylint: disable=no-member
return [shapes]
except ImportError:
raise Error("Unknown type of input shape {}.".format(type(shapes)))
return input_shapes

33
tools/ovc/openvino/tools/ovc/utils.py
View File

@ -101,39 +101,6 @@ def validate_batch_in_shape(shape, layer_name: str):
'You can also specify batch dimension by setting "layout". \n\n')
.format(layer_name, shape))
def deduce_legacy_frontend_by_namespace(argv):
if not hasattr(argv, 'framework') or not argv.framework:
if getattr(argv, 'saved_model_dir', None) or getattr(argv, 'input_meta_graph', None):
argv.framework = 'tf'
elif getattr(argv, 'input_symbol', None) or getattr(argv, 'pretrained_model_name', None):
argv.framework = 'mxnet'
elif getattr(argv, 'input_proto', None):
argv.framework = 'caffe'
elif argv.input_model is None:
raise Error('Path to input model is required: use "input_model".')
else:
argv.framework = guess_framework_by_ext(argv.input_model)
return map(lambda x: argv.framework == x, ['tf', 'caffe', 'mxnet', 'kaldi', 'onnx'])
def guess_framework_by_ext(input_model_path: str) -> int:
if re.match(r'^.*\.caffemodel$', input_model_path):
return 'caffe'
elif re.match(r'^.*\.pb$', input_model_path):
return 'tf'
elif re.match(r'^.*\.pbtxt$', input_model_path):
return 'tf'
elif re.match(r'^.*\.params$', input_model_path):
return 'mxnet'
elif re.match(r'^.*\.nnet$', input_model_path):
return 'kaldi'
elif re.match(r'^.*\.mdl', input_model_path):
return 'kaldi'
elif re.match(r'^.*\.onnx$', input_model_path):
return 'onnx'
def get_ir_version():
"""
Default IR version.

