本文共 34328 字,大约阅读时间需要 114 分钟。
我又来了!!!!
这次进行实战项目,鸟类细粒度分类识别实战。再讲细粒度分类之前,让我们先回顾一下图像分类吧。
图像分类是计算机视觉的最基础的一个任务,从最开始的入门级的mnist手写数字识别、猫狗图像二分类到后来的imagenet任务。图像分类模型随着数据集的增长,一步步提升到了今天的水平。计算机的图像分类水准已经超过了人类。
在这里我把图像分类任务分为了两种,一种是单标签的图像分类任务,一种是多标签的图像分类任务。
多标签的图像分类任务,更加符合人们的认知习惯。因为现实生活中的图片往往会包含多个类别物体。
而在单标签的图像分类任务中又可以分为三类:一种是跨物种语义级别的图像分类,即在不同物种的层次上识别不同类别对象,比如我们常见的猫狗分类。
一种是实例级图像分类即区分不同的个体,最典型的任务那就是人脸识别。 而还剩下最后一种就是细粒度分类,那么什么是细粒度分类呢?而细粒度图像分类,相比较我们前面所说 的跨物种的图像分类,级别更低一些。但相比较实例级的图像分类,级别稍高一些。
概念上的说法 是对同一大类中的子类的分类,
通俗来讲,其主要是解决 我们在日常生活中 可能看到一只狗,确分不清是哪种狗。如下图所示,我们知道下图中哪一只是阿拉斯加 哪一只是哈士奇,左边是哈士奇 右边是阿拉斯加
这里可以当做可判别性部分是 阿拉斯加犬的鼻梁是与黑色毛色是相连的,这就是discriminative part 即可判别性模块。然后再讲讲细粒度分类实验目前所遇到的挑战吧。
细粒度分类的挑战,但如今面临着如下三大问题:类内差异大
、类间差异小
,以及有限的数据集
。
由于光线,物体的姿势,视角、遮挡、背景干扰等等问题,
类内差异大,像这里的黑脚信天翁,由于光照,背景,姿势的干扰,从肉眼上很难看出属于同一个子类类间差异小 不同个体归属于不同子类可能是由于一些微小的不同,如鸟的翅膀的颜色 以及鸟喙颜色的不同
以及有限的数据集的问题,数据集的标注 通常需要专业的知识以及耗费大量的标注时间。
由于上述挑战问题,我们很难根据现有的粗粒度神经网络模型得到精准的分类结果。
那么目前的研究现状是如何呢?
目前细粒度分类主要是通过寻找可判别性的特征
来进行分类的,研究方法目前主要是可分为强监督学习
和弱监督学习
。
强监督学习
是指通常使用边界框和局部标注信息
,来获取目标的位置、大小,从而提高分类精度。 即给出了图片标注中物体的某些显著特征
,即discriminative
。
而弱监督学习
是指仅利用图像的类别标注信息
,不使用额外的标注,
目前弱监督学习的主要思路是定位出判别性的部位,取得判别性的特征做辅助来分类
。
其实这很符合人类辨别细粒度物体的流程,先看全局信息知道大类,然后根据经验把注意力放在一些关键部位来做出判断,而这些部位就是弱监督网络所要找的discriminative parts
。
目前的强监督学习方法有part-based r-cnn
基于r-cnn算法完成了局部区域的检测
,利用约束条件对r-cnn提取到的区域信息进行修正之后提取卷积特征,并将不同区域的特征进行拼接,构成最好的特征表示,然后通过SVM分类器
进行分类训练。 Posed-normalized Cnn
对每一张图片进行位置检测,然后将检测框内的图像进行裁剪,从而提取不同层次、不同位置的图像,再对提取到的图像块进行姿态对其送入CNN
,将得到的特征拼接后利用SVM分类器
进行分类。 Multi-proposal Net
通过Edge Box Crop
方法获取图像块,并引入关键点及视觉特征的输出层,进一步强化了局部特征与全部信息直接的位置关联。
弱监督方法,有图像过滤,仅借助于图像的类别信息过滤图片中与物体无关的模块,其中最有代表性的是Two-level算法。two attention level
利用物体级和局部级的信息,通过Search Selective算法
过滤掉无关背景,然后将过滤掉的背景送入CNN网络进行训练,得到物体级的分类结果,随后通过聚类算法将不同位置的特征继续区分,并将不同区域的特征拼接后送入svm分类器
进行训练。
人在认知物体和事物时,往往需要完成对其特征的理解及类别名称的记忆,B-CNN
根据大脑工作时同认知类别和关注显著特征的方法,构建了两个线性网络,协调完成局部特征提取和分类的任务。
到这里,前期的基础知识差不多就完成了,下面准备进入正题。
首先还是一如既往先介绍我们的驱动力----数据。
不对,放错图了,应该是下面这张。
本次细粒度分类所采取的数据集CUB200-2011,该数据集是由加州理工学院在2010年提出的细粒度数据集,也是目前细粒度分类识别研究的基准图像数据集,该数据集共有117888张鸟类图像,包含了200类鸟类子类,其中训练数据集有5994张图像,测试集有5794张图像,每张图像均提供了图像类标注信息,图像中鸟的bounding box,鸟的关键part信息,以及鸟的属性信息。
评判标准就是以准确率了。
好了,准备上模型了!
先用VGG16来投石问路
在此之前准备好我们的微调模型
# fine-tune 模型def fine_tune_model(model, optimizer, batch_size, epochs, freeze_num): ''' discription: 对指定预训练模型进行fine-tune,并保存为.hdf5格式 MODEL:传入的模型,VGG16, ResNet50, ... optimizer: fine-tune all layers 的优化器, first part默认用adadelta batch_size: 每一批的尺寸,建议32/64/128 epochs: fine-tune all layers的代数 freeze_num: first part冻结卷积层的数量 ''' # datagen = ImageDataGenerator( # rescale=1.255, # # shear_range=0.2, # # zoom_range=0.2, # # horizontal_flip=True, # # vertical_flip=True, # # fill_mode="nearest" # ) # datagen.fit(X_train) # first: 仅训练全连接层(权重随机初始化的) # 冻结所有卷积层 