• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
2#
3# Licensed under the Apache License, Version 2.0 (the "License");
4# you may not use this file except in compliance with the License.
5# You may obtain a copy of the License at
6#
7#     http://www.apache.org/licenses/LICENSE-2.0
8#
9# Unless required by applicable law or agreed to in writing, software
10# distributed under the License is distributed on an "AS IS" BASIS,
11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12# See the License for the specific language governing permissions and
13# limitations under the License.
14# ==============================================================================
15"""Test configs for depthwiseconv."""
16from __future__ import absolute_import
17from __future__ import division
18from __future__ import print_function
19
20import numpy as np
21import tensorflow.compat.v1 as tf
22from tensorflow.lite.testing.zip_test_utils import create_tensor_data
23from tensorflow.lite.testing.zip_test_utils import make_zip_of_tests
24from tensorflow.lite.testing.zip_test_utils import register_make_test_function
25
26
27@register_make_test_function()
28def make_depthwiseconv_tests(options):
29  """Make a set of tests to do convolution."""
30
31  # Tensorflow only supports equal strides
32  test_parameters = [
33      {
34          "input_shape": [[1, 3, 4, 3], [1, 10, 10, 3]],
35          "filter_size": [[1, 1], [1, 2], [3, 3]],
36          "strides": [[1, 1, 1, 1], [1, 3, 3, 1]],
37          "dilations": [[1, 1, 1, 1], [1, 3, 2, 1], [1, 2, 2, 1]],
38          "channel_multiplier": [1, 2],
39          "rate": [[1, 1]],
40          "padding": ["SAME", "VALID"],
41          "data_format": ["NHWC"],
42          "constant_filter": [True, False],
43          "fully_quantize": [False],
44          "quant_16x8": [False]
45      },
46      {
47          "input_shape": [[1, 3, 4, 3]],
48          "filter_size": [[1, 1]],
49          "strides": [[1, 1, 2, 1]],  # TF needs [1, x, x, 1]
50          "dilations": [[1, 1, 1, 1], [1, 2, 2, 1]],
51          "channel_multiplier": [2],
52          "rate": [[2, 2]],  #  Only [1, 1] is supported
53          "padding": ["SAME"],
54          "data_format": ["NHWC"],
55          "constant_filter": [True, False],
56          "fully_quantize": [False],
57          "quant_16x8": [False]
58      },
59      {
60          "input_shape": [[1, 3, 4, 3], [1, 10, 10, 3]],
61          "filter_size": [[1, 1], [1, 2], [3, 3]],
62          "strides": [[1, 1, 1, 1], [1, 3, 3, 1]],
63          "dilations": [[1, 1, 1, 1], [1, 3, 2, 1], [1, 2, 2, 1]],
64          "channel_multiplier": [1, 2],
65          "rate": [[1, 1]],
66          "padding": ["SAME", "VALID"],
67          "data_format": ["NHWC"],
68          "constant_filter": [True],
69          "fully_quantize": [True],
70          "quant_16x8": [False]
71      },
72      {
73          "input_shape": [[1, 3, 4, 3]],
74          "filter_size": [[1, 2]],
75          "strides": [[1, 3, 3, 1]],
76          "dilations": [[1, 3, 2, 1]],
77          "channel_multiplier": [1],
78          "rate": [[1, 1]],
79          "padding": ["SAME", "VALID"],
80          "data_format": ["NHWC"],
81          "constant_filter": [True],
82          "fully_quantize": [True],
83          "quant_16x8": [True]
84      },
85  ]
86
87  def get_tensor_shapes(parameters):
88    input_shape = parameters["input_shape"]
89    filter_size = parameters["filter_size"]
90    filter_shape = filter_size + [
91        input_shape[3], parameters["channel_multiplier"]
92    ]
93    return [input_shape, filter_shape]
94
95  def build_graph(parameters):
96    """Build a depthwise conv graph given `parameters`."""
97    input_shape, filter_shape = get_tensor_shapes(parameters)
98    input_tensor = tf.compat.v1.placeholder(
99        dtype=tf.float32, name="input", shape=input_shape)
100
101    # Get filter input either as a placeholder or constants. Also get a list of
102    # the input tensors that are represented as placeholders.
103    if parameters["constant_filter"]:
104      filter_input = create_tensor_data(np.float32, filter_shape)
105      input_tensors = [input_tensor]
106    else:
107      filter_input = tf.compat.v1.placeholder(
108          dtype=tf.float32, name="filter", shape=filter_shape)
109      input_tensors = [input_tensor, filter_input]
110
111    out = tf.nn.depthwise_conv2d(
112        input_tensor,
113        filter_input,
114        strides=parameters["strides"],
115        rate=parameters["rate"],
116        padding=parameters["padding"],
117        data_format=parameters["data_format"])
118    return input_tensors, [out]
119
120  def build_inputs(parameters, sess, inputs, outputs):
121    # pylint: disable=g-doc-return-or-yield, g-doc-args
122    """Build list of input values.
123
124    It either contains 1 tensor (input) or 2 tensors (input, filter) based on
125    whether filter is constant or variable input.
126    """
127
128    input_shape, filter_shape = get_tensor_shapes(parameters)
129    values = [
130        create_tensor_data(np.float32, input_shape, min_value=-1, max_value=1)
131    ]
132    if not parameters["constant_filter"]:
133      values.append(
134          create_tensor_data(
135              np.float32, filter_shape, min_value=-1, max_value=1))
136    return values, sess.run(outputs, feed_dict=dict(zip(inputs, values)))
137
138  make_zip_of_tests(
139      options,
140      test_parameters,
141      build_graph,
142      build_inputs,
143      expected_tf_failures=4)
144