82
tools/ovc/unit_tests/moc_tf_fe/check_info_messages_test.py
View File

@ -9,8 +9,7 @@ from contextlib import redirect_stdout
from unittest.mock import patch
from openvino.tools.ovc.main import main
from openvino.tools.ovc.get_ov_update_message import get_tf_fe_message, get_compression_message
from openvino.tools.ovc.get_ov_update_message import get_try_legacy_fe_message
from openvino.tools.ovc.get_ov_update_message import get_compression_message
def arg_parse_helper(input_model,
@ -18,41 +17,19 @@ def arg_parse_helper(input_model,
use_new_frontend,
input_model_is_text,
framework,
compress_to_fp16=False,
freeze_placeholder_with_value=None,
tensorflow_object_detection_api_pipeline_config=None):
compress_to_fp16=False):
path = os.path.dirname(__file__)
input_model = os.path.join(path, "test_models", input_model)
return argparse.Namespace(
input_model=input_model,
use_legacy_frontend=use_legacy_frontend,
use_new_frontend=use_new_frontend,
framework=framework,
input_model_is_text=input_model_is_text,
log_level='INFO',
silent=True,
model_name=None,
verbose=False,
output_model=None,
transform=[],
scale=None,
output=None,
input=None,
input_shape=None,
batch=None,
input_checkpoint=None,
saved_model_dir=None,
input_meta_graph=None,
saved_model_tags=None,
output_dir='.',
mean_values=(),
scale_values=(),
layout={},
source_layout={},
target_layout={},
freeze_placeholder_with_value=freeze_placeholder_with_value,
data_type=None,
tensorflow_custom_operations_config_update=None,
tensorflow_object_detection_api_pipeline_config=tensorflow_object_detection_api_pipeline_config,
compress_to_fp16=compress_to_fp16,
extensions=None
)
@ -63,60 +40,11 @@ class TestInfoMessagesTFFE(unittest.TestCase):
return_value=arg_parse_helper(input_model="model_int32.pbtxt",
use_legacy_frontend=False, use_new_frontend=True,
framework=None, input_model_is_text=True))
def test_api20_only(self, mock_argparse):
f = io.StringIO()
with redirect_stdout(f):
main()
std_out = f.getvalue()
tf_fe_message_found = get_tf_fe_message() in std_out
assert tf_fe_message_found
@patch('openvino.tools.ovc.convert_impl.driver', side_effect=Exception('MESSAGE'))
def run_fail_tf_fe(self, mock_driver):
from openvino.tools.ovc import convert_model
path = os.path.dirname(__file__)
convert_model(os.path.join(path, "test_models", "model_int32.pbtxt"), silent=False)
def test_suggest_legacy_fe(self):
f = io.StringIO()
with redirect_stdout(f):
try:
self.run_fail_tf_fe()
except:
pass
std_out = f.getvalue()
assert get_try_legacy_fe_message() in std_out
class TestInfoMessagesTFFEWithFallback(unittest.TestCase):
@patch('argparse.ArgumentParser.parse_args',
return_value=arg_parse_helper(input_model="model_switch_merge.pbtxt",
use_legacy_frontend=False, use_new_frontend=False,
framework=None, input_model_is_text=True,
freeze_placeholder_with_value="is_training->False"))
def test_tf_fe_message_fallback(self, mock_argparse):
f = io.StringIO()
with redirect_stdout(f):
main()
std_out = f.getvalue()
tf_fe_message_found = get_try_legacy_fe_message() in std_out
assert tf_fe_message_found, 'TF FE Info message is found for the fallback case'
@patch('argparse.ArgumentParser.parse_args',
return_value=arg_parse_helper(input_model="model_int32.pbtxt",
use_legacy_frontend=False, use_new_frontend=True,
compress_to_fp16=True,
framework=None, input_model_is_text=True,
tensorflow_object_detection_api_pipeline_config="config.yml"))
def test_tf_fe_message_fallback(self, mock_argparse):
f = io.StringIO()
with redirect_stdout(f):
main()
std_out = f.getvalue()
tf_fe_message_found = "The provided option \"tensorflow_object_detection_api_pipeline_config\" " \
"refers to legacy functionality. Please try to install openvino-dev and " \
"use convert_model() from openvino.tools.mo." in std_out
assert not tf_fe_message_found, 'TF FE Info message is found for the fallback case'
convert_model(os.path.join(path, "test_models", "model_int32.pbtxt"), verbose=True)
class TestInfoMessagesCompressFP16(unittest.TestCase):

