利用Tensorflow的Slim API实现卷积神经网络

最新推荐文章于 2024-03-21 15:50:52 发布
最新推荐文章于 2024-03-21 15:50:52 发布
阅读量2.7k 收藏 6
点赞数 2
分类专栏： tensorflow 机器学习
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/carlwu/article/details/78331721
版权

这段时间在小象学院上戎雪健老师主讲《神经网络》这门课。戎老师讲得很好。但我老没时间跑老师给的代码。老师推荐尽量用TF-SLIM实现复杂结构。


下面就是以著名的mnist数据集来实例一个神经网络的实现。

import os
import numpy as np
from scipy import ndimage
import matplotlib.pyplot as plt
import tensorflow as tf
import tensorflow.contrib.slim as slim
import time
from tensorflow.examples.tutorials.mnist import input_data
# %matplotlib inline

#装载minist数据集，请把该数据集的四个文件拷贝到程序所在目录的data子目录下
mnist = input_data.read_data_sets(r'data/', one_hot=True)
trainimg   = mnist.train.images
trainlabel = mnist.train.labels
valimg     = mnist.validation.images
vallabel   = mnist.validation.labels
testimg    = mnist.test.images
testlabel  = mnist.test.labels
print ("MNIST ready")
jupyter notebook运行结果： 

Extracting Z:\CarlWu\temp\machinelearning_course\Hadoop_cn\deeplearning\DeepLearningCourseCodes-master\04_CNN_advances\data/train-images-idx3-ubyte.gz
Extracting Z:\CarlWu\temp\machinelearning_course\Hadoop_cn\deeplearning\DeepLearningCourseCodes-master\04_CNN_advances\data/train-labels-idx1-ubyte.gz
Extracting Z:\CarlWu\temp\machinelearning_course\Hadoop_cn\deeplearning\DeepLearningCourseCodes-master\04_CNN_advances\data/t10k-images-idx3-ubyte.gz
Extracting Z:\CarlWu\temp\machinelearning_course\Hadoop_cn\deeplearning\DeepLearningCourseCodes-master\04_CNN_advances\data/t10k-labels-idx1-ubyte.gz
MNIST ready

定义神经网络模型

n_input = 784
n_classes = 10
x = tf.placeholder("float", [None, n_input])
y = tf.placeholder("float", [None, n_classes])
is_training = tf.placeholder(tf.bool)

def lrelu(x, leak=0.2, name='lrelu'):
    with tf.variable_scope(name):
        f1 = 0.5 * (1 + leak)
        f2 = 0.5 * (1 - leak)
        return f1 * x + f2 * abs(x)

def CNN(inputs, is_training=True):
    x   = tf.reshape(inputs, [-1, 28, 28, 1])
    batch_norm_params = {'is_training': is_training, 'decay': 0.9
                         , 'updates_collections': None}
    init_func = tf.truncated_normal_initializer(stddev=0.01)
    net = slim.conv2d(x, 32, [5, 5], padding='SAME'
                     , activation_fn       = lrelu
                     , weights_initializer = init_func
                     , normalizer_fn       = slim.batch_norm
                     , normalizer_params   = batch_norm_params
                     , scope='conv1')
    net = slim.max_pool2d(net, [2, 2], scope='pool1')
    net = slim.conv2d(x, 64, [5, 5], padding='SAME'
                     , activation_fn       = lrelu
                     , weights_initializer = init_func
                     , normalizer_fn       = slim.batch_norm
                     , normalizer_params   = batch_norm_params
                     , scope='conv2')
    net = slim.max_pool2d(net, [2, 2], scope='pool2')
    net = slim.flatten(net, scope='flatten3')
    net = slim.fully_connected(net, 1024
                    , activation_fn       = lrelu
                    , weights_initializer = init_func
                    , normalizer_fn       = slim.batch_norm
                    , normalizer_params   = batch_norm_params
                    , scope='fc4')
    net = slim.dropout(net, keep_prob=0.7, is_training=is_training, scope='dr')  
    out = slim.fully_connected(net, n_classes
                               , activation_fn=None, normalizer_fn=None, scope='fco')
    return out
print ("神经网络准备完毕")

