如何调用训练好的caffemodel

本文转载自查看原文 2016-04-08 17:13 8989

1.首先要准备几样东西：

（1）要预测的图像，需要32×32大小;

（2）网络配置文件，prototxt,以及每个图像的路径及其序号。

（3）训练好的caffemodel以及均值二进制文件，貌似可以定值，需要通过数据训练计算得到。

（3）预测的主程序

内容：

 1 #include<cstring>
 2 #include<cstdlib>
 3 #include<vector>
 4 #include<string>
 5 #include<iostream>
 6 #include<stdio.h>
 7 #include<assert.h>
 8 #include"caffe/caffe.hpp"
 9 #include"caffe/util/io.hpp"
10 #include"caffe/blob.hpp"
11 usingnamespace caffe;
12 usingnamespace std;
13 int main(int argc,char** argv)
14 {
15 LOG(INFO)<< argv[0]<<" [GPU] [Device ID]";
16    //Setting CPU or GPU
17 if(argc >=2&& strcmp(argv[1],"GPU")==0)
18 {
19 Caffe::set_mode(Caffe::GPU);
20 int device_id =0;
21 if(argc ==3)
22        {
23          device_id = atoi(argv[2]);
24        }
25 Caffe::SetDevice(device_id);
26   LOG(INFO)<<"Using GPU #"<< device_id;
27 }
28 else
29 {
30        LOG(INFO)<<"Using CPU";
31        Caffe::set_mode(Caffe::CPU);
32 }
33 // Set to TEST Phase
34 //Caffe::set_phase(Caffe::TEST);
35 // Load net
36 // Assume you are in Caffe master directory，神经网络的配置文件
37 Net<float> net("./examples/prediction_example/prediction_example.prototxt",TEST);
38 // Load pre-trained net (binary proto)
39 // Assume you are already trained the cifar10 example.预训练网络加载
40 net.CopyTrainedLayersFrom("./examples/cifar10/cifar10_quick_iter_5000.caffemodel");
41 float loss =0.0;
42 vector<Blob<float>*> results = net.ForwardPrefilled(&loss);
43 LOG(INFO)<<"Result size: "<< results.size();
44 // Log how many blobs were loaded
45 LOG(INFO)<<"Blob size: "<< net.input_blobs().size();
46 LOG(INFO)<<"-------------";
47 LOG(INFO)<<" prediction :  ";
48 // Get probabilities
49 const boost::shared_ptr<Blob<float>>& probLayer = net.blob_by_name("prob");
50 constfloat* probs_out = probLayer->cpu_data();
51 // Get argmax results
52 const boost::shared_ptr<Blob<float>>& argmaxLayer = net.blob_by_name("argmax");
53 // Display results
54 LOG(INFO)<<"---------------------------------------------------------------";
55 constfloat* argmaxs = argmaxLayer->cpu_data();
56 for(int i =0; i < argmaxLayer->num(); i++)
57 {
58 LOG(INFO)<<"Pattern:"<< i <<" class:"<< argmaxs[i*argmaxLayer->height()+0]<<" Prob="<< probs_out[i*probLayer->height()+0];
59 }
60 LOG(INFO)<<"-------------";
61 return0;
62 }

