forked from cinastanbean/pytorch-onnx-tensorflow-pb
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Converting A PyTorch Model to Tensorflow pb using ONNX.html
501 lines (418 loc) · 42.2 KB
/
Converting A PyTorch Model to Tensorflow pb using ONNX.html
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
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
<!DOCTYPE html><html>
<head>
<meta charset="utf-8">
<title></title>
<style>
html,body,div,span,applet,object,iframe,h1,h2,h3,h4,h5,h6,p,blockquote,pre,a,abbr,acronym,address,big,cite,code,del,dfn,em,img,ins,kbd,q,s,samp,small,strike,strong,sub,sup,tt,var,b,u,i,center,dl,dt,dd,ol,ul,li,fieldset,form,label,legend,table,caption,tbody,tfoot,thead,tr,th,td,article,aside,canvas,details,embed,figure,figcaption,footer,header,hgroup,menu,nav,output,ruby,section,summary,time,mark,audio,video{margin:0;padding:0;border:0;font:inherit;font-size:100%;vertical-align:baseline}
html{line-height:1}
ol,ul{list-style:none}
table{border-collapse:collapse;border-spacing:0;
margin-top: 0;
margin-bottom: 0.8em;
}
caption,th,td{text-align:left;font-weight:normal;vertical-align:middle}
q,blockquote{quotes:none}
q:before,q:after,blockquote:before,blockquote:after{content:"";content:none}
a img{border:none}
article,aside,details,figcaption,figure,footer,header,hgroup,menu,nav,section,summary{display:block}
a{color:#1863a1}
a:visited{color:#751590}
a:focus{color:#0181eb}
a:hover{color:#0181eb}
a:active{color:#01579f}
aside.sidebar a{color:#222}
aside.sidebar a:focus{color:#0181eb}
aside.sidebar a:hover{color:#0181eb}
aside.sidebar a:active{color:#01579f}
a{-webkit-transition:color 0.3s;-moz-transition:color 0.3s;-o-transition:color 0.3s;transition:color 0.3s}
body,h1,h2,h3,h4,h5,h6,footer{font-family:"PT Sans","Helvetica Neue","Optima","Hiragino Sans GB",sans-serif}
body{line-height:1.5em;color:#222
-webkit-text-size-adjust:none; min-width: 200px;
max-width: 760px;
margin: 0 auto; padding: 1em;}
pre,code,tt,p code,li code{font-family:Menlo,Monaco,"Andale Mono","lucida console","Courier New",monospace}
h1{font-size:2.2em;line-height:1.2em}
h1,h2,h3,h4,h5,h6{text-rendering:optimizelegibility;margin-bottom:1em;font-weight:bold}
h2,section h1{font-size:1.5em}
h3,section h2,section section h1{font-size:1.3em}
h4,section h3,section section h2,section section section h1{font-size:1em}
h5,section h4,section section h3{font-size:.9em}
h6,section h5,section section h4,section section section h3{font-size:.8em}
.markdown-body{padding: 0px 4px;}
.markdown-body h1{position:relative;padding-top:1em;padding-bottom:0.2em;margin-bottom:1em;background:url('data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAACgAAAABCAYAAACsXeyTAAAACXBIWXMAAAsTAAALEwEAmpwYAAAAFUlEQVQIHWNIS0sr/v//PwMMDzY+ADqMahlW4J91AAAAAElFTkSuQmCC') bottom left repeat-x}
.markdown-body h1 a{text-decoration:none}
.markdown-body h1 a:hover{text-decoration:underline}
