{{ message }}
This repository was archived by the owner on Nov 17, 2023. It is now read-only.
Repository navigation
Expand file tree
/
Copy pathresnet.cpp
More file actions
283 lines (245 loc) · 9.08 KB
/
Copy pathresnet.cpp
File metadata and controls
283 lines (245 loc) · 9.08 KB
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
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you 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 <map>
#include <string>
#include <fstream>
#include <vector>
#include <cstdlib>
#include "utils.h"
#include "mxnet-cpp/MxNetCpp.h"
using namespace mxnet::cpp;
Symbol ConvolutionNoBias(const std::string& symbol_name,
Symbol data,
Symbol weight,
Shape kernel,
int num_filter,
Shape stride = Shape(1, 1),
Shape dilate = Shape(1, 1),
Shape pad = Shape(0, 0),
int num_group = 1,
int64_t workspace = 512) {
return Operator("Convolution")
.SetParam("kernel", kernel)
.SetParam("num_filter", num_filter)
.SetParam("stride", stride)
.SetParam("dilate", dilate)
.SetParam("pad", pad)
.SetParam("num_group", num_group)
.SetParam("workspace", workspace)
.SetParam("no_bias", true)
.SetInput("data", data)
.SetInput("weight", weight)
.CreateSymbol(symbol_name);
}
Symbol getConv(const std::string & name, Symbol data,
int num_filter,
Shape kernel, Shape stride, Shape pad,
bool with_relu,
mx_float bn_momentum) {
Symbol conv_w(name + "_w");
Symbol conv = ConvolutionNoBias(name, data, conv_w,
kernel, num_filter, stride, Shape(1, 1),
pad, 1, 512);
Symbol gamma(name + "_gamma");
Symbol beta(name + "_beta");
Symbol mmean(name + "_mmean");
Symbol mvar(name + "_mvar");
Symbol bn = BatchNorm(name + "_bn", conv, gamma,
beta, mmean, mvar, 2e-5, bn_momentum, false);
if (with_relu) {
return Activation(name + "_relu", bn, "relu");
} else {
return bn;
}
}
Symbol makeBlock(const std::string & name, Symbol data, int num_filter,
bool dim_match, mx_float bn_momentum) {
Shape stride;
if (dim_match) {
stride = Shape(1, 1);
} else {
stride = Shape(2, 2);
}
Symbol conv1 = getConv(name + "_conv1", data, num_filter,
Shape(3, 3), stride, Shape(1, 1),
true, bn_momentum);
Symbol conv2 = getConv(name + "_conv2", conv1, num_filter,
Shape(3, 3), Shape(1, 1), Shape(1, 1),
false, bn_momentum);
Symbol shortcut;
if (dim_match) {
shortcut = data;
} else {
Symbol shortcut_w(name + "_proj_w");
shortcut = ConvolutionNoBias(name + "_proj", data, shortcut_w,
Shape(2, 2), num_filter,
Shape(2, 2), Shape(1, 1), Shape(0, 0),
1, 512);
}
Symbol fused = shortcut + conv2;
return Activation(name + "_relu", fused, "relu");
}
Symbol getBody(Symbol data, int num_level, int num_block, int num_filter, mx_float bn_momentum) {
for (int level = 0; level < num_level; level++) {
for (int block = 0; block < num_block; block++) {
data = makeBlock("level" + std::to_string(level + 1) + "_block" + std::to_string(block + 1),
data, num_filter * (std::pow(2, level)),
(level == 0 || block > 0), bn_momentum);
}
}
return data;
}
Symbol ResNetSymbol(int num_class, int num_level = 3, int num_block = 9,
int num_filter = 16, mx_float bn_momentum = 0.9,
mxnet::cpp::Shape pool_kernel = mxnet::cpp::Shape(8, 8)) {
// data and label
Symbol data = Symbol::Variable("data");
Symbol data_label = Symbol::Variable("data_label");
Symbol gamma("gamma");
Symbol beta("beta");
Symbol mmean("mmean");
Symbol mvar("mvar");
Symbol zscore = BatchNorm("zscore", data, gamma,
beta, mmean, mvar, 0.001, bn_momentum);
Symbol conv = getConv("conv0", zscore, num_filter,
Shape(3, 3), Shape(1, 1), Shape(1, 1),
true, bn_momentum);
Symbol body = getBody(conv, num_level, num_block, num_filter, bn_momentum);
Symbol pool = Pooling("pool", body, pool_kernel, PoolingPoolType::kAvg);
Symbol flat = Flatten("flatten", pool);