78
tools/ovc/unit_tests/moc_tf_fe/conversion_basic_models_test.py
View File

@ -13,18 +13,12 @@ from openvino.tools.ovc.convert import convert_model
@generator
class TestMoFreezePlaceholderTFFE(unittest.TestCase):
def basic(self, input_model, argv_input, inputs, dtype, expected, freeze_placeholder_with_value=None,
input_shape=None, only_conversion=False, input_model_is_text=True, use_new_frontend=True,
use_legacy_frontend=False):
def basic(self, input_model, argv_input, inputs, dtype, expected, only_conversion=False):
path = os.path.dirname(__file__)
input_model = os.path.join(path, "test_models", input_model)
try:
model = convert_model(input_model, input=argv_input,
freeze_placeholder_with_value=freeze_placeholder_with_value,
input_shape=input_shape, input_model_is_text=input_model_is_text,
use_new_frontend=use_new_frontend, use_legacy_frontend=use_legacy_frontend,
framework="tf")
model = convert_model(input_model, input=argv_input)
except Exception as ex:
self.fail("Model conversion failed due to error: {}".format(ex))
@ -128,31 +122,27 @@ class TestMoFreezePlaceholderTFFE(unittest.TestCase):
None
),
(
None,
"in1,in2,cond->False",
{"in1": np.array([2.0, 4.0, 6.0], dtype=np.float32),
"in2": np.array([1.0, 3.0, 5.0], dtype=np.float32)},
np.array([2, 4, 6], dtype=np.float32),
np.float32,
"cond->False",
None,
True # fill a bug to investigate why compilation of this model is hang on
),
# case: input_shape + freeze_placeholder_with_value
(
None,
"in2,in1->[2.0 4.0 6.0],cond->True",
{"in2": np.array([1.0, 3.0, 5.0], dtype=np.float32)},
np.array([2, 4, 6], dtype=np.float32),
np.float32,
"in1->[2.0 4.0 6.0],cond->True",
"[3]",
False
),
],
)
def test_bool2(self, input_freezing_value, inputs, expected,
dtype=None, freeze_placeholder_with_value=None, input_shape=None, only_conversion=False):
self.basic("model_bool2.pbtxt", input_freezing_value, inputs, dtype, expected, freeze_placeholder_with_value,
input_shape, only_conversion)
dtype=None, only_conversion=False):
self.basic("model_bool2.pbtxt", input_freezing_value, inputs, dtype, expected,
only_conversion)
@generate(
*[
@ -173,10 +163,9 @@ class TestMoFreezePlaceholderTFFE(unittest.TestCase):
],
)
def test_cutting_fp32(self, input_freezing_value, inputs, expected,
dtype=None, freeze_placeholder_with_value=None, input_shape=None, only_conversion=False):
dtype=None, only_conversion=False):
self.basic("model_three_inputs.pbtxt", input_freezing_value, inputs, dtype, expected,
freeze_placeholder_with_value,
input_shape, only_conversion, True)
only_conversion)
@generate(
*[
@ -204,11 +193,9 @@ class TestMoFreezePlaceholderTFFE(unittest.TestCase):
],
)
def test_placeholder_with_default(self, inputs, inputs_data, expected,
dtype=None, freeze_placeholder_with_value=None, input_shape=None,
only_conversion=False):
dtype=None, only_conversion=False):
self.basic("placeholder_with_default.pbtxt", inputs, inputs_data, dtype, expected,
freeze_placeholder_with_value,
input_shape, only_conversion, True)
only_conversion)
@generate(
*[
@ -229,29 +216,12 @@ class TestMoFreezePlaceholderTFFE(unittest.TestCase):
],
)
def test_freeze_placeholder_with_unknown_rank(self, inputs, inputs_data, expected,
dtype=None, freeze_placeholder_with_value=None, input_shape=None,
only_conversion=False):
dtype=None, only_conversion=False):
self.basic("mul_with_unknown_rank_y.pbtxt", inputs, inputs_data, dtype, expected,
