-
Notifications
You must be signed in to change notification settings - Fork 8
/
tensorflow2caffe.py
159 lines (149 loc) · 6.42 KB
/
tensorflow2caffe.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
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
#author: lxy
#time: 2018.4.3/ 11:30
#tool: python3
#version: 0.1
#project: face-detect
#modify:
#########################################
import numpy as np
import sys
import os
sys.path.append("/home/lxy/caffe/python")
os.environ['GLOG_minloglevel'] = '2'
import caffe
#import caffe.net as caffe_net
import tensorflow as tf
#from caffe import caffe_net
from train_models.MTCNN_config import config
from train_models.mtcnn_model import P_Net,R_Net,O_Net
import argparse
GLOG_minloglevel=1
def convert_filter(numpy_filter_weight):
return np.transpose(numpy_filter_weight,(3,2,1,0))
def convert_fc(numpy_fc_weight):
return np.transpose(numpy_fc_weight,(1,0))
def get_tf(model_file,train_net):
#sess = tf.Session()
graph = tf.Graph()
test_fg = config.train_face
key_list = []
var_dic = dict()
with graph.as_default():
if str(train_net) == "PNet":
image_op = tf.placeholder(tf.float32, shape=[1, 12, 12, 3], name='input')
net_factory = P_Net
elif str(train_net) == 'RNet':
image_op = tf.placeholder(tf.float32, shape=[1, 24, 24, 3], name='input')
net_factory = R_Net
elif str(train_net) == 'ONet':
image_op = tf.placeholder(tf.float32, shape=[1, 48, 48, 3], name='input')
net_factory = O_Net
#figure out landmark
if test_fg:
cls_prob, bbox_pred, landmark_pred = net_factory(image_op, training=False)
else:
cls_prob, bbox_pred = net_factory(image_op, training=False)
sess = tf.Session(config=tf.ConfigProto(log_device_placement=False))
#new_saver = tf.train.import_meta_graph(model_file+".meta")
saver = tf.train.Saver(max_to_keep=0)
module_ = saver.restore(sess,model_file)
all_vars = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
conv1 = all_vars[0]
bias1 = all_vars[1]
#for var_ in all_vars:
#print(var_)
for v_name in tf.global_variables():
print("name : ",v_name.name[:-2],v_name.shape)
key_list.append(v_name.name[:-2])
#print(tf.get_variable_scope())
conv_w1, bias_1 , vas= sess.run([conv1,bias1,all_vars])
print("conv ",np.shape(conv_w1))
print("bn ",np.shape(bias_1))
print(len(vas))
for i in range(len(vas)):
cur_name = key_list[i]
cur_var = vas[i]
if "weight" in cur_name and "fc" not in cur_name:
cur_var = convert_filter(cur_var)
var_dic[cur_name] = cur_var
elif "fc" in cur_name and "weight" in cur_name:
cur_var = convert_fc(cur_var)
var_dic[cur_name] = cur_var
else:
var_dic[cur_name]= cur_var
print("dic0: ",np.shape(var_dic['conv1/weights']))
return var_dic
def get_caffe(var_dic,protxt_path,caffe_model_path):
net = caffe.Net(protxt_path, caffe.TEST)
'''
for layer_name, blob in net.blobs.iteritems():
print (layer_name + '\t' + str(blob.data.shape) )
'''
print("begin to out param :")
#print(net.params('conv1', 1).get_data().shape)
for layer_name, param in net.params.iteritems():
print(layer_name)
if 'conv' in layer_name:
#print(net.params('conv1', 1).get_data())
print (layer_name,param[0].data.shape,np.shape(var_dic[layer_name+'/weights']))
print(layer_name+'/weights')
param[0].data[:,:,:,:] = var_dic[layer_name+'/weights']
#param[0].set_data(var_dic[layer_name+'/weights'])
param[1].data[:] = var_dic[layer_name+'/biases']
elif "fc1" in layer_name:
print (layer_name,param[0].data.shape,np.shape(var_dic[layer_name+'/weights']))
param[0].data[:,:] = var_dic[layer_name+'/weights']
param[1].data[:] = var_dic[layer_name+'/biases']
elif "cls" in layer_name:
print (layer_name,param[0].data.shape,np.shape(var_dic[layer_name+'/weights']))
param[0].data[:,:] = var_dic[layer_name+'/weights']
param[1].data[:] = var_dic[layer_name+'/biases']
elif "bbox" in layer_name:
print (layer_name,param[0].data.shape,np.shape(var_dic[layer_name+'/weights']))
param[0].data[:,:] = var_dic[layer_name+'/weights']
param[1].data[:] = var_dic[layer_name+'/biases']
elif "landmark" in layer_name:
print (layer_name,param[0].data.shape,np.shape(var_dic[layer_name+'/weights']))
param[0].data[:,:] = var_dic[layer_name+'/weights']
param[1].data[:] = var_dic[layer_name+'/biases']
net.save(caffe_model_path)
#def gen_caffe():
#caffe_net.Caffemodel('')
def load_caffe(pro_path,model_path):
net = caffe.Net(protxt_path,model_path, caffe.TEST)
for layer_name, blob in net.blobs.iteritems():
print (layer_name + '\t' + str(blob.data.shape) )
def args():
parser = argparse.ArgumentParser(description="tensorflow to caffe")
parser.add_argument('--test_net',type=str,default="PNet",help="which net to convert")
parser.add_argument('--test_load',type=bool,default=False,help="whether convert is successful")
return parser.parse_args()
if __name__=='__main__':
#os.environ(::google::InitGoogleLogging(" "))
arg = args()
test_net = arg.test_net
test_fg = arg.test_load
if test_net=="PNet":
#path_ = './data/MTCNN_bright_model/PNet_landmark/PNet-205'
#path_= './data/MTCNN_model/PNet_landmark/v1_trained/PNet-32'
path_= './data/MTCNN_model/PNet_landmark/resaved/PNet-5relu'
protxt_path = './caffe/PNet.prototxt'
caffe_model_path = './caffe/PNet.caffemodel'
net_factory = "PNet"
elif test_net=="RNet":
#path_= './data/MTCNN_model/RNet_landmark/v1_trained/RNet-4400'
path_= './data/MTCNN_model/RNet_landmark/resaved/RNet-40relu'
protxt_path = './caffe/RNet.prototxt'
caffe_model_path = './caffe/RNet.caffemodel'
net_factory = 'RNet'
elif test_net=="ONet":
#path_= './data/MTCNN_model/ONet_landmark/v1_trained/ONet-25'
path_= './data/MTCNN_model/ONet_landmark/resaved/ONet-60relu'
protxt_path = './caffe/ONet.prototxt'
caffe_model_path = './caffe/ONet.caffemodel'
net_factory = 'ONet'
if test_fg:
load_caffe(path_,caffe_model_path)
else:
parameters = get_tf(path_,net_factory)
get_caffe(parameters,protxt_path,caffe_model_path)