定义图结构

# PREDICTION
pred = CNN(x, is_training)

# LOSS AND OPTIMIZER
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
        labels=y, logits=pred))
optm = tf.train.AdamOptimizer(learning_rate=0.001).minimize(cost) 
corr = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))    
accr = tf.reduce_mean(tf.cast(corr, "float"))

# INITIALIZER
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
print ("FUNCTIONS READY")

#检查变量
print ("=================== TRAINABLE VARIABLES ===================")
t_weights = tf.trainable_variables()
var_names_list = [v.name for v in tf.trainable_variables()] 
for i in range(len(t_weights)):
    wval = sess.run(t_weights[i])
    print ("[%d/%d] [%s] / SAHPE IS %s"  
           % (i, len(t_weights), var_names_list[i], wval.shape,))

Jupyter notebook输出结果： 

=================== TRAINABLE VARIABLES ===================
[0/8] [conv1/weights:0] / SAHPE IS (5, 5, 1, 32)
[1/8] [conv1/BatchNorm/beta:0] / SAHPE IS (32,)
[2/8] [conv2/weights:0] / SAHPE IS (5, 5, 1, 64)
[3/8] [conv2/BatchNorm/beta:0] / SAHPE IS (64,)
[4/8] [fc4/weights:0] / SAHPE IS (12544, 1024)
[5/8] [fc4/BatchNorm/beta:0] / SAHPE IS (1024,)
[6/8] [fco/weights:0] / SAHPE IS (1024, 10)
[7/8] [fco/biases:0] / SAHPE IS (10,)



＃将模型存储在nets子目录下的一个目录中
savedir = "nets/cnn_mnist_modern/"
saver = tf.train.Saver(max_to_keep=100)
save_step = 4
if not os.path.exists(savedir):
    os.makedirs(savedir)
print ("SAVER READY")

#增加图片数据，训练模型
def augment_img(xs):
    out  = np.copy(xs)
    xs_r = np.reshape(xs, [-1, 28, 28])
    for i in range(xs_r.shape[0]):
        xs_img = xs_r[i, :, :]
        bg_value = 0
        # ROTATE
        angle = np.random.randint(-15, 15, 1).astype(float)
        xs_img = ndimage.rotate(xs_img, angle, reshape=False, cval=bg_value)
        # ZOOM
        rg = 0.1
        zoom_factor = np.random.uniform(1., 1.+rg)
        h, w = xs_img.shape[:2]
        zh   = int(np.round(zoom_factor * h))
        zw   = int(np.round(zoom_factor * w))
        top  = (zh - h) // 2
        left = (zw - w) // 2
        zoom_tuple = (zoom_factor,) * 2 + (1,) * (xs_img.ndim - 2)
        temp = ndimage.zoom(xs_img[top:top+zh, left:left+zw], zoom_tuple)
        trim_top  = ((temp.shape[0] - h) // 2)
        trim_left = ((temp.shape[1] - w) // 2)
        xs_img = temp[trim_top:trim_top+h, trim_left:trim_left+w]
        # SHIFT
        shift = np.random.randint(-3, 3, 2)
        xs_img = ndimage.shift(xs_img, shift, cval=bg_value)
        # RESHAPE
        xs_v = np.reshape(xs_img, [1, -1])
        out[i, :] = xs_v
    return out

在Jupyter notebook中运行模型，代码如下：

# PARAMETERS
training_epochs = 50
batch_size      = 50
display_step    = 3
val_acc         = 0
val_acc_max     = 0
# OPTIMIZE
currentTime = time.time()
for epoch in range(training_epochs):
    avg_cost = 0.
    total_batch = int(mnist.train.num_examples/batch_size)
    # ITERATION
    for i in range(total_batch):
        batch_xs, batch_ys = mnist.train.next_batch(batch_size)
        # AUGMENT DATA
        batch_xs = augment_img(batch_xs)
        feeds = {x: batch_xs, y: batch_ys, is_training: True}
        sess.run(optm, feed_dict=feeds)
        avg_cost += sess.run(cost, feed_dict=feeds)
    avg_cost = avg_cost / total_batch
    # DISPLAY
    if (epoch+1) % display_step == 0:
        print('time spent is ', (time.time()-currentTime))
        currentTime = time.time()
        print ("Epoch: %03d/%03d cost: %.9f" % (epoch+1, training_epochs, avg_cost))
        randidx = np.random.permutation(trainimg.shape[0])[:500]
        feeds = {x: trainimg[randidx], y: trainlabel[randidx], is_training: False}
        train_acc = sess.run(accr, feed_dict=feeds)
        print (" TRAIN ACCURACY: %.5f" % (train_acc))
        