for layer in model.layers[:freeze_num]: layer.trainable = False model.compile(optimizer=optimizer, loss="categorical_crossentropy", metrics=["accuracy"]) # model.fit_generator(datagen.flow(x_train,y_train,batch_size=batch_size), # steps_per_epoch=len(x_train)/32, # epochs=3, # shuffle=True, # verbose=1, # datagen.flow(x_valid, y_valid)) model.fit(x_train, y_train, batch_size=batch_size, epochs=3, shuffle=True, verbose=1, validation_data=(x_valid,y_valid) ) print('Finish step_1') # second: fine-tune all layers for layer in model.layers[:]: layer.trainable = True rc = ReduceLROnPlateau(monitor="val_acc", factor=0.2, patience=4, verbose=1, mode='max') model_name = model.name + ".hdf5" mc = ModelCheckpoint(model_name, monitor="val_acc", save_best_only=True, verbose=1, mode='max') el = EarlyStopping(monitor="val_acc", min_delta=0, patience=5, verbose=1, restore_best_weights=True) model.compile(optimizer=optimizer, loss='categorical_crossentropy', metrics=["accuracy"]) # history_fit = model.fit_generator(datagen.flow(x_train,y_train,batch_size=32), # steps_per_epoch=len(x_train)/32, # epochs=epochs, # shuffle=True, # verbose=1, # callbacks=[mc,rc,el], # datagen.flow(x_valid, y_valid)) history_fit = model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, shuffle=True, verbose=1, validation_data=(x_valid,y_valid), callbacks=[mc,rc,el]) print('Finish fine-tune') return history_fit
1.VGG16模型
# 定义一个VGG16的模型def vgg16_model(img_rows,img_cols): x = Input(shape=(img_rows, img_cols, 3)) x = Lambda(imagenet_utils.preprocess_input)(x) base_model = VGG16(input_tensor=x,weights="imagenet",include_top=False, pooling='avg') x = base_model.output x = Dense(1024,activation="relu",name="fc1")(x) x = Dropout(0.5)(x) predictions = Dense(n_classes,activation="softmax",name="predictions")(x) vgg16_model = Model(inputs=base_model.input,outputs=predictions,name="vgg16") return vgg16_model
# 创建VGG16模型img_rows, img_cols = 300, 300vgg16_model = vgg16_model(img_rows,img_cols)
for i,layer in enumerate(vgg16_model.layers): print(i,layer.name)
0 input_31 lambda_32 block1_conv13 block1_conv24 block1_pool5 block2_conv16 block2_conv27 block2_pool8 block3_conv19 block3_conv210 block3_conv311 block3_pool12 block4_conv113 block4_conv214 block4_conv315 block4_pool16 block5_conv117 block5_conv218 block5_conv319 block5_pool20 global_average_pooling2d_321 fc122 dropout_323 predictions
optimizer = optimizers.Adam(lr=0.0001)batch_size = 32epochs = 30freeze_num = 21%time vgg16_history = fine_tune_model(vgg16_model,optimizer,batch_size,epochs,freeze_num)
WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/optimizers.py:790: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead. Train on 7013 samples, validate on 3006 samplesEpoch 1/37013/7013 [==============================] - 53s 8ms/step - loss: 7.1095 - acc: 0.0211 - val_loss: 4.4823 - val_acc: 0.0915Epoch 2/37013/7013 [==============================] - 46s 7ms/step - loss: 4.4798 - acc: 0.0914 - val_loss: 3.6892 - val_acc: 0.2239Epoch 3/37013/7013 [==============================] - 46s 7ms/step - loss: 3.6436 - acc: 0.1925 - val_loss: 3.0040 - val_acc: 0.3440Finish step_1Train on 7013 samples, validate on 3006 samplesEpoch 1/307013/7013 [==============================] - 47s 7ms/step - loss: 2.9171 - acc: 0.3087 - val_loss: 2.1821 - val_acc: 0.4667Epoch 00001: val_loss improved from inf to 2.18212, saving model to vgg16.hdf5Epoch 2/307013/7013 [==============================] - 46s 7ms/step - loss: 1.9944 - acc: 0.4840 - val_loss: 1.8748 - val_acc: 0.5226Epoch 00002: val_loss improved from 2.18212 to 1.87480, saving model to vgg16.hdf5Epoch 3/307013/7013 [==============================] - 46s 7ms/step - loss: 1.6493 - acc: 0.5551 - val_loss: 1.7540 - val_acc: 0.5492Epoch 00003: val_loss improved from 1.87480 to 1.75400, saving model to vgg16.hdf5Epoch 4/307013/7013 [==============================] - 46s 7ms/step - loss: 1.4144 - acc: 0.6144 - val_loss: 1.6711 - val_acc: 0.5655Epoch 00004: val_loss improved from 1.75400 to 1.67106, saving model to vgg16.hdf5Epoch 5/307013/7013 [==============================] - 46s 7ms/step - loss: 1.2055 - acc: 0.6628 - val_loss: 1.6020 - val_acc: 0.5749Epoch 00005: val_loss improved from 1.67106 to 1.60200, saving model to vgg16.hdf5Epoch 00026: val_loss improved from 1.32242 to 1.32005, saving model to vgg16.hdf5Epoch 27/307013/7013 [==============================] - 46s 7ms/step - loss: 0.1979 - acc: 0.9511 - val_loss: 1.3209 - val_acc: 0.6517Epoch 00027: val_loss did not improve from 1.32005Epoch 28/307013/7013 [==============================] - 46s 7ms/step - loss: 0.1996 - acc: 0.9528 - val_loss: 1.3206 - val_acc: 0.6514Epoch 00028: val_loss did not improve from 1.32005Epoch 29/307013/7013 [==============================] - 46s 7ms/step - loss: 0.1956 - acc: 0.9555 - val_loss: 1.3216 - val_acc: 0.6517Epoch 00029: val_loss did not improve from 1.32005Epoch 00029: ReduceLROnPlateau reducing learning rate to 3.999999898951501e-06.Epoch 30/307013/7013 [==============================] - 46s 7ms/step - loss: 0.1884 - acc: 0.9558 - val_loss: 1.3194 - val_acc: 0.6514Epoch 00030: val_loss improved from 1.32005 to 1.31935, saving model to vgg16.hdf5Finish fine-tuneCPU times: user 10min, sys: 3min 58s, total: 13min 58sWall time: 25min 37s
history_plot(vgg16_history)
进过上面的一系列操作,我们可以看到VGG16的分类效果,并不是很好呀,只能刚刚及格。
那么下面有请我们的二号选手EfficientNet
2.EfficientNetB4
咚咚咚,它来了,它来了,它踩着七彩祥云来了!!!
