def __init__(self, data_class):
sign_handle = import_data.DealSign()
self.data_class = data_class
trainData, self.trainFlag, testData, self.testFlag = sign_handle.readFile(
self.data_class)
self.image_size = 15
self.channal = 16
self.num_classes = 5
self.batch_size = 16 if self.data_class == "person" else 32
self.lambda_val = 0.5
self.m_plus = 0.9
self.m_minus = 0.1
self.epsilon = 1e-9
self.iter_routing = 3
self.num_outputs_decode = 5
self.num_dims_decode = 16
self.trainData = self.two_dimension_graph(trainData)
self.testData = self.two_dimension_graph(testData)
self.capsnet_model = Capsnet(self.image_size, self.channal,
self.num_classes, self.lambda_val,
self.m_plus, self.m_minus, self.epsilon,
self.iter_routing,
self.num_outputs_decode,
self.num_dims_decode, self.batch_size)
self.sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True, log_device_placement=False))
self.sess.run(tf.global_variables_initializer())
def __init__(self,data_class):
sign_handle = import_data.DealSign()
self.data_class = data_class
self.trainData,self.trainFlag,self.testData,self.testFlag = sign_handle.readFile(self.data_class)
self.image_size = 14
self.channal = 16
self.num_classes = 5
self.max_gradient_norm = 10
self.learning_rate = 5e-4
self.batch_size = 16 if self.data_class == "person" else 32
self.cnn = CnnModel(self.image_size,self.channal, self.num_classes, self.learning_rate,self.max_gradient_norm)
self.sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True, log_device_placement=False))
self.sess.run(tf.global_variables_initializer())
action_num[i] = np.sum(predict_action[i * action_batch:(i + 1) *
action_batch])
print(action_num)
action_acc = action_num / action_batch
print(action_acc)
accuracy = np.sum(action_num) / self.testFlag.shape[0]
return accuracy
# variousTag = np.unique(self.trainFlag)
# rowacc = variousTag.shape[0] # rowacc表示动作数目
# accNumb = np.zeros((1, rowacc))
# for i in range(self.testData.shape[0]): # 判断测试集的分类
# k = np.matrix(self.testData[i, :])
# y = clf.predict(k)
# if y == self.testFlag[i]:
# accNumb[0, y-1] += 1
# accNumb = accNumb / self.testData.shape[0]*rowacc
# avg = np.sum(accNumb) / rowacc
# std = np.sqrt(np.sum((accNumb - avg) ** 2)/rowacc)
# print(avg,std)
if __name__ == "__main__":
data_class = "person"
# data_class = "people"
sign_handle = import_data.DealSign()
trainData, trainFlag, testData, testFlag = sign_handle.readFile(data_class)
print('train and predict')
svm_model = svmclassify(trainData, trainFlag, testData, testFlag)
accuracy = svm_model.classify()
print(accuracy)