        #下面这段代码计算在验证数据集上的准确度，原来的代码不能工作
        #feeds = {x: valimg, y: vallabel, is_training: False}
        #val_acc = sess.run(accr, feed_dict=feeds)

        total_batch_val=int(valimg.shape[0]/batch_size)
        print("在验证数据集上分%d批计算准确度", % total_batch_val)
        val_acc_sum = 0.0
        for j in range(total_batch_val):
            feeds = {x: valimg[j*batch_size:min((j+1)*batch_size,valimg.shape[0]-1)], 
                     y: vallabel[j*batch_size:min((j+1)*batch_size,valimg.shape[0]-1)],
                    is_training: False}
            
            val_acc = sess.run(accr, feed_dict=feeds)
            val_acc_sum = val_acc_sum + val_acc
         
        val_acc = val_acc_sum/total_batch_val
        #代码修改结束

        print (" 在验证数据集上的准确度为: %.5f" % (val_acc))
    # SAVE
    if (epoch+1) % save_step == 0:
        savename = savedir + "net-" + str(epoch) + ".ckpt"
        saver.save(sess=sess, save_path=savename)
        print (" [%s] SAVED." % (savename))
    # MAXIMUM VALIDATION ACCURACY
    if val_acc > val_acc_max:
        val_acc_max = val_acc
        best_epoch = epoch
        print ("\x1b[31m BEST EPOCH UPDATED!! [%d] \x1b[0m" % (best_epoch))
print ("OPTIMIZATION FINISHED")

在我GPU上运行了几个小时后，结果如下： 
time spent is  595.5124831199646
Epoch: 003/050 cost: 0.056146707
 TRAIN ACCURACY: 0.99200
total batch val: total_batch_val  100
 VALIDATION ACCURACY: 0.99160
 BEST EPOCH UPDATED!! [2] 
 [nets/cnn_mnist_modern/net-3.ckpt] SAVED.
time spent is  644.9777743816376
Epoch: 006/050 cost: 0.052948017
 TRAIN ACCURACY: 0.99400
total batch val: total_batch_val  100
 VALIDATION ACCURACY: 0.99020
 [nets/cnn_mnist_modern/net-7.ckpt] SAVED.
time spent is  689.395813703537
Epoch: 009/050 cost: 0.052893652
 TRAIN ACCURACY: 0.99200
total batch val: total_batch_val  100
 VALIDATION ACCURACY: 0.99180
 BEST EPOCH UPDATED!! [8] 
time spent is  598.4757721424103
...
...
Epoch: 042/050 cost: 0.037603188
 TRAIN ACCURACY: 0.99200
total batch val: total_batch_val  100
 VALIDATION ACCURACY: 0.99500
 [nets/cnn_mnist_modern/net-43.ckpt] SAVED.
time spent is  689.3062949180603
Epoch: 045/050 cost: 0.034730853
 TRAIN ACCURACY: 0.99400
total batch val: total_batch_val  100
 VALIDATION ACCURACY: 0.99520
 BEST EPOCH UPDATED!! [44] 
time spent is  616.6805007457733
Epoch: 048/050 cost: 0.035798393
 TRAIN ACCURACY: 0.99800
total batch val: total_batch_val  100
 VALIDATION ACCURACY: 0.99340
 [nets/cnn_mnist_modern/net-47.ckpt] SAVED.
OPTIMIZATION FINISHED
best_epoch = 47
restorename = savedir + "net-" + str(best_epoch) + ".ckpt"
print ("LOADING [%s]" % (restorename))
saver.restore(sess, restorename)
feeds = {x: testimg, y: testlabel, is_training: False}
test_acc = sess.run(accr, feed_dict=feeds)
print ("TEST ACCURACY: %.5f" % (test_acc))