freeze_placeholder_with_value,
input_shape, only_conversion, True)
def test_conversion_failure_fallback_use_new_frontend(self):
with self.assertRaisesRegex(Exception,
"\[TensorFlow Frontend\] Internal error, no translator found for operation\(s\)\: "
"Enter\, Exit\, LoopCond\, Merge\, NextIteration\, Switch\, TensorArrayGatherV3\, "
"TensorArraySizeV3\, TensorArrayV3"):
self.basic("ctc_model_based.pbtxt", None, None, None, None,
None, None, True, True, True, False)
@unittest.skip("88349: Fix auto-pruning in legacy FE")
def test_conversion_model_oneshot_iterator_use_legacy_frontend(self):
self.basic("model_oneshot_iterator.pbtxt", None, None, None, None,
None, None, True, True, False, True)
only_conversion)
def test_conversion_model_oneshot_iterator_default(self):
self.basic("model_oneshot_iterator.pbtxt", None, None, None, None,
None, None, True, True, False, False)
self.basic("model_oneshot_iterator.pbtxt", None, None, None, None, True)
@generate(
*[
@ -272,9 +242,7 @@ class TestMoFreezePlaceholderTFFE(unittest.TestCase):
@unittest.skip("109220: Use generating script for this test model instead of Git LFS")
def test_conversion_model_with_non_standard_extension(self, input_freezing_value, inputs, expected,
dtype):
self.basic("model_fp32.frozen", input_freezing_value, inputs, dtype, expected, only_conversion=False,
input_model_is_text=False, use_new_frontend=True,
use_legacy_frontend=False)
self.basic("model_fp32.frozen", input_freezing_value, inputs, dtype, expected, only_conversion=False)
@unittest.skip("109220: Make TF FE to return the error")
def test_conversion_dir_model(self):
@ -282,8 +250,7 @@ class TestMoFreezePlaceholderTFFE(unittest.TestCase):
"Internal error or inconsistent input model: the frontend supports "
"only frozen binary protobuf format."):
self.basic(".", None, None, None, None,
only_conversion=True, input_model_is_text=False, use_new_frontend=True,
use_legacy_frontend=False)
only_conversion=True)
@generate(
*[
@ -300,8 +267,7 @@ class TestMoFreezePlaceholderTFFE(unittest.TestCase):
],
)
def test_conversion_pbtxt_model_with_inference(self, inputs, expected, dtype):
self.basic("model_with_if.pbtxt", None, inputs, dtype, expected, only_conversion=False,
input_model_is_text=False, use_new_frontend=True, use_legacy_frontend=False)
self.basic("model_with_if.pbtxt", None, inputs, dtype, expected, only_conversion=False)
@generate(
*[
@ -311,18 +277,16 @@ class TestMoFreezePlaceholderTFFE(unittest.TestCase):
"x[2,3]",
{"x": np.array([[12, 13, 10], [11, 14, 16]], dtype=np.float32)},
np.array([[12, 13, 10], [11, 14, 16]], dtype=np.float32),
np.float32, True, False,
np.float32
),
(
"model_mul_with_undefined_constant.pbtxt",
"x[2]",
{"x": np.array([11, -12], dtype=np.int32)},
np.array([0, 0], dtype=np.int32),
np.int32, True, False,
np.int32
),
],
)
def test_conversion_model_with_undefined_constant(self, model_name, argv_input, inputs, expected, dtype,
use_new_frontend, use_legacy_frontend):
self.basic(model_name, argv_input, inputs, dtype, expected, only_conversion=False,
input_model_is_text=True, use_new_frontend=use_new_frontend, use_legacy_frontend=use_legacy_frontend)
def test_conversion_model_with_undefined_constant(self, model_name, argv_input, inputs, expected, dtype):
self.basic(model_name, argv_input, inputs, dtype, expected, only_conversion=False)