好了,不多说了,直接上代码来搭建EfficientNet网络架构。
# 定义一个EfficientNet模型def efficient_model(img_rows,img_cols): K.clear_session() x = Input(shape=(img_rows,img_cols,3)) x = Lambda(imagenet_utils.preprocess_input)(x) base_model = EfficientNetB4(input_tensor=x,weights="imagenet",include_top=False,pooling="avg") x = base_model.output x = Dense(1024,activation="relu",name="fc1")(x) x = Dropout(0.5)(x) predictions = Dense(n_classes,activation="softmax",name="predictions")(x) eB_model = Model(inputs=base_model.input,outputs=predictions,name="eB4") return eB_model
# 创建Efficient模型img_rows,img_cols=224,224eB_model = efficient_model(img_rows,img_cols)
optimizer = optimizers.Adam(lr=0.0001)batch_size = 32epochs = 30freeze_num = 469eB_model_history = fine_tune_model(eB_model,optimizer,batch_size,epochs,freeze_num)
Train on 8251 samples, validate on 1768 samplesEpoch 1/38251/8251 [==============================] - 49s 6ms/step - loss: 9.3405 - acc: 0.0053 - val_loss: 5.5664 - val_acc: 0.0051Epoch 2/38251/8251 [==============================] - 38s 5ms/step - loss: 6.8968 - acc: 0.0052 - val_loss: 5.3289 - val_acc: 0.0040Epoch 3/38251/8251 [==============================] - 39s 5ms/step - loss: 5.8723 - acc: 0.0061 - val_loss: 5.3021 - val_acc: 0.0040Finish step_1Train on 8251 samples, validate on 1768 samplesEpoch 1/308251/8251 [==============================] - 261s 32ms/step - loss: 4.4794 - acc: 0.0980 - val_loss: 2.7448 - val_acc: 0.3399Epoch 00001: val_loss improved from inf to 2.74482, saving model to eB4.hdf5Epoch 2/308251/8251 [==============================] - 155s 19ms/step - loss: 2.2635 - acc: 0.4157 - val_loss: 1.4371 - val_acc: 0.5973Epoch 00002: val_loss improved from 2.74482 to 1.43707, saving model to eB4.hdf5Epoch 3/308251/8251 [==============================] - 155s 19ms/step - loss: 1.3465 - acc: 0.6244 - val_loss: 1.1637 - val_acc: 0.6719Epoch 00003: val_loss improved from 1.43707 to 1.16373, saving model to eB4.hdf5Epoch 4/308251/8251 [==============================] - 154s 19ms/step - loss: 0.8824 - acc: 0.7488 - val_loss: 0.9904 - val_acc: 0.7110Epoch 00016: val_loss did not improve from 0.89365Epoch 17/308251/8251 [==============================] - 154s 19ms/step - loss: 0.0718 - acc: 0.9867 - val_loss: 0.8993 - val_acc: 0.7749Epoch 00017: val_loss did not improve from 0.89365Restoring model weights from the end of the best epochEpoch 00017: early stoppingFinish fine-tune
history_plot(eB_model_history)
效果很不错呀,EfficientNet不愧是谷歌出品的,必是精品。那么既然EfficientNet的效果已经这么好了,你是不是就不想接着看了,你是不是已经迫不及待想尝试EfficientNet的效果了呢。
不要急,下面还有几个小尝试,首先是在EfficientNet中加入Attention机制,至于Attention机制的话,可以去看我的博客里面有写到,那是在我未解放天性之前,写的可正经了。当然这里更是正经!!!