最后在测试集上跑一下，效果也还可以：




LOADING [nets/cnn_mnist_modern/net-47.ckpt]TEST ACCURACY: 0.99120










总结下遇到的问题及解决方法：




由于我的gpu计算能力只有3.5，老是遇到OOM及ResourceExhaustedError错误：



---------------------------------------------------------------------------ResourceExhaustedError Traceback (most recent call last)C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\client\session.py
 in _do_call(self, fn, *args) 1326try:-> 1327return
 fn(*args) 1328except errors.OpErroras
 e:C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\client\session.py in_run_fn(session, feed_dict,
 fetch_list, target_list, options, run_metadata) 1305 feed_dict, fetch_list, target_list,->
 1306 status, run_metadata) 1307C:\Users\CC-Laptop\Anaconda3\lib\contextlib.py in__exit__(self, type, value, traceback) 65try:---> 66next(self.gen) 67except StopIteration:C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\framework\errors_impl.py inraise_exception_on_not_ok_status() 465 compat.as_text(pywrap_tensorflow.TF_Message(status)),--> 466 pywrap_tensorflow.TF_GetCode(status)) 467finally:ResourceExhaustedError: OOM when allocating tensor with shape[5000,28,28,64] [[Node: conv2/convolution = Conv2D[T=DT_FLOAT, data_format="NHWC", padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/gpu:0"](Reshape, conv2/weights/read)]] [[Node: Mean_1/_117 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/cpu:0", send_device="/job:localhost/replica:0/task:0/gpu:0", send_device_incarnation=1, tensor_name="edge_255_Mean_1", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/cpu:0"]()]]During handling of the above exception, another exception occurred:ResourceExhaustedError Traceback (most recent call last)<ipython-input-19-6519d8ed8769> in<module>() 31#下面这段代码计算在验证数据集上的准确度，原来的代码不能工作 32 feeds= {x: valimg, y: vallabel, is_training:False}---> 33val_acc = sess.run(accr, feed_dict=feeds) 34 35  #total_batch_val=int(valimg.shape[0]/batch_size)C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\client\session.py inrun(self, fetches, feed_dict, options, run_metadata) 893try: 894 result = self._run(None, fetches, feed_dict, options_ptr,--> 895 run_metadata_ptr) 896 if run_metadata: 897 proto_data= tf_session.TF_GetBuffer(run_metadata_ptr)C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\client\session.py in _run(self, handle, fetches, feed_dict, options, run_metadata) 1122if final_fetches or final_targetsor  (handle and feed_dict_tensor): 1123 results = self._do_run(handle, final_targets, final_fetches,-> 1124 feed_dict_tensor, options, run_metadata) 1125else: 1126 results= []C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\client\session.py in_do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata) 1319if handle is None: 1320 return self._do_call(_run_fn, self._session, feeds, fetches, targets,-> 1321 options, run_metadata) 1322 else: 1323return self._do_call(_prun_fn, self._session, handle, feeds, fetches)C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\client\session.py in_do_call(self, fn, *args) 1338except KeyError: 1339pass-> 1340raise type(e)(node_def, op, message) 1341 1342  def _extend_graph(self):ResourceExhaustedError: OOM when allocating tensor with shape[5000,28,28,64] [[Node: conv2/convolution = Conv2D[T=DT_FLOAT, data_format="NHWC", padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/gpu:0"](Reshape, conv2/weights/read)]] [[Node: Mean_1/_117 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/cpu:0", send_device="/job:localhost/replica:0/task:0/gpu:0", send_device_incarnation=1, tensor_name="edge_255_Mean_1", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/cpu:0"]()]]Caused by op 'conv2/convolution', defined at: File "C:\Users\CC-Laptop\Anaconda3\lib\runpy.py", line 184, in _run_module_as_main "__main__", mod_spec) File "C:\Users\CC-Laptop\Anaconda3\lib\runpy.py", line 85, in _run_code exec(code, run_globals) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\ipykernel\__main__.py", line 3, in <module> app.launch_new_instance() File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\traitlets\config\application.py", line 653, in launch_instance app.start() File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\ipykernel\kernelapp.py", line 474, in start ioloop.IOLoop.instance().start() File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\zmq\eventloop\ioloop.py", line 162, in start super(ZMQIOLoop, self).start() File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tornado\ioloop.py", line 