.markdown-body h2{padding-top:0.8em;padding-bottom:0.2em;background:url('data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAACgAAAABCAYAAACsXeyTAAAACXBIWXMAAAsTAAALEwEAmpwYAAAAFUlEQVQIHWNIS0sr/v//PwMMDzY+ADqMahlW4J91AAAAAElFTkSuQmCC') bottom left repeat-x}
.markdown-body h2:first-child,.markdown-body header+h2{padding-top:0}
.markdown-body h2:first-child,.markdown-body header+h2{background:none}
p,.markdown-body blockquote,ul,ol{margin-bottom:0.8em; margin-top:0.8em;}
ul{list-style-type:disc}
ul ul{list-style-type:circle;margin-bottom:0px}
ul ul ul{list-style-type:square;margin-bottom:0px}
ol{list-style-type:decimal}
ol ol{list-style-type:lower-alpha;margin-bottom:0px}
ol ol ol{list-style-type:lower-roman;margin-bottom:0px}
ul,ul ul,ul ol,ol,ol ul,ol ol{margin-left:1.3em}
ul ul,ul ol,ol ul,ol ol{margin-bottom:0em}
strong{font-weight:bold}
em{font-style:italic}
sup,sub{font-size:0.75em;position:relative;display:inline-block;padding:0 .2em;line-height:.8em}
sup{top:-.5em}
sub{bottom:-.5em}
a[rev='footnote']{font-size:.75em;padding:0 .3em;line-height:1}
q{font-style:italic}
q:before{content:"\201C"}
q:after{content:"\201D"}
em,dfn{font-style:italic}
strong,dfn{font-weight:bold}
del,s{text-decoration:line-through}
abbr,acronym{border-bottom:1px dotted;cursor:help}
small{font-size:.8em}
big{font-size:1.2em}
.markdown-body hr {
height: 0;
margin: 15px 0;
overflow: hidden;
background: transparent;
border: 0;
border-bottom: 1px solid #ddd;
}
.markdown-body hr:before {
display: table;
content: "";
}
.markdown-body hr:after {
display: table;
clear: both;
content: "";
}
.markdown-body table {
display: block;
width: 100%;
overflow: auto;
word-break: normal;
word-break: keep-all;
}
.markdown-body table th {
font-weight: bold;
}
.markdown-body table th,
.markdown-body table td {
padding: 6px 13px;
border: 1px solid #ddd;
}
.markdown-body table tr {
background-color: #fff;
border-top: 1px solid #ccc;
}
.markdown-body table tr:nth-child(2n) {
background-color: #f8f8f8;
}
.markdown-body blockquote{font-style:italic;position:relative;font-size:1.2em;line-height:1.5em;padding-left:1em;border-left:4px solid rgba(170,170,170,0.5)}
.markdown-body blockquote cite{font-style:italic}
.markdown-body blockquote cite a{color:#aaa !important;word-wrap:break-word}
.markdown-body blockquote cite:before{content:'\2014';padding-right:.3em;padding-left:.3em;color:#aaa}
.markdown-body a{white-space:-moz-pre-wrap;white-space:-pre-wrap;white-space:-o-pre-wrap;white-space:pre-wrap;word-wrap:break-word}
body>header{font-size:1em;padding-top:1.5em;padding-bottom:1.5em}
.markdown-body{overflow:hidden}
.markdown-body>div,.markdown-body>article{width:100%}
aside.sidebar{float:none;padding:0 18px 1px;background-color:#f7f7f7;border-top:1px solid #e0e0e0}
.flex-content,article img,article video,article .flash-video,aside.sidebar img{max-width:100%;height:auto}
.basic-alignment.left,article img.left,article video.left,article .left.flash-video,aside.sidebar img.left{float:left;margin-right:1.5em}
