还在纠结垃圾分类问题？带你用Python感受ImageNet冠军模型SENet的强大

AI科技大本营

发布于 2019-07-11 14:49:15

7760

发布于 2019-07-11 14:49:15

文章被收录于专栏：AI科技大本营的专栏

作者 | beyondma

转载自CSDN博客

本月1日起，上海正式开始了“史上最严“垃圾分类的规定，扔错垃圾最高可罚200元。全国其它46个城市也要陆续步入垃圾分类新时代。各种被垃圾分类逼疯的段子在社交媒体上层出不穷。

其实从人工智能的角度看垃圾分类就是图像处理中图像分类任务的一种应用，而这在2012年以来的ImageNet图像分类任务的评比中，SENet模型以top-5测试集回归2.25%错误率的成绩可谓是技压群雄，堪称目前最强的图像分类器。

笔者刚刚还到SENet的创造者momenta公司的网站上看了一下，他们最新的方向已经是3D物体识别和标定了，效果如下：

可以说他们提出的SENet进行垃圾图像处理是完全没问题的。

Senet简介

Senet的是由momenta和牛津大学共同提出的一种基于挤压(squeeze）和激励（Excitation）的模型，每个模块通过“挤压”操作嵌入来自全局感受野的信息，并且通过“激励”操作选择性地诱导响应增强。我们可以看到历年的ImageNet冠军基本都是在使用加大模型数量和连接数量的方式来提高精度，而Senet在这种”大力出奇迹”的潮流下明显是一股清流。其论文地址如下：http://openaccess.thecvf.com/content_cvpr_2018/papers/Hu_Squeeze-and-Excitation_Networks_CVPR_2018_paper.pdf

其具体原理说明如下：

Sequeeze：对 C×H×W 进行 global average pooling，得到 1×1×C 大小的特征图，这个特征图可以理解为具有全局感受野。翻译论文原文来说：将每个二维的特征通道变成一个实数，这个实数某种程度上具有全局的感受野，并且输出的维度和输入的特征通道数相匹配。它表征着在特征通道上响应的全局分布，而且使得靠近输入的层也可以获得全局的感受野。

Excitation ：使用一个全连接神经网络，对 Sequeeze 之后的结果做一个非线性变换。它的机制一个类似于循环神经网络中的门。通过参数 w 来为每个特征通道生成权重，其中参数 w 被学习用来显式地建模特征通道间的相关性。

特征重标定：使用 Excitation 得到的结果作为权重，乘到输入特征上。将Excitation输出的权重可以认为是特征通道的重要性反应，逐通道加权到放到先前的特征上，完成对原始特征的重标定。

其模型架构如下：

SENet 构造非常简单，而且很容易被部署，不需要引入新的函数或者层。其caffe模型可以通过百度下载（https://pan.baidu.com/s/1o7HdfAE?errno=0&errmsg=Auth%20Login%20Sucess&&bduss=&ssnerror=0&traceid=）

Senet的运用

如果读者布署有caffe那么直接下载刚刚的模型直接load进来就可以使用了。如果没有装caffe而装了tensorflow也没关系，我们刚刚说了SENet没有引入新的函数和层，很方便用tensorflow实现。

下载图像集：经笔者各方查找发现了这个数据集，虽然不大也没有发挥出SENet的优势，不过也方便使用：

https://raw.githubusercontent.com/garythung/trashnet/master/data/dataset-resized.zip

建立SENet模型：使用tensorflow建立的模型在github上也有开源项目了，网址如下：https://github.com/taki0112/SENet-Tensorflow，只是他使用的是Cifar10数据集，不过这也没关系，只需要在gitclone以下将其cifar10.py中的prepare_data函数做如下修改即可。