887, in start handler_func(fd_obj, events) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tornado\stack_context.py", line 275, in null_wrapper return fn(*args, **kwargs) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\zmq\eventloop\zmqstream.py", line 440, in _handle_events self._handle_recv() File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\zmq\eventloop\zmqstream.py", line 472, in _handle_recv self._run_callback(callback, msg) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\zmq\eventloop\zmqstream.py", line 414, in _run_callback callback(*args, **kwargs) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tornado\stack_context.py", line 275, in null_wrapper return fn(*args, **kwargs) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\ipykernel\kernelbase.py", line 276, in dispatcher return self.dispatch_shell(stream, msg) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\ipykernel\kernelbase.py", line 228, in dispatch_shell handler(stream, idents, msg) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\ipykernel\kernelbase.py", line 390, in execute_request user_expressions, allow_stdin) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\ipykernel\ipkernel.py", line 196, in do_execute res = shell.run_cell(code, store_history=store_history, silent=silent) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\ipykernel\zmqshell.py", line 501, in run_cell return super(ZMQInteractiveShell, self).run_cell(*args, **kwargs) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py", line 2717, in run_cell interactivity=interactivity, compiler=compiler, result=result) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py", line 2821, in run_ast_nodes if self.run_code(code, result): File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\IPython\core\interactiveshell.py", line 2881, in run_code exec(code_obj, self.user_global_ns, self.user_ns) File "<ipython-input-4-0133824eed48>", line 2, in <module> pred = CNN(x, is_training) File "<ipython-input-3-d15e2c190a64>", line 30, in CNN , scope='conv2') File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\contrib\framework\python\ops\arg_scope.py", line 181, in func_with_args return func(*args, **current_args) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\contrib\layers\python\layers\layers.py", line 1027, in convolution outputs = layer.apply(inputs) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\layers\base.py", line 503, in apply return self.__call__(inputs, *args, **kwargs) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\layers\base.py", line 450, in __call__ outputs = self.call(inputs, *args, **kwargs) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\layers\convolutional.py", line 158, in call data_format=utils.convert_data_format(self.data_format, self.rank + 2)) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\ops\nn_ops.py", line 672, in convolution op=op) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\ops\nn_ops.py", line 338, in with_space_to_batch return op(input, num_spatial_dims, padding) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\ops\nn_ops.py", line 664, in op name=name) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\ops\nn_ops.py", line 131, in _non_atrous_convolution name=name) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\ops\gen_nn_ops.py", line 397, in conv2d data_format=data_format, name=name) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\framework\op_def_library.py", line 767, in apply_op op_def=op_def) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\framework\ops.py", line 2630, in create_op original_op=self._default_original_op, op_def=op_def) File "C:\Users\CC-Laptop\Anaconda3\lib\site-packages\tensorflow\python\framework\ops.py", line 1204, in __init__ self._traceback = self._graph._extract_stack() # pylint: disable=protected-accessResourceExhaustedError (see above for traceback): OOM when allocating tensor with shape[5000,28,28,64] [[Node: conv2/convolution = Conv2D[T=DT_FLOAT, data_format="NHWC", padding="SAME", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/gpu:0"](Reshape, conv2/weights/read)]] [[Node: Mean_1/_117 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/cpu:0", send_device="/job:localhost/replica:0/task:0/gpu:0", send_device_incarnation=1, tensor_name="edge_255_Mean_1", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/cpu:0"]()]]