30
tools/ovc/unit_tests/moc_tf_fe/conversion_incorrect_models_test.py
View File

@ -12,31 +12,9 @@ from openvino.tools.ovc.convert import convert_model
@generator
class TestMoFreezePlaceholderTFFE(unittest.TestCase):
@generate(
*[
# the default frontend
(
False, False, None
),
(
False, False, "tf"
),
# new frontend
(
True, False, None
),
(
True, False, "tf"
),
],
)
def test_conversion_fake_pb_model(self, use_new_frontend, use_legacy_frontend, framework):
with self.assertRaisesRegex(Exception,
"Internal error or inconsistent input model: the frontend supports frozen formats"
" \(.pb and .pbtxt\), SavedModel and MetaGraph \(.meta\), and v1 checkpoints."):
def test_conversion_fake_pb_model(self):
# TODO: Should FEs give detailed report why a model is rejected and should we print out the report?
with self.assertRaisesRegex(Exception, "Cannot recognize input model."):
path = os.path.dirname(__file__)
input_model = os.path.join(path, "test_models", "fake.pb")
convert_model(input_model,
use_new_frontend=use_new_frontend, use_legacy_frontend=use_legacy_frontend,
framework=framework)
convert_model(input_model)

5
tools/ovc/unit_tests/moc_tf_fe/conversion_with_checkpoint_v1_test.py
View File

@ -36,7 +36,6 @@ class TestBasicConversion(unittest.TestCase):
checkpoint_byte_stream = self.prepare_checkpoint_v1()
ckpt_file.write(bytes(checkpoint_byte_stream))
ckpt_file.close()
basic_check(input_model="model_with_variable_v1.pbtxt", argv_input=None,
basic_check(input_model=["model_with_variable_v1.pbtxt", ckpt_file.name], argv_input=None,
input_data={'input1': np.array([[1]], dtype=np.int64)},
expected_dtype=np.int64, expected_value=np.array([[14108583]], dtype=np.int64),
use_new_frontend=True, use_legacy_frontend=False, input_checkpoint=ckpt_file.name)
expected_dtype=np.int64, expected_value=np.array([[14108583]], dtype=np.int64))

83
tools/ovc/unit_tests/moc_tf_fe/conversion_with_layout_test.py
View File

@ -12,86 +12,3 @@ from openvino.runtime import Model
from openvino.runtime import PartialShape, Dimension
from openvino.tools.ovc.convert import convert_model
from openvino.tools.ovc.error import Error
@generator
class TestConversionWithBatchAndLayout(unittest.TestCase):
def basic_check(self, model_name: str, batch: int, layout: str, refs_shapes: dict):
path = os.path.dirname(__file__)
input_model = os.path.join(path, "test_models", model_name)
ov_model = convert_model(input_model, batch=batch, layout=layout)
for ov_input in ov_model.inputs:
input_name = ov_input.any_name
assert input_name in refs_shapes, "No reference input shape is found for {}".format(input_name)
input_shape = ov_input.get_partial_shape()
ref_shape = refs_shapes[input_name]
assert input_shape == ref_shape, "Incorrect shape for {} input:" \
" expected shape - {}, actual shape - {}".format(input_name, ref_shape,
input_shape)
@unittest.skip("Fix importing of openvino.test_utils in Jenkins")
def test_basic_model_no_layout(self):
from openvino.test_utils import compare_functions
path = os.path.dirname(__file__)
input_model = os.path.join(path, "test_models", "model_fp32.pbtxt")
ov_model = convert_model(input_model)
# compare with the reference graph
param1 = opset11.parameter([2, 2], name="in1", dtype=np.float32)
param2 = opset11.parameter([2, 2], name="in2", dtype=np.float32)
add = opset11.add(param1, param2, name="add")
ref_model = Model(add, [param1, param2])
flag, msg = compare_functions(ov_model, ref_model, compare_tensor_names=False)
assert flag, msg
@generate(
*[
(
"model_fp32.pbtxt", 5, "in1(cn),in2(cn)",
{"in1": PartialShape([2, 5]), "in2": PartialShape([2, 5])},
),
(
"model_fp32.pbtxt", 9, "in1(nc),in2(nc)",
{"in1": PartialShape([9, 2]), "in2": PartialShape([9, 2])},
),
(
"model_fp32.pbtxt", 7, "in1(?c),in2(?c)",
{"in1": PartialShape([2, 2]), "in2": PartialShape([2, 2])},
),
],
)
def test_basic_model_with_layout(self, model_name: str, batch: int, layout: str, refs_shapes: dict):
self.basic_check(model_name, batch, layout, refs_shapes)
@generate(
*[
(
"model_with_convolution_dynamic_rank.pbtxt", 7, "x(n???),kernel(????)",
{"x": PartialShape([7, Dimension.dynamic(), Dimension.dynamic(), Dimension.dynamic()]),
"kernel": PartialShape([2, 2, 3, 1])},
),
(
"model_with_convolution_dynamic_rank.pbtxt", 3, "x(???n),kernel(??n?)",
{"x": PartialShape([Dimension.dynamic(), Dimension.dynamic(), Dimension.dynamic(), 3]),
"kernel": PartialShape([2, 2, 3, 1])},
),
],
)
def test_model_with_convolution_dynamic_rank(self, model_name: str, batch: int, layout: str, refs_shapes: dict):
self.basic_check(model_name, batch, layout, refs_shapes)
@generate(
*[
(
"model_fp32.pbtxt", 17, "",
{},
),
],
)
def test_model_expected_failure(self, model_name: str, batch: int, layout: str, refs_shapes: dict):
# try to override batch size by default index (without specifying layout)
with self.assertRaisesRegex(Error,
"When you use \"batch\" option, Model Conversion API applies its value to the first "
"element of the shape if it is equal to -1, 0 or 1\."):
self.basic_check(model_name, batch, layout, refs_shapes)