 1def prepare_data():
 2    print("======Loading data======")
 3    download_data()
 4    data_dir = 'e:/test/'
 5    #data_dir = './cifar-10-batches-py'#改为你的文件侠
 6    image_dim = image_size * image_size * img_channels
 7    #meta = unpickle(data_dir + '/batches.meta')#本数据集不使用meta文件分类，故需要修改
 8    label_names = ['cardboard','glass','metal','trash','paper','plastic']
 9    label_count = len(label_names)
10    #train_files = ['data_batch_%d' % d for d in range(1, 6)]
11    train_files = [data_dir+s for s in label_names]#改为
12    train_data, train_labels = load_data(train_files, data_dir, label_count)
13    test_data, test_labels = load_data(['test_batch'], data_dir, label_count)
14
15    print("Train data:", np.shape(train_data), np.shape(train_labels))
16    print("Test data :", np.shape(test_data), np.shape(test_labels))
17    print("======Load finished======")
18
19    print("======Shuffling data======")
20    indices = np.random.permutation(len(train_data))
21    train_data = train_data[indices]
22    train_labels = train_labels[indices]
23    print("======Prepare Finished======")
24
25    return train_data, train_labels, test_data, test_labels

其最主要的建模代码如下，其主要工作就是将SENet的模型结构实现一下即可：

  1import tensorflow as tf
  2from tflearn.layers.conv import global_avg_pool
  3from tensorflow.contrib.layers import batch_norm, flatten
  4from tensorflow.contrib.framework import arg_scope
  5from cifar10 import *
  6import numpy as np
  7
  8weight_decay = 0.0005
  9momentum = 0.9
 10
 11init_learning_rate = 0.1
 12
 13reduction_ratio = 4
 14
 15batch_size = 128
 16iteration = 391
 17# 128 * 391 ~ 50,000
 18
 19test_iteration = 10
 20
 21total_epochs = 100
 22
 23def conv_layer(input, filter, kernel, stride=1, padding='SAME', layer_name="conv", activation=True):
 24    with tf.name_scope(layer_name):
 25        network = tf.layers.conv2d(inputs=input, use_bias=True, filters=filter, kernel_size=kernel, strides=stride, padding=padding)
 26        if activation :
 27            network = Relu(network)
 28        return network
 29
 30def Fully_connected(x, units=class_num, layer_name='fully_connected') :
 31    with tf.name_scope(layer_name) :
 32        return tf.layers.dense(inputs=x, use_bias=True, units=units)
 33
 34def Relu(x):
 35    return tf.nn.relu(x)
 36
 37def Sigmoid(x):
 38    return tf.nn.sigmoid(x)
 39
 40def Global_Average_Pooling(x):
 41    return global_avg_pool(x, name='Global_avg_pooling')
 42
 43def Max_pooling(x, pool_size=[3,3], stride=2, padding='VALID') :
 44    return tf.layers.max_pooling2d(inputs=x, pool_size=pool_size, strides=stride, padding=padding)
 45
 46def Batch_Normalization(x, training, scope):
 47    with arg_scope([batch_norm],
 48                   scope=scope,
 49                   updates_collections=None,
 50                   decay=0.9,
 51                   center=True,
 52                   scale=True,
 53                   zero_debias_moving_mean=True) :
 54        return tf.cond(training,
 55                       lambda : batch_norm(inputs=x, is_training=training, reuse=None),
 56                       lambda : batch_norm(inputs=x, is_training=training, reuse=True))
 57
 58def Concatenation(layers) :
 59    return tf.concat(layers, axis=3)
 60
 61def Dropout(x, rate, training) :
 62    return tf.layers.dropout(inputs=x, rate=rate, training=training)
 63
 64def Evaluate(sess):
 65    test_acc = 0.0
 66    test_loss = 0.0
 67    test_pre_index = 0
 68    add = 1000
 69
 70    for it in range(test_iteration):
 71        test_batch_x = test_x[test_pre_index: test_pre_index + add]