.basic-alignment.right,article img.right,article video.right,article .right.flash-video,aside.sidebar img.right{float:right;margin-left:1.5em}
.basic-alignment.center,article img.center,article video.center,article .center.flash-video,aside.sidebar img.center{display:block;margin:0 auto 1.5em}
.basic-alignment.left,article img.left,article video.left,article .left.flash-video,aside.sidebar img.left,.basic-alignment.right,article img.right,article video.right,article .right.flash-video,aside.sidebar img.right{margin-bottom:.8em}
.toggle-sidebar,.no-sidebar .toggle-sidebar{display:none}
.markdown-body img,.markdown-body video,.markdown-body .flash-video{ -webkit-border-radius:0.3em;-moz-border-radius:0.3em;-ms-border-radius:0.3em;-o-border-radius:0.3em;border-radius:0.3em;-webkit-box-shadow:rgba(0,0,0,0.15) 0 1px 4px;-moz-box-shadow:rgba(0,0,0,0.15) 0 1px 4px;box-shadow:rgba(0,0,0,0.15) 0 1px 4px;-webkit-box-sizing:border-box;-moz-box-sizing:border-box;box-sizing:border-box;border:#fff 0.5em solid;}
.markdown-body img,.markdown-body video{max-width: 100%;}
.markdown-body video,.markdown-body .flash-video{margin:0 auto 1.5em}
.markdown-body video{display:block;width:100%}
.markdown-body .flash-video>div{position:relative;display:block;padding-bottom:56.25%;padding-top:1px;height:0;overflow:hidden}
.markdown-body .flash-video>div iframe,.markdown-body .flash-video>div object,.markdown-body .flash-video>div embed{position:absolute;top:0;left:0;width:100%;height:100%}
.markdown-body>footer{padding-bottom:2.5em;margin-top:2em}
.markdown-body>footer p.meta{margin-bottom:.8em;font-size:.85em;clear:both;overflow:hidden}
body,pre{ background:#fdf6e3 url('data:image/png;base64,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') top left;}
body{ background-color: #f8f8f8;}
pre{-webkit-border-radius:0.4em;-moz-border-radius:0.4em;-ms-border-radius:0.4em;-o-border-radius:0.4em;border-radius:0.4em;border:1px solid #e7dec3;line-height:1.45em;font-size:13px;margin-bottom:2.1em;padding:.8em 1em;color:#586e75;overflow:auto}
.markdown-body code{background: none;}
h3.filename+pre{-moz-border-radius-topleft:0px;-webkit-border-top-left-radius:0px;border-top-left-radius:0px;-moz-border-radius-topright:0px;-webkit-border-top-right-radius:0px;border-top-right-radius:0px}
p code,li code{display:inline-block;white-space:no-wrap;background:#fff;font-size:.8em;line-height:1.5em;color:#555;border:1px solid #ddd;-webkit-border-radius:0.4em;-moz-border-radius:0.4em;-ms-border-radius:0.4em;-o-border-radius:0.4em;border-radius:0.4em;padding:0 .3em;margin:-1px 0}
p pre code,li pre code{font-size:1em !important;background:none;border:none}
/*
Orginal Style from ethanschoonover.com/solarized (c) Jeremy Hull <[email protected]>
*/
.hljs {
display: block;
overflow-x: auto;
padding: 0.5em;
background: #fdf6e3;
color: #657b83;
-webkit-text-size-adjust: none;
}
.hljs-comment,
.diff .hljs-header,
.hljs-doctype,
.hljs-pi,
.lisp .hljs-string {
color: #93a1a1;
}
/* Solarized Green */
.hljs-keyword,
.hljs-winutils,
.method,
.hljs-addition,