14
tools/ovc/unit_tests/moc_tf_fe/utils.py
View File

@ -9,17 +9,15 @@ from openvino.runtime import Core
from openvino.tools.ovc.convert import convert_model
def basic_check(input_model, argv_input, input_data, expected_dtype, expected_value, freeze_placeholder_with_value=None,
input_shape=None, only_conversion=False, input_model_is_text=True, use_new_frontend=True,
use_legacy_frontend=False, extensions=None, input_checkpoint=None):
def basic_check(input_model, argv_input, input_data, expected_dtype, expected_value, \
only_conversion=False, input_model_is_text=True, use_new_frontend=True, extensions=None):
path = os.path.dirname(__file__)
if isinstance(input_model, (tuple, list)):
input_model = tuple(os.path.join(path, "test_models", part) for part in input_model)
else:
input_model = os.path.join(path, "test_models", input_model)
ov_model = convert_model(input_model, input=argv_input,
freeze_placeholder_with_value=freeze_placeholder_with_value,
input_shape=input_shape, input_model_is_text=input_model_is_text,
use_new_frontend=use_new_frontend, use_legacy_frontend=use_legacy_frontend,
framework="tf", extensions=extensions, input_checkpoint=input_checkpoint)
ov_model = convert_model(input_model, input=argv_input, extensions=extensions)
if only_conversion:
return ov_model

14
tools/ovc/unit_tests/ovc/convert/import_from_mo_test.py
View File

@ -6,10 +6,11 @@ import tempfile
from pathlib import Path
from generator import generator, generate
from openvino.runtime import serialize, InputCutInfo, LayoutMap
from openvino.runtime import serialize
from openvino.tools.ovc import InputCutInfo
from unit_tests.ovc.unit_test_with_mocked_telemetry import UnitTestWithMockedTelemetry
from utils import create_onnx_model, save_to_onnx
from unit_tests.ovc.convert.utils import create_onnx_model, save_to_onnx
@generator
@ -57,11 +58,10 @@ class ConvertImportMOTest(UnitTestWithMockedTelemetry):
@generate(*[
({}),
({'input': InputCutInfo(name='LeakyRelu_out', shape=None, type=None, value=None)}),
({'layout': {'input': LayoutMap(source_layout='NCHW', target_layout='NHWC')}}),
])
# Checks convert import from openvino.tools.mo
def test_import(self, params):
from openvino.runtime import convert_model
from openvino.tools.ovc import convert_model
with tempfile.TemporaryDirectory(dir=self.test_directory) as tmpdir:
model = create_onnx_model()
@ -73,7 +73,7 @@ class ConvertImportMOTest(UnitTestWithMockedTelemetry):
assert os.path.exists(out_xml)
def test_input_model_path(self):
from openvino.runtime import convert_model
from openvino.tools.ovc import convert_model
with tempfile.TemporaryDirectory(dir=self.test_directory) as tmpdir:
model = self.create_onnx_model()
@ -87,14 +87,14 @@ class ConvertImportMOTest(UnitTestWithMockedTelemetry):
def test_unnamed_input_model(self):
from openvino.runtime import convert_model
from openvino.tools.ovc import convert_model
with tempfile.TemporaryDirectory(dir=self.test_directory) as tmpdir:
model = self.create_onnx_model()
model_path = save_to_onnx(model, tmpdir)
out_xml = os.path.join(tmpdir, "model.xml")
ov_model = convert_model(model_path)
serialize(ov_model, out_xml.encode('utf-8'), out_xml.replace('.xml', '.bin').encode('utf-8'))
#serialize(ov_model, out_xml.encode('utf-8'), out_xml.replace('.xml', '.bin').encode('utf-8'))
#TODO: check that model is correct