 72        test_batch_y = test_y[test_pre_index: test_pre_index + add]
 73        test_pre_index = test_pre_index + add
 74
 75        test_feed_dict = {
 76            x: test_batch_x,
 77            label: test_batch_y,
 78            learning_rate: epoch_learning_rate,
 79            training_flag: False
 80        }
 81
 82        loss_, acc_ = sess.run([cost, accuracy], feed_dict=test_feed_dict)
 83
 84        test_loss += loss_
 85        test_acc += acc_
 86
 87    test_loss /= test_iteration # average loss
 88    test_acc /= test_iteration # average accuracy
 89
 90    summary = tf.Summary(value=[tf.Summary.Value(tag='test_loss', simple_value=test_loss),
 91                                tf.Summary.Value(tag='test_accuracy', simple_value=test_acc)])
 92
 93    return test_acc, test_loss, summary
 94
 95class SE_Inception_resnet_v2():
 96    def __init__(self, x, training):
 97        self.training = training
 98        self.model = self.Build_SEnet(x)
 99
100    def Stem(self, x, scope):
101        with tf.name_scope(scope) :
102            x = conv_layer(x, filter=32, kernel=[3,3], stride=2, padding='VALID', layer_name=scope+'_conv1')
103            x = conv_layer(x, filter=32, kernel=[3,3], padding='VALID', layer_name=scope+'_conv2')
104            block_1 = conv_layer(x, filter=64, kernel=[3,3], layer_name=scope+'_conv3')
105
106            split_max_x = Max_pooling(block_1)
107            split_conv_x = conv_layer(block_1, filter=96, kernel=[3,3], stride=2, padding='VALID', layer_name=scope+'_split_conv1')
108            x = Concatenation([split_max_x,split_conv_x])
109
110            split_conv_x1 = conv_layer(x, filter=64, kernel=[1,1], layer_name=scope+'_split_conv2')
111            split_conv_x1 = conv_layer(split_conv_x1, filter=96, kernel=[3,3], padding='VALID', layer_name=scope+'_split_conv3')
112
113            split_conv_x2 = conv_layer(x, filter=64, kernel=[1,1], layer_name=scope+'_split_conv4')
114            split_conv_x2 = conv_layer(split_conv_x2, filter=64, kernel=[7,1], layer_name=scope+'_split_conv5')
115            split_conv_x2 = conv_layer(split_conv_x2, filter=64, kernel=[1,7], layer_name=scope+'_split_conv6')
116            split_conv_x2 = conv_layer(split_conv_x2, filter=96, kernel=[3,3], padding='VALID', layer_name=scope+'_split_conv7')
117
118            x = Concatenation([split_conv_x1,split_conv_x2])
119
120            split_conv_x = conv_layer(x, filter=192, kernel=[3,3], stride=2, padding='VALID', layer_name=scope+'_split_conv8')
121            split_max_x = Max_pooling(x)
122
123            x = Concatenation([split_conv_x, split_max_x])
124
125            x = Batch_Normalization(x, training=self.training, scope=scope+'_batch1')
126            x = Relu(x)
127
128            return x
129
130    def Inception_resnet_A(self, x, scope):
131        with tf.name_scope(scope) :
132            init = x
133
134            split_conv_x1 = conv_layer(x, filter=32, kernel=[1,1], layer_name=scope+'_split_conv1')
135
136            split_conv_x2 = conv_layer(x, filter=32, kernel=[1,1], layer_name=scope+'_split_conv2')
137            split_conv_x2 = conv_layer(split_conv_x2, filter=32, kernel=[3,3], layer_name=scope+'_split_conv3')
138
139            split_conv_x3 = conv_layer(x, filter=32, kernel=[1,1], layer_name=scope+'_split_conv4')
140            split_conv_x3 = conv_layer(split_conv_x3, filter=48, kernel=[3,3], layer_name=scope+'_split_conv5')
141            split_conv_x3 = conv_layer(split_conv_x3, filter=64, kernel=[3,3], layer_name=scope+'_split_conv6')
142
143            x = Concatenation([split_conv_x1,split_conv_x2,split_conv_x3])
