def parseTfTodataset():
"""
将tfrecors文件的数据读出转化为slim需要的datasets
:return:
"""
dataset = flowers.get_split("train", flowersDataPath)
dataProvider = provider.DatasetDataProvider(dataset=dataset,
common_queue_capacity=32,
common_queue_min=1)
image, label = dataProvider.get(["image", "label"])
return image, label
def trainModelWithMyClass():
"""
修改最终的输出分类【1001】==》【5】,有1001个分类改为5个分类
:return:
"""
fintuing = tf.Graph()
with fintuing.as_default():
tf.logging.set_verbosity(tf.logging.INFO)
# 获取数据集
dataset = flowers.get_split("train", flowersDatDir)
images, labels = loadBatch(dataset=dataset)
# 载入模型，此时模型还没有载入参数
with slim.arg_scope(V2model.inception_resnet_v2_arg_scope()):
logits, end_points = V2model.inception_resnet_v2(inputs=images,
num_classes=1001)
#增加一个全连接层，将输出1001分类改为输出5个分类
logits = slim.fully_connected(inputs=logits, num_outputs=5)
prob = tf.nn.softmax(logits)
oneHotLabels = slim.one_hot_encoding(labels=labels, num_classes=5)
# 创建损失函数
slim.losses.softmax_cross_entropy(logits=prob,
onehot_labels=oneHotLabels)
totalLoss = slim.losses.get_total_loss()
# 创建训练节点
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01)
trainOp = slim.learning.create_train_op(total_loss=totalLoss,
optimizer=optimizer)
# 准备载入模型权重的函数
modelPath = os.path.join(inceptionV2ModelPath,
inceptionV2ModelCkName)
variables = slim.get_model_variables("InceptionResnetV2")
init_fn = slim.assign_from_checkpoint_fn(model_path=modelPath,
var_list=variables)
# 正式载入模型并开始训练
with tf.Session() as sess:
init_fn(sess)
print("done!")
final_loss = slim.learning.train(
train_op=trainOp,
logdir=None,
number_of_steps=10 # 训练10波
)
def trainModel():
tf.logging.set_verbosity(tf.logging.INFO)
dataset = flowers.get_split("train", flowersDataPath)
images, labels = loadBatch(dataset=dataset)
prob = Slim_cnn(images=images, numClass=5)
prob = tf.nn.softmax(prob.net)
oneHotLabel = slim.one_hot_encoding(labels=labels, num_classes=5)
slim.losses.softmax_cross_entropy(prob, onehot_labels=oneHotLabel)
totalLoss = slim.losses.get_total_loss()
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01)
trainOp = slim.learning.create_train_op(total_loss=totalLoss,
optimizer=optimizer)
finalLoss = slim.learning.train(train_op=trainOp,
logdir=configure.saveModel,
number_of_steps=100)
def showFlowers():
"""
展示数据集中的图片
:return:
"""
with tf.Graph().as_default():
dataset = flowers.get_split("train", flowersDataPath)
dataProvider = provider.DatasetDataProvider(dataset=dataset,
common_queue_capacity=32,
common_queue_min=1)
image, label = dataProvider.get(["image", "label"])
with tf.Session() as sess:
with slim.queues.QueueRunners(sess):
for i in range(5):
npImage, npLabel = sess.run([image, label])
height, width, channel = npImage.shape
className = name = dataset.labels_to_names[npLabel]
plt.figure()
plt.imshow(npImage)
plt.title("%s,%d x %d" % (name, height, width))
plt.axis("off")
plt.show()
def trainModelWithNew():
"""
从头训练自己的模型
:return:
"""
graph = tf.Graph()
with graph.as_default():
tf.logging.set_verbosity(tf.logging.INFO)
# 获取数据集
dataset = flowers.get_split("train", flowersDatDir)
images, labels = loadBatch(dataset=dataset)
# 载入模型，此时模型还没有载入参数
with slim.arg_scope(V2model.inception_resnet_v2_arg_scope()):
logits, end_points = V2model.inception_resnet_v2(inputs=images,
num_classes=5)
prob = tf.nn.softmax(logits)
oneHotLabels = slim.one_hot_encoding(labels=labels, num_classes=5)
print("oneHotLabels:", oneHotLabels.shape)
print("prob:", prob.shape)
# 创建损失函数
slim.losses.softmax_cross_entropy(logits=prob,
onehot_labels=oneHotLabels)
totalLoss = slim.losses.get_total_loss()
# 创建训练节点
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01)
trainOp = slim.learning.create_train_op(total_loss=totalLoss,
optimizer=optimizer)
final_loss = slim.learning.train(
train_op=trainOp,
logdir=None,
number_of_steps=10 # 训练10波
)
print("trianing done and finished")