4
tools/ovc/unit_tests/ovc/convert/meta_data_test.py
View File

@ -7,7 +7,8 @@ from pathlib import Path
from generator import generator
from openvino.runtime import get_version as get_rt_version
from openvino.runtime import serialize, convert_model
from openvino.runtime import serialize
from openvino.tools.ovc import convert_model
from unit_tests.ovc.unit_test_with_mocked_telemetry import UnitTestWithMockedTelemetry
from utils import save_to_onnx
@ -58,7 +59,6 @@ class MetaDataTest(UnitTestWithMockedTelemetry):
def ref_meta_data():
return {
'Runtime_version': get_rt_version(),
'legacy_frontend': "False",
'conversion_parameters': {
'input_model': Path.joinpath(Path("DIR"), Path("model.onnx")),
}

92
tools/ovc/unit_tests/ovc/utils/args_to_string_test.py
View File

@ -2,60 +2,13 @@
# SPDX-License-Identifier: Apache-2.0
import numpy as np
from openvino.runtime import Layout, PartialShape, Dimension, Shape, Type
from openvino.runtime import Layout, Dimension
from openvino.runtime import InputCutInfo, LayoutMap
from openvino.tools.ovc.cli_parser import mean_scale_value_to_str, \
transform_param_to_str, str_list_to_str, source_target_layout_to_str, layout_param_to_str
from openvino.tools.ovc.cli_parser import str_list_to_str
from unit_tests.ovc.unit_test_with_mocked_telemetry import UnitTestWithMockedTelemetry
class TestConvertingConvertArgumentsToString(UnitTestWithMockedTelemetry):
def test_mean_scale_value_to_str(self):
values = [0.5, 1.3, 0.67]
self.assertTrue(mean_scale_value_to_str(values) == "[0.5,1.3,0.67]")
values = {"input": [0.5, 1.3, 0.67]}
self.assertTrue(mean_scale_value_to_str(values) == "input[0.5,1.3,0.67]")
values = {"input1": [0.5, 1.3, 0.67], "input2": [4.2, 6.7, 3.15], "input3": [0.757, 4.6, 7.3]}
self.assertTrue(mean_scale_value_to_str(values) ==
"input1[0.5,1.3,0.67],input2[4.2,6.7,3.15],input3[0.757,4.6,7.3]")
self.assertRaises(Exception, mean_scale_value_to_str, **{"value": {("a", "b"): [0.5, 1.3, 0.67]}})
self.assertRaises(Exception, mean_scale_value_to_str, **{"value": {"name": Dimension(1)}})
self.assertRaises(Exception, mean_scale_value_to_str, **{"value": Dimension(1)})
def test_transform_param_to_str(self):
transform = 'MakeStateful'
self.assertTrue(transform_param_to_str(transform) == "MakeStateful")
transform1 = ('LowLatency2', {'use_const_initializer': False})
self.assertTrue(transform_param_to_str(transform1) ==
"LowLatency2[use_const_initializer=False]")
transform2 = ('MakeStateful', {'param_res_names': {
'input_name_1': 'output_name_1', 'input_name_2': 'output_name_2'}})
self.assertTrue(transform_param_to_str(transform2) ==
"MakeStateful[param_res_names={\'input_name_1\':\'output_name_1\',"
"\'input_name_2\':\'output_name_2\'}]")
transform = [transform1, transform2]
self.assertTrue(transform_param_to_str(transform) == "LowLatency2[use_const_initializer=False],"
"MakeStateful[param_res_names={"
"\'input_name_1\':\'output_name_1\',"
"\'input_name_2\':\'output_name_2\'}]")
self.assertRaises(Exception, transform_param_to_str, **{"value": ('LowLatency2',