Symbol fc_w("fc_w"), fc_b("fc_b");
Symbol fc = FullyConnected("fc", flat, fc_w, fc_b, num_class);
return SoftmaxOutput("softmax", fc, data_label);
}
NDArray ResizeInput(NDArray data, const Shape new_shape) {
NDArray pic = data.Reshape(Shape(0, 1, 28, 28));
NDArray pic_1channel;
Operator("_contrib_BilinearResize2D")
.SetParam("height", new_shape[2])
.SetParam("width", new_shape[3])
(pic).Invoke(pic_1channel);
NDArray output;
Operator("tile")
.SetParam("reps", Shape(1, 3, 1, 1))
(pic_1channel).Invoke(output);
return output;
}
int main(int argc, char const *argv[]) {
int max_epoch = argc > 1 ? strtol(argv[1], nullptr, 10) : 100;
float learning_rate = 1e-4;
float weight_decay = 1e-4;
TRY
auto resnet = ResNetSymbol(10);
std::map<std::string, NDArray> args_map;
std::map<std::string, NDArray> aux_map;
/*context*/
auto ctx = Context::cpu();
int num_gpu;
MXGetGPUCount(&num_gpu);
int batch_size = 8;
#if MXNET_USE_CUDA
if (num_gpu > 0) {
ctx = Context::gpu();
batch_size = 32;
}
#endif
const Shape data_shape = Shape(batch_size, 3, 224, 224),
label_shape = Shape(batch_size);
args_map["data"] = NDArray(data_shape, ctx);
args_map["data_label"] = NDArray(label_shape, ctx);
resnet.InferArgsMap(ctx, &args_map, args_map);
std::vector<std::string> data_files = { "./data/mnist_data/train-images-idx3-ubyte",
"./data/mnist_data/train-labels-idx1-ubyte",
"./data/mnist_data/t10k-images-idx3-ubyte",
"./data/mnist_data/t10k-labels-idx1-ubyte"
};
auto train_iter = MXDataIter("MNISTIter");
if (!setDataIter(&train_iter, "Train", data_files, batch_size)) {
return 1;
}
auto val_iter = MXDataIter("MNISTIter");
if (!setDataIter(&val_iter, "Label", data_files, batch_size)) {
return 1;
}
// initialize parameters
auto initializer = Uniform(0.07);
for (auto& arg : args_map) {
initializer(arg.first, &arg.second);
}
Optimizer* opt = OptimizerRegistry::Find("sgd");
opt->SetParam("lr", learning_rate)
->SetParam("wd", weight_decay)
->SetParam("momentum", 0.9)
->SetParam("rescale_grad", 1.0 / batch_size)
->SetParam("clip_gradient", 10);
auto *exec = resnet.SimpleBind(ctx, args_map);
auto arg_names = resnet.ListArguments();
// Create metrics
Accuracy train_acc, val_acc;
LogLoss logloss_train, logloss_val;
for (int epoch = 0; epoch < max_epoch; ++epoch) {
LG << "Epoch: " << epoch;
train_iter.Reset();
train_acc.Reset();
int iter = 0;
while (train_iter.Next()) {
auto data_batch = train_iter.GetDataBatch();
ResizeInput(data_batch.data, data_shape).CopyTo(&args_map["data"]);
data_batch.label.CopyTo(&args_map["data_label"]);
NDArray::WaitAll();
exec->Forward(true);
exec->Backward();
for (size_t i = 0; i < arg_names.size(); ++i) {
if (arg_names[i] == "data" || arg_names[i] == "data_label") continue;
opt->Update(i, exec->arg_arrays[i], exec->grad_arrays[i]);
}
NDArray::WaitAll();
train_acc.Update(data_batch.label, exec->outputs[0]);
logloss_train.Reset();
logloss_train.Update(data_batch.label, exec->outputs[0]);
++iter;
LG << "EPOCH: " << epoch << " ITER: " << iter
<< " Train Accuracy: " << train_acc.Get()
<< " Train Loss: " << logloss_train.Get();
}
LG << "EPOCH: " << epoch << " Train Accuracy: " << train_acc.Get();
val_iter.Reset();
val_acc.Reset();
iter = 0;
while (val_iter.Next()) {
auto data_batch = val_iter.GetDataBatch();
ResizeInput(data_batch.data, data_shape).CopyTo(&args_map["data"]);
data_batch.label.CopyTo(&args_map["data_label"]);
NDArray::WaitAll();
exec->Forward(false);
NDArray::WaitAll();
val_acc.Update(data_batch.label, exec->outputs[0]);
LG << "EPOCH: " << epoch << " ITER: " << iter << " Val Accuracy: " << val_acc.Get();
++iter;
}
LG << "Validation Accuracy: " << val_acc.Get();
}
delete exec;
delete opt;
MXNotifyShutdown();
CATCH
return 0;
}
You can’t perform that action at this time.