View Code

2.结果：

  1 anmeng@cam:/opt/caffe-master$  ./build/tools/prediction_example
  2 WARNING:Logging before InitGoogleLogging() is written to STDERR
  3 I0326 00:32:58.843086  3090 prediction_example.cpp:19]./build/tools/prediction_example [GPU][Device ID]
  4 I0326 00:32:58.843185  3090 prediction_example.cpp:35]Using CPU
  5 I0326 00:32:58.990399  3090 upgrade_proto.cpp:52]Attempting to upgrade input file specified using deprecated V1LayerParameter:./examples/prediction_example/prediction_example.prototxt
  6 I0326 00:32:58.990525  3090 upgrade_proto.cpp:60]Successfully upgraded file specified using deprecated V1LayerParameter
  7 I0326 00:32:58.990702  3090 net.cpp:49]Initializing net from parameters:
  8 name:"CIFAR10_quick"
  9 state {
 10  phase: TEST
 11 }
 12 layer {
 13  name:"data"
 14  type:"ImageData"
 15  top:"data"
 16  top:"label"
 17  transform_param {
 18    mean_file:"./examples/cifar10/mean.binaryproto"
 19  }
 20  image_data_param {
 21    source:"./examples/prediction_example/file_list.txt"
 22    batch_size:7
 23    new_height:32
 24    new_width:32
 25  }
 26 }
 27 layer {
 28  name:"conv1"
 29  type:"Convolution"
 30  bottom:"data"
 31  top:"conv1"
 32  param {
 33    lr_mult:1
 34  }
 35  param {
 36    lr_mult:2
 37  }
 38  convolution_param {
 39    num_output:32
 40    pad:2
 41    kernel_size:5
 42    stride:1
 43    weight_filler {
 44      type:"gaussian"
 45      std:0.0001
 46    }
 47    bias_filler {
 48      type:"constant"
 49    }
 50  }
 51 }
 52 layer {
 53  name:"pool1"
 54  type:"Pooling"
 55  bottom:"conv1"
 56  top:"pool1"
 57  pooling_param {
 58    pool: MAX
 59    kernel_size:3
 60    stride:2
 61  }
 62 }
 63 layer {
 64  name:"relu1"
 65  type:"ReLU"
 66  bottom:"pool1"
 67  top:"pool1"
 68 }
 69 layer {
 70  name:"conv2"
 71  type:"Convolution"
 72  bottom:"pool1"
 73  top:"conv2"
 74  param {
 75    lr_mult:1
 76  }
 77  param {
 78    lr_mult:2
 79  }
 80  convolution_param {
 81    num_output:32
 82    pad:2
 83    kernel_size:5
 84    stride:1
 85    weight_filler {
 86      type:"gaussian"
 87      std:0.01
 88    }
 89    bias_filler {
 90      type:"constant"
 91    }
 92  }
 93 }
 94 layer {
 95  name:"relu2"
 96  type:"ReLU"
 97  bottom:"conv2"
 98  top:"conv2"
 99 }
100 layer {
101  name:"pool2"
102  type:"Pooling"
103  bottom:"conv2"
104  top:"pool2"
105  pooling_param {
106    pool: AVE
107    kernel_size:3
108    stride:2
109  }
110 }
111 layer {
112  name:"conv3"
113  type:"Convolution"
114  bottom:"pool2"
115  top:"conv3"
116  param {
117    lr_mult:1
118  }
119  param {
120    lr_mult:2
121  }
122  convolution_param {
123    num_output:64
124    pad:2
125    kernel_size:5
126    stride:1
127    weight_filler {
128      type:"gaussian"
129      std:0.01
130    }
131    bias_filler {
132      type:"constant"
133    }
134  }
135 }
136 layer {
137  name:"relu3"
138  type:"ReLU"
139  bottom:"conv3"
140  top:"conv3"
141 }
142 layer {
143  name:"pool3"
144  type:"Pooling"
145  bottom:"conv3"
146  top:"pool3"
147  pooling_param {
148    pool: AVE
149    kernel_size:3
150    stride:2
151  }
152 }
153 layer {
154  name:"ip1"
155  type:"InnerProduct"
156  bottom:"pool3"
157  top:"ip1"
158  param {
159    lr_mult:1
160  }
161  param {
162    lr_mult:2
163  }
164  inner_product_param {
165    num_output:64
166    weight_filler {
167      type:"gaussian"
168      std:0.1
169    }
170    bias_filler {
171      type:"constant"
172    }
173  }
174 }
175 layer {
176  name:"ip2"