144            x = conv_layer(x, filter=384, kernel=[1,1], layer_name=scope+'_final_conv1', activation=False)
145
146            x = x*0.1
147            x = init + x
148
149            x = Batch_Normalization(x, training=self.training, scope=scope+'_batch1')
150            x = Relu(x)
151
152            return x
153
154    def Inception_resnet_B(self, x, scope):
155        with tf.name_scope(scope) :
156            init = x
157
158            split_conv_x1 = conv_layer(x, filter=192, kernel=[1,1], layer_name=scope+'_split_conv1')
159
160            split_conv_x2 = conv_layer(x, filter=128, kernel=[1,1], layer_name=scope+'_split_conv2')
161            split_conv_x2 = conv_layer(split_conv_x2, filter=160, kernel=[1,7], layer_name=scope+'_split_conv3')
162            split_conv_x2 = conv_layer(split_conv_x2, filter=192, kernel=[7,1], layer_name=scope+'_split_conv4')
163
164            x = Concatenation([split_conv_x1, split_conv_x2])
165            x = conv_layer(x, filter=1152, kernel=[1,1], layer_name=scope+'_final_conv1', activation=False)
166            # 1154
167            x = x * 0.1
168            x = init + x
169
170            x = Batch_Normalization(x, training=self.training, scope=scope+'_batch1')
171            x = Relu(x)
172
173            return x
174
175    def Inception_resnet_C(self, x, scope):
176        with tf.name_scope(scope) :
177            init = x
178
179            split_conv_x1 = conv_layer(x, filter=192, kernel=[1,1], layer_name=scope+'_split_conv1')
180
181            split_conv_x2 = conv_layer(x, filter=192, kernel=[1, 1], layer_name=scope + '_split_conv2')
182            split_conv_x2 = conv_layer(split_conv_x2, filter=224, kernel=[1, 3], layer_name=scope + '_split_conv3')
183            split_conv_x2 = conv_layer(split_conv_x2, filter=256, kernel=[3, 1], layer_name=scope + '_split_conv4')
184
185            x = Concatenation([split_conv_x1,split_conv_x2])
186            x = conv_layer(x, filter=2144, kernel=[1,1], layer_name=scope+'_final_conv2', activation=False)
187            # 2048
188            x = x * 0.1
189            x = init + x
190
191            x = Batch_Normalization(x, training=self.training, scope=scope+'_batch1')
192            x = Relu(x)
193
194            return x
195
196    def Reduction_A(self, x, scope):
197        with tf.name_scope(scope) :
198            k = 256
199            l = 256
200            m = 384
201            n = 384
202
203            split_max_x = Max_pooling(x)
204
205            split_conv_x1 = conv_layer(x, filter=n, kernel=[3,3], stride=2, padding='VALID', layer_name=scope+'_split_conv1')
206
207            split_conv_x2 = conv_layer(x, filter=k, kernel=[1,1], layer_name=scope+'_split_conv2')
208            split_conv_x2 = conv_layer(split_conv_x2, filter=l, kernel=[3,3], layer_name=scope+'_split_conv3')
209            split_conv_x2 = conv_layer(split_conv_x2, filter=m, kernel=[3,3], stride=2, padding='VALID', layer_name=scope+'_split_conv4')
210
211            x = Concatenation([split_max_x, split_conv_x1, split_conv_x2])
212
213            x = Batch_Normalization(x, training=self.training, scope=scope+'_batch1')
214            x = Relu(x)
215
216            return x
217
218    def Reduction_B(self, x, scope):
219        with tf.name_scope(scope) :
220            split_max_x = Max_pooling(x)
221
222            split_conv_x1 = conv_layer(x, filter=256, kernel=[1,1], layer_name=scope+'_split_conv1')
223            split_conv_x1 = conv_layer(split_conv_x1, filter=384, kernel=[3,3], stride=2, padding='VALID', layer_name=scope+'_split_conv2')
224