这类问题的一般解决方法是将数据集分成更小的batch，然后再训练或测试。例如上面的问题，把下面两句话：


feeds = {x: valimg, y: vallabel, is_training: False}


val_acc = sess.run(accr, feed_dict=feeds)  


改成下面一段即可：







        total_batch_val=int(valimg.shape[0]/batch_size)
        print("在验证数据集上分%d批计算准确度", % total_batch_val)
        val_acc_sum = 0.0
        for j in range(total_batch_val):
            feeds = {x: valimg[j*batch_size:min((j+1)*batch_size,valimg.shape[0]-1)], 
                     y: vallabel[j*batch_size:min((j+1)*batch_size,valimg.shape[0]-1)],
                    is_training: False}
            
            val_acc = sess.run(accr, feed_dict=feeds)
            val_acc_sum = val_acc_sum + val_acc
         
        val_acc = val_acc_sum/total_batch_val
        #代码修改结束











                

        

        

    

  


    

      

        

            

              
                
                  carlwu
                
              
            

            

                关注
              
            

        

        

          
    
            
  • 
              
                
                
                
                
                    2
                
              
              
    点赞
    
            
    • 
            
  • 
              
                
                
                
              
              
    踩
    
            
    • 
            
  • 
              
                
                
                
                
                    6
                
              
              
    
                
    
                  
                    收藏
                  
                
    
              
    
              
    
                
    
                  觉得还不错?
                  
                    一键收藏
                  
                 
                
    
              
    
            
    • 
          
  • 
            
    
              
              
            
    
              
              
              
                    0
              
            
            
    评论
    
          
    • 
          
  • 
          
    • 
          
  • 
            
              
            
              
    
            
              
                
            
    
          
    • 
        

      

      

            

                专栏目录
          










                



	

		

			

				
					
Python TensorFlow，卷积神经网络(CNN)，手动实现卷积神经网络

				
			

			

				

					houyanhua1的专栏
				

				

					03-07
					
					3501
					
				

			

		

		

			
				
普通的深层神经网络，层与层之间通过全连接进行稠密矩阵运算，矩阵中的权重系数比较多(参数多)，影响效率且容易出现过拟合。

卷积神经网络的结构：卷积层、激活函数、池化层、全连接层。

卷积层：将原始图片通过Filter(过滤器，权重矩阵，卷积核，观察窗口)，分割出局部信息，过滤器经过多次平移取样(步长)形成一个个局部信息然后组成卷积层，卷积层中的每一个单位点对应前面层的局部信息。（卷积的意义：一个点...

			
		

	



                

              
                



	

		

			

				
					
[Tensorflow][6]图像识别与卷积神经网络

				
			

			

				

					T0UGH
				

				

					03-10
					
					1036
					
				

			

		

		

			
				
第6章 图像识别与卷积神经网络卷积神经网络(Convolutional Neural Network,CNN)
6.1 图像识别问题简介及经典数据集
图像识别问题简介

图像识别问题希望借助计算机程序来处理、分析和理解图片中的内容，使得计算机可以从图片中自动识别各种不同模式的目标和对象
卷积神经网络是图像识别问题取得突破性进展的最主要...

			
		

	



                


  
              

                


	

		

			

				
					
TensorFlow-slim之CNN实现手写数字识别

				
			

			

				

					
				

				

					04-23
					
					282
					
				

			

		

		

			
				
TensorFlow四种写法之一：Slim

代码如下：


#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @Time    : 2018/8/23 19:40
# @Author  : Seven
# @Site    : 
# @File    : CNN-slim.py
# @Software: PyCharm

# 0.导入环境
from ...

			
		

	





	

		

			

				
					
tensorflow构建网络-----TF-Slim

				
			

			

				

					wnloverforever的博客
				

				

					11-11
					
					407
					
				

			

		

		

			
				
方式1使用轻量级库TF-Slim
import tensorflow.contrib.slim as slim

1.使用slim创建变量
#常规变量，step那种
weights = slim.variable('weights',
			shape=[384,12,12,3].
			#方差为0.1
			initializer=tf.truncated_normal_initializer(...

			
		

	



		

			
		



	

		

			

				
					
用tensorflow的slim模块快速实现mnist手写体识别分类

				
			

			

				

					m0_37041325的博客
				

				

					09-11
					
					2998
					
				

			

		

		

			
				
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
from tensorflow.examples.tutorials.mnist import mnist
import tensorflow.contrib.slim as slim

mnist=input_data.read_d

			
		

	





	

		

			

				
					
【TF-Slim使用】

				
			

			

				

					图挖掘领域，新晋砖家 ☞ 未来可期，欢迎和静静一起学习交流吖 
				

				

					08-18
					
					1048
					
				

			

		

		

			
				
转载 https://blog.csdn.net/MOU_IT/article/details/82717745

            
                            
                                            1、简介
    对于tensorflow.contrib这个库，tensorflow官方对它的描述是：此目录中的任何代码未经官方支持，可能会随时更改或删除。每个目录下都有指定的所有者。它旨在包含额外功能