177  type:"InnerProduct"
178  bottom:"ip1"
179  top:"ip2"
180  param {
181    lr_mult:1
182  }
183  param {
184    lr_mult:2
185  }
186  inner_product_param {
187    num_output:10
188    weight_filler {
189      type:"gaussian"
190      std:0.1
191    }
192    bias_filler {
193      type:"constant"
194    }
195  }
196 }
197 layer {
198  name:"accuracy"
199  type:"Accuracy"
200  bottom:"ip2"
201  bottom:"label"
202  top:"accuracy"
203  include {
204    phase: TEST
205  }
206 }
207 layer {
208  name:"loss"
209  type:"SoftmaxWithLoss"
210  bottom:"ip2"
211  bottom:"label"
212  top:"loss"
213 }
214 layer {
215  name:"prob"
216  type:"Softmax"
217  bottom:"ip2"
218  top:"prob"
219  include {
220    phase: TEST
221  }
222 }
223 layer {
224  name:"argmax"
225  type:"ArgMax"
226  bottom:"prob"
227  top:"argmax"
228  include {
229    phase: TEST
230  }
231  argmax_param {
232    top_k:5
233  }
234 }
235 I0326 00:32:58.991350  3090 layer_factory.hpp:77]Creating layer data
236 I0326 00:32:58.991396  3090 net.cpp:91]CreatingLayer data
237 I0326 00:32:58.991411  3090 net.cpp:399] data -> data
238 I0326 00:32:58.991430  3090 net.cpp:399] data -> label
239 I0326 00:32:58.991448  3090 data_transformer.cpp:25]Loading mean file from:./examples/cifar10/mean.binaryproto
240 I0326 00:32:58.991549  3090 image_data_layer.cpp:38]Opening file ./examples/prediction_example/file_list.txt
241 I0326 00:32:58.991603  3090 image_data_layer.cpp:53] A total of 7 images.
242 I0326 00:32:58.995582  3090 image_data_layer.cpp:80] output data size:7,3,32,32
243 I0326 00:32:58.996433  3090 net.cpp:141]Setting up data
244 I0326 00:32:58.996472  3090 net.cpp:148]Top shape:733232(21504)
245 I0326 00:32:58.996482  3090 net.cpp:148]Top shape:7(7)
246 I0326 00:32:58.996489  3090 net.cpp:156]Memory required for data:86044
247 I0326 00:32:58.996500  3090 layer_factory.hpp:77]Creating layer label_data_1_split
248 I0326 00:32:58.996515  3090 net.cpp:91]CreatingLayer label_data_1_split
249 I0326 00:32:58.996523  3090 net.cpp:425] label_data_1_split <- label
250 I0326 00:32:58.996536  3090 net.cpp:399] label_data_1_split -> label_data_1_split_0
251 I0326 00:32:58.996549  3090 net.cpp:399] label_data_1_split -> label_data_1_split_1
252 I0326 00:32:58.996562  3090 net.cpp:141]Setting up label_data_1_split
253 I0326 00:32:58.996569  3090 net.cpp:148]Top shape:7(7)
254 I0326 00:32:58.996577  3090 net.cpp:148]Top shape:7(7)
255 I0326 00:32:58.996582  3090 net.cpp:156]Memory required for data:86100
256 I0326 00:32:58.996587  3090 layer_factory.hpp:77]Creating layer conv1
257 I0326 00:32:58.996606  3090 net.cpp:91]CreatingLayer conv1
258 I0326 00:32:58.996613  3090 net.cpp:425] conv1 <- data
259 I0326 00:32:58.996620  3090 net.cpp:399] conv1 -> conv1
260 I0326 00:32:59.145514  3090 net.cpp:141]Setting up conv1
261 I0326 00:32:59.145567  3090 net.cpp:148]Top shape:7323232(229376)
262 I0326 00:32:59.145577  3090 net.cpp:156]Memory required for data:1003604
263 I0326 00:32:59.145602  3090 layer_factory.hpp:77]Creating layer pool1
264 I0326 00:32:59.145617  3090 net.cpp:91]CreatingLayer pool1
265 I0326 00:32:59.145628  3090 net.cpp:425] pool1 <- conv1
266 I0326 00:32:59.145637  3090 net.cpp:399] pool1 -> pool1
267 I0326 00:32:59.145658  3090 net.cpp:141]Setting up pool1
268 I0326 00:32:59.145668  3090 net.cpp:148]Top shape:7321616(57344)
269 I0326 00:32:59.145673  3090 net.cpp:156]Memory required for data:1232980
270 I0326 00:32:59.145679  3090 layer_factory.hpp:77]Creating layer relu1