.css .hljs-tag,
.hljs-request,
.hljs-status,
.nginx .hljs-title {
color: #859900;
}
/* Solarized Cyan */
.hljs-number,
.hljs-command,
.hljs-string,
.hljs-tag .hljs-value,
.hljs-rule .hljs-value,
.hljs-doctag,
.tex .hljs-formula,
.hljs-regexp,
.hljs-hexcolor,
.hljs-link_url {
color: #2aa198;
}
/* Solarized Blue */
.hljs-title,
.hljs-localvars,
.hljs-chunk,
.hljs-decorator,
.hljs-built_in,
.hljs-identifier,
.vhdl .hljs-literal,
.hljs-id,
.css .hljs-function,
.hljs-name {
color: #268bd2;
}
/* Solarized Yellow */
.hljs-attribute,
.hljs-variable,
.lisp .hljs-body,
.smalltalk .hljs-number,
.hljs-constant,
.hljs-class .hljs-title,
.hljs-parent,
.hljs-type,
.hljs-link_reference {
color: #b58900;
}
/* Solarized Orange */
.hljs-preprocessor,
.hljs-preprocessor .hljs-keyword,
.hljs-pragma,
.hljs-shebang,
.hljs-symbol,
.hljs-symbol .hljs-string,
.diff .hljs-change,
.hljs-special,
.hljs-attr_selector,
.hljs-subst,
.hljs-cdata,
.css .hljs-pseudo,
.hljs-header {
color: #cb4b16;
}
/* Solarized Red */
.hljs-deletion,
.hljs-important {
color: #dc322f;
}
/* Solarized Violet */
.hljs-link_label {
color: #6c71c4;
}
.tex .hljs-formula {
background: #eee8d5;
}
</style>
<style> @media print{ .hljs{overflow: visible; word-wrap: break-word !important;} }</style></head><body><div class="markdown-body">
<h1 id="toc_0">Converting A PyTorch Model to Tensorflow pb using ONNX</h1>
<p align="right">[email protected]</p>
<ul>
<li>
<a href="#toc_0">Converting A PyTorch Model to Tensorflow pb using ONNX</a>
</li>
<li>
<a href="#toc_1">1. Pre-installation</a>
</li>
<li>
<a href="#toc_2">2. 转换过程</a>
<ul>
<li>
<a href="#toc_3">2.1 Step 1.2.3.</a>
</li>
<li>
<a href="#toc_4">2.2 Verification</a>
</li>
</ul>
</li>
<li>
<a href="#toc_5">3. Related Info</a>
<ul>
<li>
<a href="#toc_6">3.1 ONNX</a>
</li>
<li>
<a href="#toc_7">3.2 Microsoft/MMdnn</a>
</li>
</ul>
</li>
<li>
<a href="#toc_8">Reference</a>
</li>
</ul>
<h1 id="toc_1">1. Pre-installation</h1>
<p><strong>Version Info</strong></p>
<pre><code>pytorch 0.4.0 py27_cuda0.0_cudnn0.0_1 pytorch
torchvision 0.2.1 py27_1 pytorch
tensorflow 1.8.0 <pip>
onnx 1.2.2 <pip>
onnx-tf 1.1.2 <pip>
</code></pre>
<p>注意：</p>
<ol>
<li>ONNX1.1.2版本太低会引发BatchNormalization错误，当前pip已经支持1.3.0版本；也可以考虑源码安装 <code>pip install -U git+https://github.com/onnx/onnx.git@master</code>。</li>
<li>本实验验证ONNX1.2.2版本可正常运行</li>
<li>onnx-tf采用源码安装；要求 Tensorflow>=1.5.0.；</li>
</ol>
<h1 id="toc_2">2. 转换过程</h1>
<h2 id="toc_3">2.1 Step 1.2.3.</h2>
<p><strong>pipeline: pytorch model --> onnx modle --> tensorflow graph pb.</strong></p>
<pre><code># step 1, load pytorch model and export onnx during running.
modelname = 'resnet18'
weightfile = 'models/model_best_checkpoint_resnet18.pth.tar'
modelhandle = DIY_Model(modelname, weightfile, class_numbers)
model = modelhandle.model
#model.eval() # useless
dummy_input = Variable(torch.randn(1, 3, 224, 224)) # nchw
onnx_filename = os.path.split(weightfile)[-1] + ".onnx"
torch.onnx.export(model, dummy_input,
onnx_filename,
verbose=True)
