1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
|
/*
* Copyright (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <tensorflow/contrib/lite/model.h>
#include <tensorflow/core/public/session.h>
#include "tflite/Assert.h"
#include "tflite/Session.h"
#include "tflite/InterpreterSession.h"
#include "tflite/NNAPISession.h"
#include "tflite/ext/kernels/register.h"
#include "misc/fp32.h"
#include <iostream>
#include <string>
#include <vector>
#define TF_ENSURE(e) { \
if(!(e).ok()) \
{ \
throw std::runtime_error{"'" #e "' FAILED"}; \
} \
}
using namespace tflite;
using namespace tflite::ops::builtin;
std::unique_ptr<FlatBufferModel> BuildModelFromFile(const std::string &path)
{
static StderrReporter reporter;
return FlatBufferModel::BuildFromFile(path.c_str(), &reporter);
}
std::unique_ptr<Interpreter> BuildInterpFromModel(const std::unique_ptr<FlatBufferModel> &model)
{
std::unique_ptr<Interpreter> interp;
BuiltinOpResolver resolver;
InterpreterBuilder builder(*model, resolver);
TFLITE_ENSURE(builder(&interp));
return std::move(interp);
}
tensorflow::TensorShape asTensorflowShape(const TfLiteTensor *tensor)
{
tensorflow::TensorShape shape;
const int rank = tensor->dims->size;
for (int axis = 0; axis < rank; ++axis)
{
shape.AddDim(tensor->dims->data[axis]);
}
return shape;
}
uint32_t count_elements(const TfLiteTensor *tensor)
{
const int rank = tensor->dims->size;
if (rank == 0)
{
return 0;
}
uint32_t res = 1;
for (int axis = 0; axis < rank; ++axis)
{
res *= tensor->dims->data[axis];
}
return res;
}
int main(int argc, char **argv)
{
bool use_nnapi = false;
if (std::getenv("USE_NNAPI") != nullptr)
{
use_nnapi = true;
}
if (argc < 3)
{
std::cerr << "USAGE: " << argv[0] << " [T/F lite model] [T/F model]" << std::endl;
return 255;
}
//
// Prepare Tensorflow Lite session
//
const std::string lite_model_path{argv[1]};
auto lite_model = BuildModelFromFile(lite_model_path);
auto lite_interp = BuildInterpFromModel(lite_model);
std::shared_ptr<nnfw::tflite::Session> lite_sess;
if (use_nnapi)
{
lite_sess = std::make_shared<nnfw::tflite::NNAPISession>(lite_interp.get());
}
else
{
lite_sess = std::make_shared<nnfw::tflite::InterpreterSession>(lite_interp.get());
}
//
// Prepare Tensorflow session
//
const std::string full_model_path{argv[2]};
tensorflow::Session* full_sess;
tensorflow::GraphDef full_model;
TF_ENSURE(tensorflow::NewSession(tensorflow::SessionOptions(), &full_sess));
TF_ENSURE(ReadBinaryProto(tensorflow::Env::Default(), full_model_path, &full_model));
TF_ENSURE(full_sess->Create(full_model));
//
//
//
std::vector<tensorflow::Tensor> input_nodes;
std::vector<std::string> input_names;
for (uint32_t n = 0; n < lite_interp->inputs().size(); ++n)
{
const TfLiteTensor *tensor = lite_interp->tensor(lite_interp->inputs().at(n));
input_nodes.emplace_back(tensorflow::DT_FLOAT, asTensorflowShape(tensor));
input_names.emplace_back(tensor->name);
}
assert(input_nodes.size() == input_names.size());
assert(input_nodes.size() == lite_interp->inputs().size());
std::vector<std::string> output_names;
std::vector<tensorflow::Tensor> output_nodes;
for (uint32_t n = 0; n < lite_interp->outputs().size(); ++n)
{
const TfLiteTensor *tensor = lite_interp->tensor(lite_interp->outputs().at(n));
output_names.emplace_back(tensor->name);
}
assert(output_names.size() == lite_interp->outputs().size());
// output_nodes will be initialized after Tensorflow Session run
assert(output_nodes.size() == 0);
//
// Prepare inference
//
lite_sess->prepare();
// TODO Feed Inputs (for both Tensorflow and Tensorflow Lite)
std::vector<std::pair<std::string, tensorflow::Tensor>> inputs;
for (uint32_t n = 0; n < input_nodes.size(); ++n)
{
inputs.emplace_back(input_names.at(0), input_nodes.at(0));
}
//
// Run inference
//
TF_ENSURE(full_sess->Run(inputs, output_names, {}, &output_nodes));
lite_sess->run();
//
// Compare Output
//
auto equals = [] (float lhs, float rhs) {
// TODO Allow users to set tolerance
if (nnfw::misc::fp32::absolute_epsilon_equal(lhs, rhs))
{
return true;
}
return nnfw::misc::fp32::epsilon_equal(lhs, rhs);
};
const uint32_t output_count = output_names.size();
bool matched = true;
for (uint32_t n = 0; n < output_count; ++n)
{
const TfLiteTensor *tensor = lite_interp->tensor(lite_interp->outputs().at(n));
// TODO Compare shape
const auto element_count = count_elements(tensor);
std::cout << "Compare output #" << n << "(" << tensor->name << ", " << element_count << " elements)" << std::endl;
for (uint32_t index = 0; index < element_count; ++index)
{
const auto full_value = output_nodes.at(n).flat<float>().data()[index];
const auto lite_value = lite_sess->interp()->typed_output_tensor<float>(n)[index];
if (!equals(full_value, lite_value))
{
std::cerr << full_value << " is expected, but " << lite_value << " is obtaeind (at " << n << ":" << index << ")" << std::endl;
matched = false;
}
}
}
//
// Cleanup
//
lite_sess->teardown();
return matched ? 0 : 255;
}
|
