2 Copyright (c) 2018 Intel Corporation
4 Licensed under the Apache License, Version 2.0 (the "License");
5 you may not use this file except in compliance with the License.
6 You may obtain a copy of the License at
8 http://www.apache.org/licenses/LICENSE-2.0
10 Unless required by applicable law or agreed to in writing, software
11 distributed under the License is distributed on an "AS IS" BASIS,
12 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 See the License for the specific language governing permissions and
14 limitations under the License.
21 from extensions.back.ShufflenetReLUReorder import ShufflenetReLUReorder
22 from mo.utils.unittest.graph import build_graph, compare_graphs
24 # The dictionary with nodes attributes used to build various graphs. A key is the name of the node and the value is the
25 # dictionary with node attributes.
27 'placeholder_1': {'shape': None, 'type': 'Placeholder', 'kind': 'op', 'op': 'Placeholder'},
28 'placeholder_1_data': {'value': None, 'shape': None, 'kind': 'data', 'data_type': None},
30 'relu_1': {'type': 'ReLU', 'kind': 'op', 'op': 'ReLU'},
31 'relu_1_data': {'value': None, 'shape': None, 'kind': 'data'},
33 'reshape_1': {'type': 'Reshape', 'kind': 'op', 'op': 'Reshape'},
34 'reshape_1_data': {'value': None, 'shape': None, 'kind': 'data'},
35 'reshape_2': {'type': 'Reshape', 'kind': 'op', 'op': 'Reshape'},
36 'reshape_2_data': {'value': None, 'shape': None, 'kind': 'data'},
37 'reshape_3': {'type': 'Reshape', 'kind': 'op', 'op': 'Reshape'},
38 'reshape_3_data': {'value': None, 'shape': None, 'kind': 'data'},
40 'transpose_1': {'type': 'Permute', 'kind': 'op', 'op': 'Transpose'},
41 'transpose_1_data': {'value': None, 'shape': None, 'kind': 'data'},
43 'conv_1': {'type': 'Convolution', 'kind': 'op', 'op': 'Conv2d'},
44 'conv_1_data': {'value': None, 'shape': None, 'kind': 'data'},
48 class ShufflenetReLUReorderTests(unittest.TestCase):
50 graph = build_graph(nodes_attributes,
51 [('placeholder_1', 'placeholder_1_data'),
52 ('placeholder_1_data', 'relu_1'),
53 ('relu_1', 'relu_1_data'),
54 ('relu_1_data', 'reshape_1'),
55 ('reshape_1', 'reshape_1_data'),
56 ('reshape_1_data', 'transpose_1'),
57 ('transpose_1', 'transpose_1_data'),
58 ('transpose_1_data', 'reshape_2'),
59 ('reshape_2', 'reshape_2_data'),
60 ('reshape_2_data', 'conv_1'),
61 ('conv_1', 'conv_1_data')
63 {'placeholder_1_data': {'shape': np.array([1, 227, 227, 112])},
64 'relu_1_data': {'shape': np.array([1, 227, 227, 112])},
65 'reshape_1_data': {'shape': np.array([227, 227, 4, 28])},
66 'transpose_1': {'order': np.array([0, 1, 3, 2])},
67 'transpose_1_data': {'shape': np.array([227, 227, 28, 4])},
68 'reshape_2_data': {'shape': np.array([1, 227, 227, 112])},
69 'conv_1_data': {'shape': np.array([1, 227, 227, 112])},
70 'conv_1': {'pad': np.array([1, 1])}
72 graph.graph['layout'] = 'NHWC'
74 graph_ref = build_graph(nodes_attributes,
75 [('placeholder_1', 'placeholder_1_data'),
76 ('placeholder_1_data', 'reshape_1'),
77 ('reshape_1', 'reshape_1_data'),
78 ('reshape_1_data', 'transpose_1'),
79 ('transpose_1', 'transpose_1_data'),
80 ('transpose_1_data', 'reshape_2'),
81 ('reshape_2', 'reshape_2_data'),
82 ('reshape_2_data', 'relu_1'),
83 ('relu_1', 'relu_1_data'),
84 ('relu_1_data', 'conv_1'),
85 ('conv_1', 'conv_1_data')
87 {'placeholder_1_data': {'shape': np.array([1, 227, 227, 112])},
88 'relu_1_data': {'shape': np.array([1, 227, 227, 112])},
89 'reshape_1_data': {'shape': np.array([227, 227, 4, 28])},
90 'transpose_1': {'order': np.array([0, 1, 3, 2])},
91 'transpose_1_data': {'shape': np.array([227, 227, 28, 4])},
92 'reshape_2_data': {'shape': np.array([1, 227, 227, 112])},
93 'conv_1_data': {'shape': np.array([1, 227, 227, 112])},
96 pattern = ShufflenetReLUReorder()
97 pattern.find_and_replace_pattern(graph)
99 (flag, resp) = compare_graphs(graph, graph_ref, 'conv_1_data', check_op_attrs=True)
100 self.assertTrue(flag, resp)
102 def test_2_neg(self):
103 graph = build_graph(nodes_attributes,
104 [('placeholder_1', 'placeholder_1_data'),
105 ('placeholder_1_data', 'reshape_1'),
106 ('reshape_1', 'reshape_1_data'),
107 ('reshape_1_data', 'transpose_1'),
108 ('transpose_1', 'transpose_1_data'),
109 ('transpose_1_data', 'reshape_2'),
110 ('reshape_2', 'reshape_2_data'),
111 ('reshape_2_data', 'conv_1'),
112 ('conv_1', 'conv_1_data')
114 {'placeholder_1_data': {'shape': np.array([1, 227, 227, 112])},
115 'relu_1_data': {'shape': np.array([1, 227, 227, 112])},
116 'reshape_1_data': {'shape': np.array([227, 227, 4, 28])},
117 'transpose_1': {'order': np.array([0, 1, 3, 2])},
118 'transpose_1_data': {'shape': np.array([227, 227, 28, 4])},
119 'reshape_2_data': {'shape': np.array([1, 227, 227, 112])},
120 'conv_1_data': {'shape': np.array([1, 227, 227, 112])},
122 graph.graph['layout'] = 'NHWC'
124 graph_ref = build_graph(nodes_attributes,
125 [('placeholder_1', 'placeholder_1_data'),
126 ('placeholder_1_data', 'reshape_1'),
127 ('reshape_1', 'reshape_1_data'),
128 ('reshape_1_data', 'transpose_1'),
129 ('transpose_1', 'transpose_1_data'),
130 ('transpose_1_data', 'reshape_2'),
131 ('reshape_2', 'reshape_2_data'),
132 ('reshape_2_data', 'conv_1'),
133 ('conv_1', 'conv_1_data')
135 {'placeholder_1_data': {'shape': np.array([1, 227, 227, 112])},
136 'relu_1_data': {'shape': np.array([1, 227, 227, 112])},
137 'reshape_1_data': {'shape': np.array([227, 227, 4, 28])},
138 'transpose_1': {'order': np.array([0, 1, 3, 2])},
139 'transpose_1_data': {'shape': np.array([227, 227, 28, 4])},
140 'reshape_2_data': {'shape': np.array([1, 227, 227, 112])},
141 'conv_1_data': {'shape': np.array([1, 227, 227, 112])},
144 pattern = ShufflenetReLUReorder()
145 pattern.find_and_replace_pattern(graph)
147 (flag, resp) = compare_graphs(graph, graph_ref, 'conv_1_data', check_op_attrs=True)
148 self.assertTrue(flag, resp)