3.efficientnet-with-attention
# 定义一个加入Attention模块的Efficient网络架构即efficientnet-with-attentiondef efficient_attention_model(img_rows,img_cols): K.clear_session() in_lay = Input(shape=(img_rows,img_cols,3)) base_model = EfficientNetB3(input_shape=(img_rows,img_cols,3),weights="imagenet",include_top=False) pt_depth = base_model.get_output_shape_at(0)[-1] pt_features = base_model(in_lay) bn_features = BatchNormalization()(pt_features) # here we do an attention mechanism to turn pixels in the GAP on an off atten_layer = Conv2D(64,kernel_size=(1,1),padding="same",activation="relu")(Dropout(0.5)(bn_features)) atten_layer = Conv2D(16,kernel_size=(1,1),padding="same",activation="relu")(atten_layer) atten_layer = Conv2D(8,kernel_size=(1,1),padding="same",activation="relu")(atten_layer) atten_layer = Conv2D(1,kernel_size=(1,1),padding="valid",activation="sigmoid")(atten_layer)# H,W,1 # fan it out to all of the channels up_c2_w = np.ones((1,1,1,pt_depth)) #1,1,C up_c2 = Conv2D(pt_depth,kernel_size=(1,1),padding="same",activation="linear",use_bias=False,weights=[up_c2_w]) up_c2.trainable = False atten_layer = up_c2(atten_layer)# H,W,C mask_features = multiply([atten_layer,bn_features])# H,W,C gap_features = GlobalAveragePooling2D()(mask_features)# 1,1,C # gap_mask = GlobalAveragePooling2D()(atten_layer)# 1,1,C # # to account for missing values from the attention model # gap = Lambda(lambda x:x[0]/x[1],name="RescaleGAP")([gap_features,gap_mask]) gap_dr = Dropout(0.25)(gap_features) dr_steps = Dropout(0.25)(Dense(1000,activation="relu")(gap_dr)) out_layer = Dense(200,activation="softmax")(dr_steps) eb_atten_model = Model(inputs=[in_lay],outputs=[out_layer]) return eb_atten_model
img_rows,img_cols = 224,224eB_atten_model = efficient_attention_model(img_rows,img_cols)
eB_atten_model.save("eb_atten_model.h5")
for i,layer in enumerate(eB_atten_model.layers): print(i,layer.name)
0 input_11 efficientnet-b32 batch_normalization_13 dropout_14 conv2d_15 conv2d_26 conv2d_37 conv2d_48 conv2d_59 multiply_110 global_average_pooling2d_111 dropout_212 dense_113 dropout_314 dense_2
optimizer = optimizers.Adam(lr=0.0001)batch_size = 32epochs = 30freeze_num = 12eB_atten_model_history = fine_tune_model(eB_atten_model,optimizer,batch_size,epochs,freeze_num)
WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/optimizers.py:793: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.Train on 8251 samples, validate on 1768 samplesEpoch 1/38251/8251 [==============================] - 39s 5ms/step - loss: 5.2083 - acc: 0.0221 - val_loss: 16.0324 - val_acc: 0.0040Epoch 2/38251/8251 [==============================] - 28s 3ms/step - loss: 4.7719 - acc: 0.1130 - val_loss: 16.0147 - val_acc: 0.0057Epoch 3/38251/8251 [==============================] - 28s 3ms/step - loss: 4.3135 - acc: 0.2112 - val_loss: 16.0056 - val_acc: 0.0062Finish step_1Train on 8251 samples, validate on 1768 samplesEpoch 1/308251/8251 [==============================] - 168s 20ms/step - loss: 2.1612 - acc: 0.4549 - val_loss: 1.1888 - val_acc: 0.6725Epoch 00001: val_loss improved from inf to 1.18880, saving model to model_1.hdf5Epoch 