{'use_const_initializer': False},
"param")})
self.assertRaises(Exception, transform_param_to_str, **{"value": (("a", "b"), {})})
self.assertRaises(Exception, transform_param_to_str, **{"value": ('LowLatency2', Dimension(1))})
self.assertRaises(Exception, transform_param_to_str, **{"value": ('LowLatency2',
{('a', 'b'): False})})
self.assertRaises(Exception, transform_param_to_str, **{"value": Dimension(1)})
def test_str_list_to_str(self):
list_str = ["data1", "data2", "data3"]
self.assertTrue(str_list_to_str(list_str) == "data1,data2,data3")
@ -65,44 +18,3 @@ class TestConvertingConvertArgumentsToString(UnitTestWithMockedTelemetry):
self.assertRaises(Exception, str_list_to_str, **{"values": [int, 1]})
self.assertRaises(Exception, str_list_to_str, **{"values": Dimension(1)})
def test_source_target_layout_to_str(self):
layout = {"input1": Layout("nhwc"), "input2": Layout("n??"), "input3": "nchw"}
self.assertTrue(source_target_layout_to_str(layout) == "input1([N,H,W,C]),input2([N,?,?]),input3(nchw)")
self.assertRaises(Exception, source_target_layout_to_str, **{"value": {"op": Dimension(1)}})
self.assertRaises(Exception, source_target_layout_to_str, **{"value": {("a", "b"): Layout("nhwc")}})
self.assertRaises(Exception, source_target_layout_to_str, **{"value": Dimension(1)})
def test_layout_param_to_str_to_str(self):
layout = {"input1": Layout("nhwc"), "input2": Layout("n??"), "input3": "nchw"}
self.assertTrue(layout_param_to_str(layout) == "input1([N,H,W,C]),input2([N,?,?]),input3(nchw)")
layout_map1 = LayoutMap(source_layout=Layout("n??"), target_layout=None)
layout_map2 = LayoutMap(source_layout=Layout("nhwc"), target_layout=("nchw"))
layout_map3 = LayoutMap(source_layout="abc", target_layout="cab")
layout = {"input1": layout_map1, "input2": layout_map2, "input3": layout_map3, "input4": Layout("nhwc"),
"input5": "n?"}
self.assertTrue(layout_param_to_str(layout) == "input1([N,?,?]),input2([N,H,W,C]->nchw),"
"input3(abc->cab),input4([N,H,W,C]),input5(n?)")
self.assertRaises(Exception, layout_param_to_str, **{"value": {"op": Dimension(1)}})
self.assertRaises(Exception, layout_param_to_str, **{"value": {("a", "b"): Layout("nhwc")}})
self.assertRaises(Exception, layout_param_to_str, **{"value": Dimension(1)})
layout = ["nhwc", "[n,c]"]
self.assertTrue(layout_param_to_str(layout) == "nhwc,[n,c]")
layout = ["abc->cab", "..nc"]
self.assertTrue(layout_param_to_str(layout) == "abc->cab,..nc")
layout_map1 = LayoutMap(source_layout=Layout("n??"), target_layout=None)
layout = [layout_map1, "..nc"]
self.assertTrue(layout_param_to_str(layout) == "[N,?,?],..nc")
layout_map2 = LayoutMap(source_layout=Layout("nhwc"), target_layout=("nchw"))
layout_map3 = LayoutMap(source_layout="abc", target_layout="cab")
layout = [layout_map2, layout_map3]
self.assertTrue(layout_param_to_str(layout) == "[N,H,W,C]->nchw,abc->cab")

1941
tools/ovc/unit_tests/ovc/utils/cli_parser_test.py
View File

File diff suppressed because it is too large Load Diff