225            split_conv_x2 = conv_layer(x, filter=256, kernel=[1,1], layer_name=scope+'_split_conv3')
226            split_conv_x2 = conv_layer(split_conv_x2, filter=288, kernel=[3,3], stride=2, padding='VALID', layer_name=scope+'_split_conv4')
227
228            split_conv_x3 = conv_layer(x, filter=256, kernel=[1,1], layer_name=scope+'_split_conv5')
229            split_conv_x3 = conv_layer(split_conv_x3, filter=288, kernel=[3,3], layer_name=scope+'_split_conv6')
230            split_conv_x3 = conv_layer(split_conv_x3, filter=320, kernel=[3,3], stride=2, padding='VALID', layer_name=scope+'_split_conv7')
231
232            x = Concatenation([split_max_x, split_conv_x1, split_conv_x2, split_conv_x3])
233
234            x = Batch_Normalization(x, training=self.training, scope=scope+'_batch1')
235            x = Relu(x)
236
237            return x
238
239    def Squeeze_excitation_layer(self, input_x, out_dim, ratio, layer_name):
240        with tf.name_scope(layer_name) :
241
242
243            squeeze = Global_Average_Pooling(input_x)
244
245            excitation = Fully_connected(squeeze, units=out_dim / ratio, layer_name=layer_name+'_fully_connected1')
246            excitation = Relu(excitation)
247            excitation = Fully_connected(excitation, units=out_dim, layer_name=layer_name+'_fully_connected2')
248            excitation = Sigmoid(excitation)
249
250            excitation = tf.reshape(excitation, [-1,1,1,out_dim])
251            scale = input_x * excitation
252
253            return scale
254
255    def Build_SEnet(self, input_x):
256        input_x = tf.pad(input_x, [[0, 0], [32, 32], [32, 32], [0, 0]])
257        # size 32 -> 96
258        print(np.shape(input_x))
259        # only cifar10 architecture
260
261        x = self.Stem(input_x, scope='stem')
262
263        for i in range(5) :
264            x = self.Inception_resnet_A(x, scope='Inception_A'+str(i))
265            channel = int(np.shape(x)[-1])
266            x = self.Squeeze_excitation_layer(x, out_dim=channel, ratio=reduction_ratio, layer_name='SE_A'+str(i))
267
268        x = self.Reduction_A(x, scope='Reduction_A')
269
270        channel = int(np.shape(x)[-1])
271        x = self.Squeeze_excitation_layer(x, out_dim=channel, ratio=reduction_ratio, layer_name='SE_A')
272
273        for i in range(10)  :
274            x = self.Inception_resnet_B(x, scope='Inception_B'+str(i))
275            channel = int(np.shape(x)[-1])
276            x = self.Squeeze_excitation_layer(x, out_dim=channel, ratio=reduction_ratio, layer_name='SE_B'+str(i))
277
278        x = self.Reduction_B(x, scope='Reduction_B')
279
280        channel = int(np.shape(x)[-1])
281        x = self.Squeeze_excitation_layer(x, out_dim=channel, ratio=reduction_ratio, layer_name='SE_B')
282
283        for i in range(5) :
284            x = self.Inception_resnet_C(x, scope='Inception_C'+str(i))
285            channel = int(np.shape(x)[-1])
286            x = self.Squeeze_excitation_layer(x, out_dim=channel, ratio=reduction_ratio, layer_name='SE_C'+str(i))
287
288
289        # channel = int(np.shape(x)[-1])
290        # x = self.Squeeze_excitation_layer(x, out_dim=channel, ratio=reduction_ratio, layer_name='SE_C')
291
292        x = Global_Average_Pooling(x)
293        x = Dropout(x, rate=0.2, training=self.training)
294        x = flatten(x)
295
296        x = Fully_connected(x, layer_name='final_fully_connected')
297        return x
298
299
300train_x, train_y, test_x, test_y = prepare_data()
301train_x, test_x = color_preprocessing(train_x, test_x)
302
303
304# image_size = 32, img_channels = 3, class_num = 10 in cifar10