2/308251/8251 [==============================] - 121s 15ms/step - loss: 0.9003 - acc: 0.7442 - val_loss: 0.9400 - val_acc: 0.7330Epoch 00002: val_loss improved from 1.18880 to 0.94002, saving model to model_1.hdf5Epoch 3/308251/8251 [==============================] - 121s 15ms/step - loss: 0.5455 - acc: 0.8467 - val_loss: 0.8569 - val_acc: 0.7574Epoch 00013: val_loss did not improve from 0.78748Epoch 14/308251/8251 [==============================] - 121s 15ms/step - loss: 0.0417 - acc: 0.9924 - val_loss: 0.7958 - val_acc: 0.7924Epoch 00014: val_loss did not improve from 0.78748Epoch 00014: ReduceLROnPlateau reducing learning rate to 3.999999898951501e-06.Epoch 15/308251/8251 [==============================] - 121s 15ms/step - loss: 0.0370 - acc: 0.9936 - val_loss: 0.7938 - val_acc: 0.7941Epoch 00015: val_loss did not improve from 0.78748Epoch 16/308251/8251 [==============================] - 121s 15ms/step - loss: 0.0379 - acc: 0.9933 - val_loss: 0.7932 - val_acc: 0.7952Epoch 00016: val_loss did not improve from 0.78748Restoring model weights from the end of the best epochEpoch 00016: early stoppingFinish fine-tune
history_plot(eB_atten_model_history)
[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-
效果还是提升了一点点的,下面是又尝试了另外一种attention的写法,用到senet和cbam。当然如果你不了解的话,还是老规矩,去看我的博客,卷积神经网络发展史里面有提到。
4.EfficientNetB3 with attention v2
from keras.layers import GlobalAveragePooling2D, GlobalMaxPooling2D, Reshape, Dense, multiply, Permute, Concatenate, Conv2D, Add, Activation, Lambdafrom keras import backend as Kfrom keras.activations import sigmoiddef attach_attention_module(net, attention_module): if attention_module == 'se_block': # SE_block net = se_block(net) elif attention_module == 'cbam_block': # CBAM_block net = cbam_block(net) else: raise Exception("'{}' is not supported attention module!".format(attention_module)) return netdef se_block(input_feature, ratio=8): """Contains the implementation of Squeeze-and-Excitation(SE) block. As described in https://arxiv.org/abs/1709.01507. """ channel_axis = 1 if K.image_data_format() == "channels_first" else -1 channel = input_feature._keras_shape[channel_axis] se_feature = GlobalAveragePooling2D()(input_feature) se_feature = Reshape((1, 1, channel))(se_feature) assert se_feature._keras_shape[1:] == (1,1,channel) se_feature = Dense(channel // ratio, activation='relu', kernel_initializer='he_normal', use_bias=True, bias_initializer='zeros')(se_feature) assert se_feature._keras_shape[1:] == (1,1,channel//ratio) se_feature = Dense(channel, activation='sigmoid', kernel_initializer='he_normal', use_bias=True, bias_initializer='zeros')(se_feature) assert se_feature._keras_shape[1:] == (1,1,channel) if K.image_data_format() == 'channels_first': se_feature = Permute((3, 1, 2))(se_feature) se_feature = multiply([input_feature, se_feature]) return se_featuredef cbam_block(cbam_feature, ratio=8): """Contains the implementation of Convolutional Block Attention Module(CBAM) block. As described in https://arxiv.org/abs/1807.06521. """ cbam_feature = channel_attention(cbam_feature, ratio) cbam_feature = spatial_attention(cbam_feature) return cbam_featuredef channel_attention(input_feature, ratio=8): channel_axis = 1 if K.image_data_format() == "channels_first" else -1 channel = input_feature._keras_shape[channel_axis] shared_layer_one = Dense(channel//ratio, activation='relu', kernel_initializer='he_normal', use_bias=True, bias_initializer='zeros') shared_layer_two = Dense(channel, kernel_initializer='he_normal', use_bias=True, bias_initializer='zeros') avg_pool = GlobalAveragePooling2D()(input_feature) avg_pool = Reshape((1,1,channel))(avg_pool) assert avg_pool._keras_shape[1:] == (1,1,channel) avg_pool = shared_layer_one(avg_pool) assert avg_pool._keras_shape[1:] == (1,1,channel//ratio) avg_pool = shared_layer_two(avg_pool) assert avg_pool._keras_shape[1:] == (1,1,channel) max_pool = GlobalMaxPooling2D()(input_feature) max_pool = Reshape((1,1,channel))(max_pool) assert max_pool._keras_shape[1:] == (1,1,channel) max_pool = shared_layer_one(max_pool) assert max_pool._keras_shape[1:] == (1,1,channel//ratio) max_pool = shared_layer_two(max_pool) assert max_pool._keras_shape[1:] == (1,1,channel) cbam_feature = Add()([avg_pool,max_pool]) cbam_feature = Activation('sigmoid')(cbam_feature) if K.image_data_format() == "channels_first": cbam_feature = Permute((3, 1, 2))(cbam_feature) return multiply([input_feature, cbam_feature])def spatial_attention(input_feature): kernel_size = 7 if K.image_data_format() == "channels_first": channel = input_feature._keras_shape[1] cbam_feature = Permute((2,3,1))(input_feature) else: channel = input_feature._keras_shape[-1] cbam_feature = input_feature avg_pool = Lambda(lambda x: K.mean(x, axis=3, keepdims=True))(cbam_feature) assert avg_pool._keras_shape[-1] == 1 max_pool = Lambda(lambda x: K.max(x, axis=3, keepdims=True))(cbam_feature) assert max_pool._keras_shape[-1] == 1 concat = Concatenate(axis=3)([avg_pool, max_pool]) assert concat._keras_shape[-1] == 2 cbam_feature = Conv2D(filters = 1, kernel_size=kernel_size, strides=1, padding='same', activation='sigmoid', kernel_initializer='he_normal', use_bias=False)(concat) assert cbam_feature._keras_shape[-1] == 1 if K.image_data_format() == "channels_first": cbam_feature = Permute((3, 1, 2))(cbam_feature) return multiply([input_feature, cbam_feature])
# 定义一个EfficientNet模型def efficient__atten2_model(img_rows,img_cols): K.clear_session() in_lay = Input(shape=(img_rows,img_cols,3)) base_model = EfficientNetB3(input_shape=(img_rows,img_cols,3),weights="imagenet",include_top=False) pt_features = base_model(in_lay) bn_features = BatchNormalization()(pt_features) atten_features = attach_attention_module(bn_features,"se_block") gap_features = GlobalAveragePooling2D()(atten_features) gap_dr = Dropout(0.25)(gap_features) dr_steps = Dropout(0.25)(Dense(1000,activation="relu")(gap_dr)) out_layer = Dense(n_classes,activation="softmax")(dr_steps) eb_atten_model = Model(inputs=[in_lay],outputs=[out_layer]) return eb_atten_model
img_rows,img_cols = 224,224eB_atten2_model = efficient__atten2_model(img_rows,img_cols)
optimizer = optimizers.Adam(lr=0.0001)batch_size = 32epochs = 30freeze_num = 19eB_atten2_model_history = fine_tune_model(eB_atten2_model,optimizer,batch_size,epochs,freeze_num)
