def data_init(self):
save_path = r"model_saver\AE_circle"
out_dir_prefix = os.path.join(save_path, "model")
height = 64
width = 64
epochs = 50
GPU_ratio = 0.5
batch_size = 12
train_ratio = 0.7
pic_path = r'./good samples'
(self.x_train, self.x_train_label, no1, no2) = cm.data_load(pic_path, train_ratio=1, resize=(width, height), \
has_dir=False)
print(self.x_train.shape)
# self.ui.plainTextEdit.appendPlainText('x_train.shape = {}'.format(x_train.shape))
print('x_train.shape = {}'.format(self.x_train.shape))
# self.ui.label.setText('x_train.shape = {}'.format(x_train.shape))
# prepare test data
pic_path = r'./xxx'
(self.x_train_2, self.x_train_label_2, self.x_test_2, self.x_test_label_2) = cm.data_load(pic_path, train_ratio,
resize=(width, height), shuffle=True,
normalize=True)
print('x_test shape = ', self.x_test_2.shape)
# self.ui.plainTextEdit.appendPlainText('x_test shape = {}'.format(x_test_2.shape))
print('x_test shape = {}'.format(self.x_test_2.shape))
print('x_test_label shape = ', self.x_test_label_2.shape)
# self.ui.plainTextEdit.appendPlainText('xx_test_label shape = {}'.format(x_test_label_2.shape))
print('x_test_label shape = {}'.format(self.x_test_label_2.shape))
def data_init(self):
pic_path = r'D:\dataset\xxx'
(x_train, x_train_label, no1, no2) = cm.data_load(pic_path,
train_ratio=1,
resize=(width,
height),
has_dir=True)
print(x_train.shape)
print(x_train_label)
def run(self):
try:
print("Start train data pre-process\n")
(x_train, y_train_label, no1, no2) = cm.data_load(self.train_dir_name, train_ratio=self.train_ratio,
resize=(self.resize_width,self.resize_height),
has_dir=self.train_has_dir)
print('Training data shape = {}'.format(x_train.shape))
print("Start test data pre-process\n")
(x_train_2, y_train_label_2, x_test, y_test_label) = cm.data_load(self.test_dir_name, train_ratio=0,
resize=(self.resize_width, self.resize_height),
has_dir=self.test_has_dir)
print('Test data shape = {}'.format(x_test.shape))
print('Test label shape = {}'.format(y_test_label.shape))
except:
print("在進行訓練資料前處理時發生錯誤")
self.ae = AI.AE(input_dim=[None, self.resize_width, self.resize_height, 3])
self.ae.train(x_train,x_test,y_test_label,self.GPU_ratio,self.epoch,
self.batch_size,self.fine_tune,self.save_ckpt)
self.finish_flag.emit()
# 'model_saver/'為置放的資料夾，'combined_model.pb'為檔名
# with tf.gfile.GFile("model_saver/pb_CLASF_circle__model.pb", "wb") as f:
# f.write(output_graph_def.SerializeToString())
# print("PB file saved successfully")
if __name__ == "__main__":
save_path = r"model_saver\binary_CLASF_circle"
out_dir_prefix = os.path.join(save_path, "model")
height = 64
width = 64
epochs = 20
GPU_ratio = 0.8
batch_size = 8
train_ratio = 0.7
# prepare training data
pic_path = r'D:\dataset\Halcon_circle'
classifier_num = 2
(x_train, x_train_label, x_test, x_test_label) = cm.data_load(pic_path, train_ratio=train_ratio, resize=(width, height), has_dir=True)
print("x_train shape = ",x_train.shape)
print("x_test shape = ", x_test.shape)
x_train_label = np_utils.to_categorical(x_train_label, classifier_num)
x_test_label = np_utils.to_categorical(x_test_label, classifier_num)
print("train label shape = ",x_train_label.shape)
print("test label shape = ", x_test_label.shape)
CLASF = Classification(classifier_num,input_dim = [None,width,height,3], save_path=save_path)
CLASF.train(x_train, x_train_label, x_test, x_test_label, GPU_ratio=GPU_ratio, epochs=epochs,
batch_size=batch_size, fine_tune=True)
return train_ave_loss, test_ave_loss, acc
if __name__ == "__main__":
#prepare training data
# pic_path = r'E:\dataset\forAE\pill\train\Good'
pic_path = r'E:\dataset\forAE\circle\train\Good'
#pic_path = r"G:/我的雲端硬碟/Python/Code/Pycharm/P_UI_Test/goodsamples"
# pic_path = r"G://我的雲端硬碟//Python//Code//Pycharm//P_UI_Test//goodsamples"
#pic_path = "G:\\我的雲端硬碟\\Python\\Code\\Pycharm\\P_UI_Test\\goodsamples"
# pic_path = 'C:\Users\highh\OneDrive\Desktop\goodsamples'
#pic_path = "E:/dataset/Surface_detection/0"
(x_train, x_train_label, no1, no2) = cm.data_load(pic_path,
train_ratio=1,
resize=(width, height),
has_dir=False)
print(x_train.shape)
#print("這是打包exe中文的輸出測試")
# print(x_train_label)
#prepare test data
#pic_path = r'E:\dataset\forAE\pill\test'
pic_path = r'E:\dataset\forAE\circle\test'
(no_care, no_care_2, x_test, x_test_label) = cm.data_load(pic_path,
train_ratio=0,
resize=(width,
height),
shuffle=True,
normalize=True)
# print('x_train shape = ',x_train_2.shape)
ref:
https://zhuanlan.zhihu.com/p/32887066
'''
import tensorflow as tf
from tensorflow.python.platform import gfile
import common as cm
from keras.utils import np_utils
model_filename = "model_saver/pb_CLASF_circle__model.pb"
#data preprocess
pic_path = r'.\circle_test'
classifier_num = 2
(x_train, x_train_label, x_test, x_test_label) = cm.data_load(pic_path,
train_ratio=0,
resize=(64, 64),
has_dir=False)
print(x_test_label.shape)
x_test_label = np_utils.to_categorical(x_test_label, classifier_num)
print("test label shape = ", x_test_label.shape)
with tf.Session() as sess:
with gfile.FastGFile(model_filename, 'rb') as f:
graph_def = tf.GraphDef()
graph_def.ParseFromString(f.read())
sess.graph.as_default()
tf.import_graph_def(graph_def, name='') #導入計算圖
sess.run(tf.global_variables_initializer()) #將變數初始化