# step 2, create onnx_model using tensorflow as backend. check if right and export graph.
onnx_model = onnx.load(onnx_filename)
tf_rep = prepare(onnx_model, strict=False)
# install onnx-tensorflow from github，and tf_rep = prepare(onnx_model, strict=False)
# Reference https://github.com/onnx/onnx-tensorflow/issues/167
#tf_rep = prepare(onnx_model) # whthout strict=False leads to KeyError: 'pyfunc_0'
image = Image.open('pants.jpg')
# debug, here using the same input to check onnx and tf.
output_pytorch, img_np = modelhandle.process(image)
print('output_pytorch = {}'.format(output_pytorch))
output_onnx_tf = tf_rep.run(img_np)
print('output_onnx_tf = {}'.format(output_onnx_tf))
# onnx --> tf.graph.pb
tf_pb_path = onnx_filename + '_graph.pb'
tf_rep.export_graph(tf_pb_path)
# step 3, check if tf.pb is right.
with tf.Graph().as_default():
graph_def = tf.GraphDef()
with open(tf_pb_path, "rb") as f:
graph_def.ParseFromString(f.read())
tf.import_graph_def(graph_def, name="")
with tf.Session() as sess:
#init = tf.initialize_all_variables()
init = tf.global_variables_initializer()
#sess.run(init)
# print all ops, check input/output tensor name.
# uncomment it if you donnot know io tensor names.
'''
print('-------------ops---------------------')
op = sess.graph.get_operations()
for m in op:
print(m.values())
print('-------------ops done.---------------------')
'''
input_x = sess.graph.get_tensor_by_name("0:0") # input
outputs1 = sess.graph.get_tensor_by_name('add_1:0') # 5
outputs2 = sess.graph.get_tensor_by_name('add_3:0') # 10
output_tf_pb = sess.run([outputs1, outputs2], feed_dict={input_x:img_np})
#output_tf_pb = sess.run([outputs1, outputs2], feed_dict={input_x:np.random.randn(1, 3, 224, 224)})
print('output_tf_pb = {}'.format(output_tf_pb))
</code></pre>
<h2 id="toc_4">2.2 Verification</h2>
<p><strong>确保输出结果一致</strong></p>
<pre><code>output_pytorch = [array([ 2.5359073 , -1.4261041 , -5.2394 , -0.62402934, 4.7426634 ], dtype=float32), array([ 7.6249304, 5.1203837, 1.8118637, 1.5143847, -4.9409146, 1.1695148, -6.2375665, -1.6033885, -1.4286405, -2.964429 ], dtype=float32)]
output_onnx_tf = Outputs(_0=array([[ 2.5359051, -1.4261056, -5.239397 , -0.6240269, 4.7426634]], dtype=float32), _1=array([[ 7.6249285, 5.12038 , 1.811865 , 1.5143874, -4.940915 , 1.1695154, -6.237564 , -1.6033876, -1.4286422, -2.964428 ]], dtype=float32))
output_tf_pb = [array([[ 2.5359051, -1.4261056, -5.239397 , -0.6240269, 4.7426634]], dtype=float32), array([[ 7.6249285, 5.12038 , 1.811865 , 1.5143874, -4.940915 , 1.1695154, -6.237564 , -1.6033876, -1.4286422, -2.964428 ]], dtype=float32)]
</code></pre>
<p><strong>独立TF验证程序</strong></p>
<pre><code>def get_img_np_nchw(filename):
try:
image = Image.open(filename).convert('RGB').resize((224, 224))
miu = np.array([0.485, 0.456, 0.406])
std = np.array([0.229, 0.224, 0.225])
#miu = np.array([0.5, 0.5, 0.5])
#std = np.array([0.22, 0.22, 0.22])
# img_np.shape = (224, 224, 3)
img_np = np.array(image, dtype=float) / 255.