305x = tf.placeholder(tf.float32, shape=[None, image_size, image_size, img_channels])
306label = tf.placeholder(tf.float32, shape=[None, class_num])
307
308training_flag = tf.placeholder(tf.bool)
309
310
311learning_rate = tf.placeholder(tf.float32, name='learning_rate')
312
313logits = SE_Inception_resnet_v2(x, training=training_flag).model
314cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=label, logits=logits))
315
316l2_loss = tf.add_n([tf.nn.l2_loss(var) for var in tf.trainable_variables()])
317optimizer = tf.train.MomentumOptimizer(learning_rate=learning_rate, momentum=momentum, use_nesterov=True)
318train = optimizer.minimize(cost + l2_loss * weight_decay)
319
320correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(label, 1))
321accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
322
323saver = tf.train.Saver(tf.global_variables())
324
325with tf.Session() as sess:
326    ckpt = tf.train.get_checkpoint_state('./model')
327    if ckpt and tf.train.checkpoint_exists(ckpt.model_checkpoint_path):
328        saver.restore(sess, ckpt.model_checkpoint_path)
329    else:
330        sess.run(tf.global_variables_initializer())
331
332    summary_writer = tf.summary.FileWriter('./logs', sess.graph)
333
334    epoch_learning_rate = init_learning_rate
335    for epoch in range(1, total_epochs + 1):
336        if epoch % 30 == 0 :
337            epoch_learning_rate = epoch_learning_rate / 10
338
339        pre_index = 0
340        train_acc = 0.0
341        train_loss = 0.0
342
343        for step in range(1, iteration + 1):
344            if pre_index + batch_size < 50000:
345                batch_x = train_x[pre_index: pre_index + batch_size]
346                batch_y = train_y[pre_index: pre_index + batch_size]
347            else:
348                batch_x = train_x[pre_index:]
349                batch_y = train_y[pre_index:]
350
351            batch_x = data_augmentation(batch_x)
352
353            train_feed_dict = {
354                x: batch_x,
355                label: batch_y,
356                learning_rate: epoch_learning_rate,
357                training_flag: True
358            }
359
360            _, batch_loss = sess.run([train, cost], feed_dict=train_feed_dict)
361            batch_acc = accuracy.eval(feed_dict=train_feed_dict)
362
363            train_loss += batch_loss
364            train_acc += batch_acc
365            pre_index += batch_size
366
367
368        train_loss /= iteration # average loss
369        train_acc /= iteration # average accuracy
370
371        train_summary = tf.Summary(value=[tf.Summary.Value(tag='train_loss', simple_value=train_loss),
372                                          tf.Summary.Value(tag='train_accuracy', simple_value=train_acc)])
373
374        test_acc, test_loss, test_summary = Evaluate(sess)
375
376        summary_writer.add_summary(summary=train_summary, global_step=epoch)
377        summary_writer.add_summary(summary=test_summary, global_step=epoch)
378        summary_writer.flush()
379
380        line = "epoch: %d/%d, train_loss: %.4f, train_acc: %.4f, test_loss: %.4f, test_acc: %.4f \n" % (
381            epoch, total_epochs, train_loss, train_acc, test_loss, test_acc)
382        print(line)
383
384        with open('logs.txt', 'a') as f:
385            f.write(line)
386
387        saver.save(sess=sess, save_path='./model/Inception_resnet_v2.ckpt')

其实使用SENet做垃圾分类真是大才小用了，不过大家也可以感受一下他的实力强大。

原文链接：

https://blog.csdn.net/BEYONDMA/article/details/94888771

（*本文为 AI科技大本营转载文章，转载请联系原作者）

本文参与腾讯云自媒体同步曝光计划，分享自微信公众号。

原始发表：2019-07-07，如有侵权请联系 cloudcommunity@tencent.com 删除

数据分析