			
		

	





	

		

			

				
					
学习笔记(01):第二章：卷积神经网络案例-基于slim的神经网模型训练1

				
			

			

				

					qq_26491375的博客
				

				

					03-06
					
					179
					
				

			

		

		

			
				
本章介绍卷积神经网络案例代码

			
		

	





	

		

			

				
					
深度学习实践Tensorflow搭建卷积神经网络图像识别笔记

				
			

			

				

					piglake的博客
				

				

					12-19
					
					1664
					
				

			

		

		

			
				
目录DeepLearning Practise of building CNN with TensorFlow-slim概述摘要1.必要的:2.我的硬件：架构1.model：2.train:3.robots4.cnntest5.svmtest6.无关紧要的：结果1.loss曲线：2.测试集准确率：3.改进TensorFlow-slim爬虫request正则表达式CNN模型搭建what is CNN...

			
		

	





	

		

			

				
					
基于TensorFlow Object Detection API实现RetinaNet目标检测网络（附源码）

				
			

			

				

					专注AI领域
				

				

					05-26
					
					1299
					
				

			

		

		

			
				
RetinaNet是作者Tsung-Yi Lin和Kaiming He于2018年发表的论文Focal Loss for Dense Object Detection中提出的网络。RetinaNet凭借其结构精简清晰、可扩展性强、检测准确率高等特点，成为了很多目标检测算法的baseline。在模型的训练过程中，经常会遇到这样的场景：我们需要保存某个时刻训练好的模型参数，这样在对模型进行微调时，才能使用保存好的参数继续训练，而不是再随机生成参数大范围的拟合。

			
		

	





	

		

			

				
					
卷积神经网络的模型融合与集成学习

					
最新发布

				
			

			

				

					AGI通用人工智能之禅
				

				

					03-21
					
					485
					
				

			

		

		

			
				
"卷积神经网络的模型融合与集成学习"
1.背景介绍
1.1 深度学习和卷积神经网络概述
深度学习作为机器学习的一个新的研究热点,已经取得了令人瞩目的成就。卷积神经网络(Convolutional Neural 

			
		

	





	

		

			

				
					
Tensorflow 使用slim框架下的分类模型进行分类

				
			

			

				

					cugxyy6的博客
				

				

					10-27
					
					2862
					
				

			

		

		

			
				
Tensorflow的slim框架可以写出像keras一样简单的代码来实现网络结构（虽然现在keras也已经集成在tf.contrib中了），而且models/slim提供了类似之前说过的object detection接口类似的image classification接口，可以很方便的进行fine-tuning利用自己的数据集训练自己所需的模型。官方文档提供了比较详细的从数据准备，预训练模型的mo

			
		

	





	

		

			

				
					
第二章：卷积神经网络案例

				
			

			

				

					01-29
				

			

		

		

			
				
本章介绍卷积神经网络案例代码

			
		

	





	

		

			

				
					
TensorFlow之Slim

				
			

			

				

					健身房
				

				

					05-28
					
					1038
					
				

			

		

		

			
				
  TF-Slim是Tensorflow中一个轻量级的库，用于定义、训练和评估复杂的模型。TF-Slim中的组件可以与Tensorflow中原生的函数一起使用，与其他的框架(比如tf.contrib.learn)也可以一起使用。使用方法如下：
import tensorflow.contrib.slim as slim

  TF-Slim可以使建立、训练和评估神经网络更加简单，具有如下特点：

允许用户通过减少模板代码使得模型更加简洁。这个可以通过使用argument scoping和大量的高层layer

			
		

	





	

		

			

				
					
第二十四节，TensorFlow下slim库函数的使用以及使用VGG网络进行预训练、迁移学习(附代码)...

				
			

			

				

					weixin_30670965的博客
				

				

					06-08
					
					502
					
				

			

		

		

			
				
在介绍这一节之前，需要你对slim模型库有一些基本了解，具体可以参考第二十二节，TensorFlow中的图片分类模型库slim的使用、数据集处理，这一节我们会详细介绍slim模型库下面的一些函数的使用。
一 简介
slim被放在tensorflow.contrib这个库下面，导入的方法如下：

import tensorflow.contrib.slim as slim

这样我们就可...

