-
Notifications
You must be signed in to change notification settings - Fork 0
/
conv_inference.py
executable file
·97 lines (85 loc) · 4.56 KB
/
conv_inference.py
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
import os
import numpy as np
import tensorflow as tf
from tensorflow.python.tools import freeze_graph
import tensorflow.contrib.slim as slim
import matplotlib.pyplot as plt
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
######################################
test_x = np.linspace(-0.5,0.5,50).reshape(-1,1,1,1)
test_y = np.array(np.ceil(np.sin(test_x*np.pi*6)).reshape(-1,1), dtype=np.int32)
# plt.scatter(test_x.reshape(-1).tolist(), test_y.reshape(-1),)
# plt.show()
#######################################
tf_x = tf.placeholder(dtype=tf.float32, shape=[None,1,1,1], name='h_input')
tf_label = tf.placeholder(dtype=tf.float32, shape=[None,1], name='h_label')
is_training = False
x = tf.layers.conv2d(inputs=tf_x, filters=12,
kernel_size=1, strides=[1, 1], kernel_initializer=tf.glorot_uniform_initializer(),
padding='VALID',
kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=5e-4), use_bias=True, name='h_conv_1')
x = tf.layers.batch_normalization(inputs=x, momentum=0.5, epsilon=1e-3,
scale=True,
center=True,
training=is_training,
fused=False,
name='h_bn_1')
x = tf.nn.tanh(x, name='h_act_1')
# x = tf.fake_quant_with_min_max_args(x, min=-1., max=1., name='h_fake_1')
x = tf.layers.conv2d(inputs=x, filters=12,
kernel_size=1, strides=[1, 1], kernel_initializer=tf.glorot_uniform_initializer(),
padding='VALID',
kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=5e-4), use_bias=False, name='h_conv_2')
x = tf.layers.batch_normalization(inputs=x, momentum=0.5, epsilon=1e-3,
scale=True,
center=True,
training=is_training,
fused=False,
name='h_bn_2')
x = tf.nn.relu6(x, name='h_act_2')
x = tf.fake_quant_with_min_max_args(x, min=0., max=6., name='h_fake_2')
x = tf.layers.conv2d(inputs=x, filters=12,
kernel_size=1, strides=[1, 1], kernel_initializer=tf.glorot_uniform_initializer(),
padding='VALID',
kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=5e-4), use_bias=True, name='h_conv_3')
x = tf.nn.sigmoid(x, name='h_act_3')
# x = tf.fake_quant_with_min_max_args(x, min=0., max=1., name='h_fake_3')
x = tf.layers.conv2d(inputs=x, filters=1,
kernel_size=1, strides=[1, 1], kernel_initializer=tf.glorot_uniform_initializer(),
padding='VALID',
kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=5e-4), use_bias=True, name='h_conv_4')
x = tf.reshape(x,[-1,1], 'h_reshape')
y = tf.identity(x, 'h_output')
loss = tf.reduce_mean(tf.square(tf_label-y), name='h_reducemean')
#############################################
tf.contrib.quantize.create_eval_graph(input_graph=tf.get_default_graph())
with tf.Session() as sess:
tf.global_variables_initializer().run(session=sess)
saver = tf.compat.v1.train.Saver()
saver.restore(sess=sess, save_path='./model/hxh')
# 保存图
tf.train.write_graph(sess.graph_def, 'output_model/pb_model', 'model.pb')
# 把图和参数结构一起
freeze_graph.freeze_graph('output_model/pb_model/model.pb', '', False, './model/hxh', 'h_output',
'save/restore_all', 'save/Const:0', 'output_model/pb_model/frozen_model.pb',
True, "")
writer = tf.summary.FileWriter('./board', sess.graph)
converter = tf.lite.TFLiteConverter.from_session(sess, input_tensors=[tf_x], output_tensors=[y])
converter.inference_type = tf.lite.constants.QUANTIZED_UINT8
input_arrays = converter.get_input_arrays()
converter.quantized_input_stats = {input_arrays[0]: (128., 118.)} # mean, std_dev
tflite_model = converter.convert()
open("./output_model/converted_model.tflite", "wb").write(tflite_model)
interpreter = tf.lite.Interpreter(model_content=tflite_model)
interpreter.allocate_tensors()
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()
res_y = []
for tmp_x in test_x:
interpreter.set_tensor(input_details[0]['index'], [np.array((tmp_x+0.5)*255,dtype=np.uint8)])
interpreter.invoke()
output_data = interpreter.get_tensor(output_details[0]['index'])
res_y.append(output_data[0])
print('loss = {}'.format(np.mean(test_y-res_y)))
print('done....')