271 I0326 00:32:59.145689  3090 net.cpp:91]CreatingLayer relu1
272 I0326 00:32:59.145694  3090 net.cpp:425] relu1 <- pool1
273 I0326 00:32:59.145701  3090 net.cpp:386] relu1 -> pool1 (in-place)
274 I0326 00:32:59.145912  3090 net.cpp:141]Setting up relu1
275 I0326 00:32:59.145936  3090 net.cpp:148]Top shape:7321616(57344)
276 I0326 00:32:59.145943  3090 net.cpp:156]Memory required for data:1462356
277 I0326 00:32:59.145951  3090 layer_factory.hpp:77]Creating layer conv2
278 I0326 00:32:59.145963  3090 net.cpp:91]CreatingLayer conv2
279 I0326 00:32:59.145973  3090 net.cpp:425] conv2 <- pool1
280 I0326 00:32:59.145982  3090 net.cpp:399] conv2 -> conv2
281 I0326 00:32:59.146879  3090 net.cpp:141]Setting up conv2
282 I0326 00:32:59.146905  3090 net.cpp:148]Top shape:7321616(57344)
283 I0326 00:32:59.146913  3090 net.cpp:156]Memory required for data:1691732
284 I0326 00:32:59.146924  3090 layer_factory.hpp:77]Creating layer relu2
285 I0326 00:32:59.146934  3090 net.cpp:91]CreatingLayer relu2
286 I0326 00:32:59.146939  3090 net.cpp:425] relu2 <- conv2
287 I0326 00:32:59.146949  3090 net.cpp:386] relu2 -> conv2 (in-place)
288 I0326 00:32:59.147078  3090 net.cpp:141]Setting up relu2
289 I0326 00:32:59.147100  3090 net.cpp:148]Top shape:7321616(57344)
290 I0326 00:32:59.147106  3090 net.cpp:156]Memory required for data:1921108
291 I0326 00:32:59.147114  3090 layer_factory.hpp:77]Creating layer pool2
292 I0326 00:32:59.147121  3090 net.cpp:91]CreatingLayer pool2
293 I0326 00:32:59.147127  3090 net.cpp:425] pool2 <- conv2
294 I0326 00:32:59.147135  3090 net.cpp:399] pool2 -> pool2
295 I0326 00:32:59.147347  3090 net.cpp:141]Setting up pool2
296 I0326 00:32:59.147372  3090 net.cpp:148]Top shape:73288(14336)
297 I0326 00:32:59.147379  3090 net.cpp:156]Memory required for data:1978452
298 I0326 00:32:59.147387  3090 layer_factory.hpp:77]Creating layer conv3
299 I0326 00:32:59.147398  3090 net.cpp:91]CreatingLayer conv3
300 I0326 00:32:59.147408  3090 net.cpp:425] conv3 <- pool2
301 I0326 00:32:59.147416  3090 net.cpp:399] conv3 -> conv3
302 I0326 00:32:59.148576  3090 net.cpp:141]Setting up conv3
303 I0326 00:32:59.148602  3090 net.cpp:148]Top shape:76488(28672)
304 I0326 00:32:59.148608  3090 net.cpp:156]Memory required for data:2093140
305 I0326 00:32:59.148620  3090 layer_factory.hpp:77]Creating layer relu3
306 I0326 00:32:59.148628  3090 net.cpp:91]CreatingLayer relu3
307 I0326 00:32:59.148635  3090 net.cpp:425] relu3 <- conv3
308 I0326 00:32:59.148643  3090 net.cpp:386] relu3 -> conv3 (in-place)
309 I0326 00:32:59.148766  3090 net.cpp:141]Setting up relu3
310 I0326 00:32:59.148788  3090 net.cpp:148]Top shape:76488(28672)
311 I0326 00:32:59.148795  3090 net.cpp:156]Memory required for data:2207828
312 I0326 00:32:59.148802  3090 layer_factory.hpp:77]Creating layer pool3
313 I0326 00:32:59.148810  3090 net.cpp:91]CreatingLayer pool3
314 I0326 00:32:59.148816  3090 net.cpp:425] pool3 <- conv3
315 I0326 00:32:59.148824  3090 net.cpp:399] pool3 -> pool3
316 I0326 00:32:59.149027  3090 net.cpp:141]Setting up pool3
317 I0326 00:32:59.149052  3090 net.cpp:148]Top shape:76444(7168)
318 I0326 00:32:59.149058  3090 net.cpp:156]Memory required for data:2236500
319 I0326 00:32:59.149065  3090 layer_factory.hpp:77]Creating layer ip1
320 I0326 00:32:59.149077  3090 net.cpp:91]CreatingLayer ip1
321 I0326 00:32:59.149086  3090 net.cpp:425] ip1 <- pool3
322 I0326 00:32:59.149094  3090 net.cpp:399] ip1 -> ip1
323 I0326 00:32:59.149904  3090 net.cpp:141]Setting up ip1
