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 conv with activations.""" 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 27def make_conv_activation_tests(activation_op): 28 """Make a set of tests to do convolution with activation.""" 29 30 def f(options): 31 """Actual function that generates examples.""" 32 test_parameters = [ 33 { 34 "input_shape": [[1, 3, 4, 3], [4, 6, 6, 1]], 35 "filter_shape": [[1, 1], [2, 3], [3, 3]], 36 "strides": [[1, 1, 1, 1], [1, 2, 3, 1]], 37 "dilations": [[1, 1, 1, 1], [1, 3, 2, 1], [1, 2, 2, 1]], 38 "padding": ["SAME", "VALID"], 39 "data_format": ["NHWC"], # TODO(aselle): NCHW would be good 40 "constant_filter": [True, False], 41 "channel_multiplier": [1, 2], 42 "fully_quantize": [False], 43 "quant_16x8": [False], 44 "dynamic_range_quantize": [False], 45 }, 46 # TODO(b/134702301): The fully_quantize param is just ignored by the 47 # MLIR testing path now, resulting in duplicate tests. Either ignore 48 # these tests or handle it properly in the mlir_convert() function. 49 { 50 "input_shape": [[1, 3, 4, 3], [4, 6, 6, 1]], 51 "filter_shape": [[1, 1], [2, 3]], 52 "strides": [[1, 1, 1, 1], [1, 2, 3, 1]], 53 "dilations": [[1, 1, 1, 1], [1, 3, 2, 1]], 54 "padding": ["SAME", "VALID"], 55 "data_format": ["NHWC"], # TODO(aselle): NCHW would be good 56 "constant_filter": [True], 57 "channel_multiplier": [1, 2], 58 "fully_quantize": [True], 59 "quant_16x8": [False, True], 60 "dynamic_range_quantize": [False], 61 }, 62 { 63 "input_shape": [[1, 3, 4, 3]], 64 "filter_shape": [[1, 1], [2, 3], [3, 3]], 65 "strides": [[1, 1, 1, 1], [1, 2, 3, 1]], 66 "dilations": [[1, 1, 1, 1]], 67 "padding": ["SAME", "VALID"], 68 "data_format": ["NHWC"], 69 "constant_filter": [True], 70 "channel_multiplier": [1, 2], 71 "fully_quantize": [False], 72 "quant_16x8": [False], 73 "dynamic_range_quantize": [True], 74 }, 75 ] 76 77 def get_tensor_shapes(parameters): 78 input_shape = parameters["input_shape"] 79 filter_size = parameters["filter_shape"] 80 filter_shape = filter_size + [ 81 input_shape[3], parameters["channel_multiplier"] 82 ] 83 return [input_shape, filter_shape] 84 85 def build_graph(parameters): 86 """Build a conv graph given `parameters`.""" 87 input_shape, filter_shape = get_tensor_shapes(parameters) 88 input_tensor = tf.compat.v1.placeholder( 89 dtype=tf.float32, name="input", shape=input_shape) 90 91 # Get filter input either as a placeholder or constants. Also get a list 92 # of the input tensors that are represented as placeholders. 93 if parameters["constant_filter"]: 94 filter_input = create_tensor_data( 95 np.float32, filter_shape, min_value=-10, max_value=10) 96 input_tensors = [input_tensor] 97 else: 98 filter_input = tf.compat.v1.placeholder( 99 dtype=tf.float32, name="filter", shape=filter_shape) 100 input_tensors = [input_tensor, filter_input] 101 102 out = tf.nn.conv2d( 103 input_tensor, 104 filter_input, 105 strides=parameters["strides"], 106 dilations=parameters["dilations"], 107 padding=parameters["padding"], 108 data_format=parameters["data_format"]) 109 out = activation_op(out) 110 return input_tensors, [out] 111 112 def build_inputs(parameters, sess, inputs, outputs): 113 """Build inputs for conv with activation.""" 114 115 input_shape, filter_shape = get_tensor_shapes(parameters) 116 values = [ 117 create_tensor_data( 118 np.float32, input_shape, min_value=-1, max_value=1) 119 ] 120 if not parameters["constant_filter"]: 121 values.append(create_tensor_data(np.float32, filter_shape)) 122 return values, sess.run(outputs, feed_dict=dict(zip(inputs, values))) 123 124 make_zip_of_tests( 125 options, 126 test_parameters, 127 build_graph, 128 build_inputs, 129 expected_tf_failures=48) 130 131 return f 132 133 134@register_make_test_function() 135def make_conv_relu6_tests(options): 136 """Make a set of tests to do conv_relu6.""" 137 return make_conv_activation_tests(tf.nn.relu6)(options) 138 139 140@register_make_test_function() 141def make_conv_relu_tests(options): 142 """Make a set of tests to do conv_relu.""" 143 return make_conv_activation_tests(tf.nn.relu)(options) 144 145 146def relu1(input_tensor): 147 # Note that the following is not supported: 148 # out = tf.maximum(-1.0, tf.minimum(input_tensor, 1.0)) 149 out = tf.minimum(1.0, tf.maximum(input_tensor, -1.0)) 150 return out 151 152 153@register_make_test_function() 154def make_conv_relu1_tests(options): 155 """Make a set of tests to do conv_relu1.""" 156 return make_conv_activation_tests(relu1)(options) 157