r = (img_np[:,:,0] - miu[0]) / std[0]
g = (img_np[:,:,1] - miu[1]) / std[1]
b = (img_np[:,:,2] - miu[2]) / std[2]
img_np_t = np.array([r,g,b])
img_np_nchw = np.expand_dims(img_np_t, axis=0)
return img_np_nchw
except:
print("RuntimeError: get_img_np_nchw({}).".format(filename))
# NoneType
if __name__ == '__main__':
tf_pb_path = 'model_best_checkpoint_resnet18.pth.tar.onnx_graph.pb'
filename = 'pants.jpg'
img_np_nchw = get_img_np_nchw(filename)
# step 3, check if tf.pb is right.
with tf.Graph().as_default():
graph_def = tf.GraphDef()
with open(tf_pb_path, "rb") as f:
graph_def.ParseFromString(f.read())
tf.import_graph_def(graph_def, name="")
with tf.Session() as sess:
init = tf.global_variables_initializer()
#init = tf.initialize_all_variables()
sess.run(init)
# print all ops, check input/output tensor name.
# uncomment it if you donnot know io tensor names.
'''
print('-------------ops---------------------')
op = sess.graph.get_operations()
for m in op:
print(m.values())
print('-------------ops done.---------------------')
'''
input_x = sess.graph.get_tensor_by_name("0:0") # input
outputs1 = sess.graph.get_tensor_by_name('add_1:0') # 5
outputs2 = sess.graph.get_tensor_by_name('add_3:0') # 10
output_tf_pb = sess.run([outputs1, outputs2], feed_dict={input_x:img_np_nchw})
print('output_tf_pb = {}'.format(output_tf_pb))
</code></pre>
<h1 id="toc_5">3. Related Info</h1>
<h2 id="toc_6">3.1 ONNX</h2>
<p>Open Neural Network Exchange<br/>
<a href="https://github.com/onnx">https://github.com/onnx</a><br/>
<a href="https://onnx.ai/">https://onnx.ai/</a></p>
<p>The ONNX exporter is a <mark><strong>trace-based</strong></mark> exporter, which means that it operates by executing your model once, and exporting the operators which were actually run during this run. <a href="https://pytorch.org/docs/stable/onnx.html#example-end-to-end-alexnet-from-pytorch-to-caffe2">Limitations</a></p>
<p><a href="https://github.com/onnx/tensorflow-onnx">https://github.com/onnx/tensorflow-onnx</a><br/>
<a href="https://github.com/onnx/onnx-tensorflow">https://github.com/onnx/onnx-tensorflow</a></p>
<h2 id="toc_7">3.2 Microsoft/MMdnn</h2>
<p>当前网络没有调通<br/>
<a href="https://github.com/Microsoft/MMdnn/blob/master/mmdnn/conversion/pytorch/README.md">https://github.com/Microsoft/MMdnn/blob/master/mmdnn/conversion/pytorch/README.md</a></p>
<h1 id="toc_8">Reference</h1>
<ol>
<li>Open Neural Network Exchange <a href="https://github.com/onnx">https://github.com/onnx</a></li>
<li><a href="https://github.com/onnx/tutorials/blob/master/tutorials/PytorchOnnxExport.ipynb">Exporting model from PyTorch to ONNX</a></li>
<li><a href="https://github.com/onnx/tutorials/blob/master/tutorials/OnnxTensorflowImport.ipynb">Importing ONNX models to Tensorflow(ONNX)</a></li>
<li><a href="https://zhuanlan.zhihu.com/p/26136080">Tensorflow + tornado服务</a></li>
<li><a href="https://github.com/llSourcell/tensorflow_image_classifier/blob/master/src/label_image.py">graph_def = tf.GraphDef() graph_def.ParseFromString(f.read())</a></li>
<li><a href="https://www.tensorflow.org/extend/tool_developers/">A Tool Developer's Guide to TensorFlow Model Files</a></li>
<li><a href="https://www.jianshu.com/p/613c3b08faea">TensorFlow学习笔记：Retrain Inception_v3</a></li>
</ol>
</div></body>
</html>