324 I0326 00:32:59.149914  3090 net.cpp:148]Top shape:764(448)
325 I0326 00:32:59.149919  3090 net.cpp:156]Memory required for data:2238292
326 I0326 00:32:59.149927  3090 layer_factory.hpp:77]Creating layer ip2
327 I0326 00:32:59.149935  3090 net.cpp:91]CreatingLayer ip2
328 I0326 00:32:59.149941  3090 net.cpp:425] ip2 <- ip1
329 I0326 00:32:59.149948  3090 net.cpp:399] ip2 -> ip2
330 I0326 00:32:59.149968  3090 net.cpp:141]Setting up ip2
331 I0326 00:32:59.149976  3090 net.cpp:148]Top shape:710(70)
332 I0326 00:32:59.149981  3090 net.cpp:156]Memory required for data:2238572
333 I0326 00:32:59.149991  3090 layer_factory.hpp:77]Creating layer ip2_ip2_0_split
334 I0326 00:32:59.149998  3090 net.cpp:91]CreatingLayer ip2_ip2_0_split
335 I0326 00:32:59.150004  3090 net.cpp:425] ip2_ip2_0_split <- ip2
336 I0326 00:32:59.150012  3090 net.cpp:399] ip2_ip2_0_split -> ip2_ip2_0_split_0
337 I0326 00:32:59.150019  3090 net.cpp:399] ip2_ip2_0_split -> ip2_ip2_0_split_1
338 I0326 00:32:59.150027  3090 net.cpp:399] ip2_ip2_0_split -> ip2_ip2_0_split_2
339 I0326 00:32:59.150035  3090 net.cpp:141]Setting up ip2_ip2_0_split
340 I0326 00:32:59.150043  3090 net.cpp:148]Top shape:710(70)
341 I0326 00:32:59.150048  3090 net.cpp:148]Top shape:710(70)
342 I0326 00:32:59.150055  3090 net.cpp:148]Top shape:710(70)
343 I0326 00:32:59.150060  3090 net.cpp:156]Memory required for data:2239412
344 I0326 00:32:59.150066  3090 layer_factory.hpp:77]Creating layer accuracy
345 I0326 00:32:59.150076  3090 net.cpp:91]CreatingLayer accuracy
346 I0326 00:32:59.150082  3090 net.cpp:425] accuracy <- ip2_ip2_0_split_0
347 I0326 00:32:59.150089  3090 net.cpp:425] accuracy <- label_data_1_split_0
348 I0326 00:32:59.150095  3090 net.cpp:399] accuracy -> accuracy
349 I0326 00:32:59.150106  3090 net.cpp:141]Setting up accuracy
350 I0326 00:32:59.150113  3090 net.cpp:148]Top shape:(1)
351 I0326 00:32:59.150118  3090 net.cpp:156]Memory required for data:2239416
352 I0326 00:32:59.150125  3090 layer_factory.hpp:77]Creating layer loss
353 I0326 00:32:59.150133  3090 net.cpp:91]CreatingLayer loss
354 I0326 00:32:59.150140  3090 net.cpp:425] loss <- ip2_ip2_0_split_1
355 I0326 00:32:59.150146  3090 net.cpp:425] loss <- label_data_1_split_1
356 I0326 00:32:59.150152  3090 net.cpp:399] loss -> loss
357 I0326 00:32:59.150166  3090 layer_factory.hpp:77]Creating layer loss
358 I0326 00:32:59.150297  3090 net.cpp:141]Setting up loss
359 I0326 00:32:59.150308  3090 net.cpp:148]Top shape:(1)
360 I0326 00:32:59.150315  3090 net.cpp:151]     with loss weight 1
361 I0326 00:32:59.150331  3090 net.cpp:156]Memory required for data:2239420
362 I0326 00:32:59.150337  3090 layer_factory.hpp:77]Creating layer prob
363 I0326 00:32:59.150346  3090 net.cpp:91]CreatingLayer prob
364 I0326 00:32:59.150352  3090 net.cpp:425] prob <- ip2_ip2_0_split_2
365 I0326 00:32:59.150360  3090 net.cpp:399] prob -> prob
366 I0326 00:32:59.150557  3090 net.cpp:141]Setting up prob
367 I0326 00:32:59.150569  3090 net.cpp:148]Top shape:710(70)
368 I0326 00:32:59.150575  3090 net.cpp:156]Memory required for data:2239700
369 I0326 00:32:59.150581  3090 layer_factory.hpp:77]Creating layer argmax
370 I0326 00:32:59.150593  3090 net.cpp:91]CreatingLayer argmax
371 I0326 00:32:59.150599  3090 net.cpp:425] argmax <- prob
372 I0326 00:32:59.150605  3090 net.cpp:399] argmax -> argmax
373 I0326 00:32:59.150619  3090 net.cpp:141]Setting up argmax
374 I0326 00:32:59.150626  3090 net.cpp:148]Top shape:715(35)