			
		

	





	

		

			

				
					
简单的一个两层的一维卷积网络

				
			

			

				

					2301_77540108的博客
				

				

					09-01
					
					215
					
				

			

		

		

			
				
这个模型包括两层卷积层，每一层都是由一个卷积层和一个最大池化层构成的。输入层是一个一维张量，形状为(batch_size, input_length, input_channels)，输出层是一个softmax分类器。其中x_train和y_train是训练数据和标签，x_val和y_val是验证数据和标签。

			
		

	





	

		

			

				
					
使用 TF-Slim 设计复杂网络

				
			

			

				

					风吴痕的博客
				

				

					09-27
					
					2036
					
				

			

		

		

			
				
原文地址：https://mp.weixin.qq.com/s/gUDJuS1HKkxUZaCHHUj7awTF-Slim 是一个新的轻量的 TensorFlow 高级 API（tensorflow.contrib.slim），用于定义、训练、评估复杂模型。参考材料【1】包含几个广泛使用的 CNN 图像分类模型代码，可以利用提供的脚本从头训练一个模型，或者使用预训练的网络权值微调。TF-Slim 附

			
		

	





	

		

			

				
					
TF：卷积层的实现方式

				
			

			

				

					sinoai
				

				

					04-15
					
					496
					
				

			

		

		

			
				
卷积层的实现方式：

tf.nn.conv2d：最基础的用法，需要自定义权重和偏置。用法示例：
	
def create_weights(shape):
    return tf.Variable(tf.truncated_normal(shape,stddev=0.1))

def create_bias(num_filters):
    return tf.Variable(tf.cons...

			
		

	





	

		

			

				
					
Python pandas秘籍第一、二、三章笔记

				
			

			

				

					程志伟的博客
				

				

					08-04
					
					703
					
				

			

		

		

			
				
Python 3.7.6 (default, Jan 8 2020, 20:23:39) [MSC v.1916 64 bit (AMD64)]
Type "copyright", "credits" or "license" for more information.

IPython 7.12.0 -- An enhanced Interactive Python.

import pandas as pd



第一章



1.1 从 CSV 文件中读取数据
broken_df = pd.re...

			
		

	





	

		

			

				
					
TF-slim快速搭建cnn

				
			

			

				

					风吴痕的博客
				

				

					11-01
					
					2997
					
				

			

		

		

			
				
参考： 
1、https://github.com/tensorflow/models/blob/master/research/slim/nets/cifarnet.py 
2、https://github.com/tensorflow/models/blob/master/research/slim模型：参考slim里的cifarnet.py快速搭建CNN 
使用数据：mnist#!/usr/b

			
		

	





	

		

			

				
					
TensorFlow-Slim：轻量级深度学习库

				
			

			

				

					
				

			

		

		

			
				
"TensorFlow-Slim 是一个轻量级库，专用于在TensorFlow中高效地定义、训练和评估复杂的神经网络模型。它允许用户通过消除样板代码来更紧凑地定义模型，通过参数作用域和高级层及变量提高代码可读性和可维护性。此外...

			
		

	



              




          




        



    





    



      

      

        
      

      





	

		评论	
	

	
	




  

    

      添加红包
      
    

    

      

        
        

          
          
        

        
请填写红包祝福语或标题

      

      

        
        

          
          个
        

        
红包个数最小为10个

      

      

        
        

          
          元
        

        
红包金额最低5元

      

      

        
        

          当前余额3.43元
          前往充值 >
        

      

      

        

          需支付：10.00元
        

        
        
      

    

  





  

    
    
  

  

    
    
  








  

    

      

        

          

            
            
成就一亿技术人!

          

          

        

        

          

          

            领取后你会自动成为博主和红包主的粉丝
            规则
          

        

      

      

        

          

            
              
            
            

              
hope_wisdom
 发出的红包
            

          

        

        

          

          

          

        

      

    

    

  



      
      

      
实付元

      
使用余额支付

      

        

          

            
            点击重新获取
          

        

        
扫码支付

      

      

        
          
            
          
          
        
      

      

        
        钱包余额
          0
          

            
            

              

                
抵扣说明：

                
 1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
 2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

              

            

          

      

      余额充值