迁移学习InceptionV3

tf_model_path = '/Users/akarizo/Desktop/inception_dec_2015/tensorflow_inception_graph.pb'
with open(tf_model_path, 'rb') as f:
    serialized = f.read()
tf.reset_default_graph()
original_gdef = tf.GraphDef()
original_gdef.ParseFromString(serialized)
##2 可以在这里查看到全部的图信息(前提是你有转换成功)
with tf.Graph().as_default() as g:
    tf.import_graph_def(original_gdef, name='')
    ops = g.get_operations()
    try:
        for i in range(10000):
            print('op id {} : op name: {}, op type: "{}"'.format(str(i),ops[i].name, ops[i].type))
    except:
        pass

def colab_to_drive(file_path, save_name):
  '''
   ** 从Colab到drive ** 
  '''
  auth.authenticate_user()
  drive_service = build('drive', 'v3')
  file_metadata = {
    'name': save_name,
    'mimeType': 'text/plain'
  }
  media = MediaFileUpload(file_path, 
                          mimetype='text/plain',
                          resumable=True)
  drive_service.files().create(body=file_metadata,
                               media_body=media,
                               fields='id').execute()
  print("传输完成.")

def upload_demol():
  '''
   ** 封装上传 **
  '''
  time_str = datetime.now().strftime('_%d_%H_%M_%S')
  name = 'RSM_'+time_str+'.gz'
  tar_cmd = 'tar zcvf {a} {b}'.format(a=name, b=MODEL_SAVE_DIR)
  rm_cmd  = 'rm -rf {}'.format(name)
  get_ipython().system(tar_cmd)
  colab_to_drive(name, name)
  get_ipython().system(rm_cmd)
  print("上传成功,包名为",name)
# upload_demol()

def get_test_bottlenecks(sess,image_lists,n_class,jpeg_data_tensor,bottleneck_tensor):
    '''
    获取全部的测试数据,计算输出
    :param sess: 
    :param image_lists: 
    :param n_class: 
    :param jpeg_data_tensor: 
    :param bottleneck_tensor: 
    :return:                   瓶颈输出 & label
    '''
    bottlenecks  = []
    ground_truths = []
    label_name_list = list(image_lists.keys())
    #  ** 原先方法会报错:
    #     for label_index,label_name in enumerate(image_lists[label_name_list]):
    #  **修改如下(3行):
    label_index = random.randrange(n_class)  # 标签索引随机生成
    label_name = label_name_list[label_index]
    label_index= label_name_list.index(label_name)
   category = 'testing'
    for index, unused_base_name in enumerate(image_lists[label_name][category]): # 索引, {文件名}
        bottleneck = get_or_create_bottleneck(
            sess, image_lists, label_name, index,
            category, jpeg_data_tensor, bottleneck_tensor)
        ground_truth = np.zeros(n_class, dtype=np.float32)
        ground_truth[label_index] = 1.0
        bottlenecks.append(bottleneck)
        ground_truths.append(ground_truth)
    return bottlenecks, ground_truths

def run_bottleneck_on_images(sess,image_data,jpeg_data_tensor,bottleneck_tensor):
    '''
    使用加载的训练好的Inception-v3模型处理一张图片，得到这个图片的特征向量。
    :param sess:              会话句柄
    :param image_data:        图片文件句柄
    :param jpeg_data_tensor:  输入张量句柄
    :param bottleneck_tensor: 瓶颈张量句柄
    :return:                  瓶颈张量值
    '''
    bottleneck_values = sess.run(bottleneck_tensor,feed_dict={jpeg_data_tensor:image_data})
    print(bottleneck_values.shape) # 从这里也能知道它返回的shape
    bottleneck_values = np.squeeze(bottleneck_values)
    return bottleneck_values

bottleneck_input   = tf.placeholder(tf.float32, [None,BOTTLENECK_TENSOR_SIZE], name='BottleneckInputPlaceholder')
ground_truth_input = tf.placeholder(tf.float32, [None,n_class], name='GroundTruthInput')
# 上面的操作后,我们定义好自己的全连接层,接入即可
with tf.name_scope('final_train_ops'):
    # Weight-bias初始化
    Weights = tf.Variable(tf.truncated_normal([BOTTLENECK_TENSOR_SIZE,n_class],stddev=0.001)) 
    biases  = tf.Variable(tf.zeros([n_class]))
    logits  = tf.matmul(bottleneck_input,Weights) + biases
    final_tensor = tf.nn.softmax(logits, name='softMax_last')
tf.add_to_collection(name='final_tensor', value=final_tensor)

cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=logits,labels=ground_truth_input))
train_step    = tf.train.RMSPropOptimizer(LEARNING_RATE).minimize(cross_entropy)
# 正确率
with tf.name_scope('evaluation'):
    correct_prediction = tf.equal(tf.argmax(final_tensor,1,name='argmax_softMax_last'),tf.argmax(ground_truth_input,1))
    evaluation_step    = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    saver = tf.train.Saver()
    if keepon:
        print('继续加载模型:')
        saver.restore(sess, tf.train.latest_checkpoint(MODEL_SAVE_DIR))

if i % 1000 == 0:
                    print("进行存储上传")
                    tf.train.write_graph(sess.graph, MODEL_SAVE_DIR, 'model.pbtxt')
                    upload_demol()
                saver.save(sess, MODEL_SAVEPATH, global_step=50) # 50次保存一次

import tensorflow as tf
import numpy as np

# ***********************define***********************
MODEL_PATH   = '/Users/akarizo/TEMP/SaveData/model/101_class_model.ckpt-50.meta'
MODEL_DIR    =  '/Users/akarizo/TEMP/SaveData/model'
image_path   = '/Users/akarizo/Desktop/pisa.jpeg'
CATO_PATH    = '/Users/akarizo/TEMP/SaveData/model/catogory.info'
catogo_list  = open(CATO_PATH, 'r').read().split('|')
# ***********************LOADGraph***********************
sess = tf.Session()
saver = tf.train.import_meta_graph(MODEL_PATH)
saver.restore(sess, tf.train.latest_checkpoint(MODEL_DIR))
graph = tf.get_default_graph()
# ***********************Define-Node***********************
inpiut_x     = graph.get_tensor_by_name('import/DecodeJpeg/contents:0')
poo3         = graph.get_tensor_by_name('import/pool_3/_reshape:0')
change_input = graph.get_tensor_by_name('BottleneckInputPlaceholder:0')
predict      = graph.get_tensor_by_name('final_train_ops/softMax_last:0')
print('加载模型成功')
def classification_photo():
    # step 打开图片切换格式
    image = tf.gfile.FastGFile(image_path, 'rb').read()        # inceptionV3不需要转换图片, 它自己换处理图片
    poo3_frist = sess.run(poo3, feed_dict={inpiut_x: image})   # 按照模型的顺序要, 先喂给它图片, 然后图片提取到瓶颈的tensor
    result = sess.run(predict, feed_dict={change_input:poo3_frist}) # 瓶颈的tensor再转入input传入, 得到我们最后的predict
    ar = np.array(result).flatten().tolist()
    inde = ar.index(max(ar))
    print("it's a {} photo".format(catogo_list[inde]))
classification_photo()