375 I0326 00:32:59.150631  3090 net.cpp:156]Memory required for data:2239840
376 I0326 00:32:59.150637  3090 net.cpp:219] argmax does not need backward computation.
377 I0326 00:32:59.150643  3090 net.cpp:219] prob does not need backward computation.
378 I0326 00:32:59.150650  3090 net.cpp:217] loss needs backward computation.
379 I0326 00:32:59.150655  3090 net.cpp:219] accuracy does not need backward computation.
380 I0326 00:32:59.150661  3090 net.cpp:217] ip2_ip2_0_split needs backward computation.
381 I0326 00:32:59.150666  3090 net.cpp:217] ip2 needs backward computation.
382 I0326 00:32:59.150672  3090 net.cpp:217] ip1 needs backward computation.
383 I0326 00:32:59.150678  3090 net.cpp:217] pool3 needs backward computation.
384 I0326 00:32:59.150683  3090 net.cpp:217] relu3 needs backward computation.
385 I0326 00:32:59.150689  3090 net.cpp:217] conv3 needs backward computation.
386 I0326 00:32:59.150696  3090 net.cpp:217] pool2 needs backward computation.
387 I0326 00:32:59.150701  3090 net.cpp:217] relu2 needs backward computation.
388 I0326 00:32:59.150707  3090 net.cpp:217] conv2 needs backward computation.
389 I0326 00:32:59.150712  3090 net.cpp:217] relu1 needs backward computation.
390 I0326 00:32:59.150717  3090 net.cpp:217] pool1 needs backward computation.
391 I0326 00:32:59.150723  3090 net.cpp:217] conv1 needs backward computation.
392 I0326 00:32:59.150729  3090 net.cpp:219] label_data_1_split does not need backward computation.
393 I0326 00:32:59.150735  3090 net.cpp:219] data does not need backward computation.
394 I0326 00:32:59.150741  3090 net.cpp:261]This network produces output accuracy
395 I0326 00:32:59.150746  3090 net.cpp:261]This network produces output argmax
396 I0326 00:32:59.150753  3090 net.cpp:261]This network produces output loss
397 I0326 00:32:59.150768  3090 net.cpp:274]Network initialization done.
398 I0326 00:32:59.151636  3090 upgrade_proto.cpp:52]Attempting to upgrade input file specified using deprecated V1LayerParameter:./examples/cifar10/cifar10_quick_iter_5000.caffemodel
399 I0326 00:32:59.152022  3090 upgrade_proto.cpp:60]Successfully upgraded file specified using deprecated V1LayerParameter
400 I0326 00:32:59.152045  3090 net.cpp:753]Ignoring source layer cifar
401 W0326 00:32:59.152138  3090 net.hpp:41] DEPRECATED:ForwardPrefilled() will be removed in a future version.UseForward().
402 I0326 00:32:59.190719  3090 prediction_example.cpp:52]Result size:3
403 I0326 00:32:59.190753  3090 prediction_example.cpp:55]Blob size:0
404 I0326 00:32:59.190762  3090 prediction_example.cpp:58]-------------
405 I0326 00:32:59.190768  3090 prediction_example.cpp:59]  prediction :  
406 I0326 00:32:59.190779  3090 prediction_example.cpp:69]---------------------------------------------------------------
407 I0326 00:32:59.190786  3090 prediction_example.cpp:73]Pattern:0 class:5Prob=2.69745e-06
408 I0326 00:32:59.190804  3090 prediction_example.cpp:73]Pattern:1 class:3Prob=4.0121e-05
409 I0326 00:32:59.190811  3090 prediction_example.cpp:73]Pattern:2 class:9Prob=0.00531695
410 I0326 00:32:59.190819  3090 prediction_example.cpp:73]Pattern:3 class:6Prob=0.00282228
411 I0326 00:32:59.190825  3090 prediction_example.cpp:73]Pattern:4 class:0Prob=1.68176e-05
412 I0326 00:32:59.190832  3090 prediction_example.cpp:73]Pattern:5 class:4Prob=0.991599
413 I0326 00:32:59.190840  3090 prediction_example.cpp:73]Pattern:6 class:2Prob=9.24821e-05
414 I0326 00:32:59.190847  3090 prediction_example.cpp:75]-------------

View Code

各个类别图示：

3.后记

上面是用CPU跑的，我还等了几秒钟，用了下GPU处理，瞬间，真的很快，Enter完就出结果了。

参考文献：

（1）https://github.com/ikkiChung/Prediction-Example-With-Caffe

（2） http://ibuzzlog.blogspot.tw/2015/01/c-prediction-example-with-caffe-deep.html

（3） http://www.cnblogs.com/denny402/p/5111018.html

（4） https://github.com/BVLC/caffe/issues/2487

（5）http://pz124578126.lofter.com/post/1cd68c28_72625df

（6） https://groups.google.com/forum/#!topic/caffe-users/WO4gHU8rTCI

（7） https://groups.google.com/forum/#!topic/caffe-users/wWSGX4vmAh4

来自为知笔记(Wiz)

免责声明！

本站转载的文章为个人学习借鉴使用，本站对版权不负任何法律责任。如果侵犯了您的隐私权益，请联系本站邮箱yoyou2525@163.com删除。

猜您在找 用训练好的caffemodel对单个/批量图片进行分类 Caffe学习系列(20)：用训练好的caffemodel来进行分类【tensorflow2.0】使用spark-scala调用tensorflow2.0训练好的模型 Tensorflow 用训练好的模型预测在 C/C++ 中使用 TensorFlow 预训练好的模型—— 直接调用Ｃ++ 接口实现 caffe——利用训练好的模型批量预测将训练好的Tensorflow模型部署到web应用中使用训练好的word2vector进行文本聚类将训练好的tensorflow模型移植到android应用中 100天搞定机器学习：模型训练好了，然后呢？