Train on 8251 samples, validate on 1768 samplesEpoch 1/38251/8251 [==============================] - 33s 4ms/step - loss: 5.3202 - acc: 0.0061 - val_loss: 16.0269 - val_acc: 0.0057Epoch 2/38251/8251 [==============================] - 26s 3ms/step - loss: 5.3261 - acc: 0.0051 - val_loss: 16.0269 - val_acc: 0.0057Epoch 3/38251/8251 [==============================] - 26s 3ms/step - loss: 5.3248 - acc: 0.0048 - val_loss: 16.0269 - val_acc: 0.0057Finish step_1Train on 8251 samples, validate on 1768 samplesEpoch 1/308251/8251 [==============================] - 153s 19ms/step - loss: 3.9559 - acc: 0.1742 - val_loss: 2.1066 - val_acc: 0.4712Epoch 00001: val_loss improved from inf to 2.10657, saving model to model_1.hdf5Epoch 2/308251/8251 [==============================] - 119s 14ms/step - loss: 1.6183 - acc: 0.5708 - val_loss: 1.1768 - val_acc: 0.6618Epoch 00002: val_loss improved from 2.10657 to 1.17679, saving model to model_1.hdf5Epoch 3/308251/8251 [==============================] - 119s 14ms/step - loss: 0.9172 - acc: 0.7374 - val_loss: 0.9507 - val_acc: 0.7189Epoch 00003: val_loss improved from 1.17679 to 0.95071, saving model to model_1.hdf5Epoch 4/308251/8251 [==============================] - 119s 14ms/step - loss: 0.5897 - acc: 0.8317 - val_loss: 0.8628 - val_acc: 0.7562Epoch 00004: val_loss improved from 0.95071 to 0.86283, saving model to model_1.hdf5Epoch 5/308251/8251 [==============================] - 119s 14ms/step - loss: 0.3838 - acc: 0.8956 - val_loss: 0.8359 - val_acc: 0.7636Epoch 00005: val_loss improved from 0.86283 to 0.83592, saving model to model_1.hdf5Epoch 6/308251/8251 [==============================] - 119s 14ms/step - loss: 0.2797 - acc: 0.9234 - val_loss: 0.8280 - val_acc: 0.7647Epoch 00006: val_loss improved from 0.83592 to 0.82797, saving model to model_1.hdf5Epoch 7/308251/8251 [==============================] - 119s 14ms/step - loss: 0.1997 - acc: 0.9495 - val_loss: 0.8620 - val_acc: 0.7602Epoch 00007: val_loss did not improve from 0.82797Epoch 8/308251/8251 [==============================] - 119s 14ms/step - loss: 0.1408 - acc: 0.9667 - val_loss: 0.8602 - val_acc: 0.7800Epoch 00008: val_loss did not improve from 0.82797Epoch 9/308251/8251 [==============================] - 119s 14ms/step - loss: 0.1103 - acc: 0.9739 - val_loss: 0.9202 - val_acc: 0.7545Epoch 00009: val_loss did not improve from 0.82797Epoch 00009: ReduceLROnPlateau reducing learning rate to 1.9999999494757503e-05.Epoch 10/308251/8251 [==============================] - 119s 14ms/step - loss: 0.0803 - acc: 0.9824 - val_loss: 0.8677 - val_acc: 0.7709Epoch 00010: val_loss did not improve from 0.82797Epoch 11/308251/8251 [==============================] - 119s 14ms/step - loss: 0.0772 - acc: 0.9833 - val_loss: 0.8560 - val_acc: 0.7771Epoch 00011: val_loss did not improve from 0.82797Restoring model weights from the end of the best epochEpoch 00011: early stoppingFinish fine-tune
history_plot(eB_atten2_model_history)
咱也不知道为什么,这个效果比上面那个attention的写法,会提升一点点,这就是炼丹吧。
5.双线性EfficientNet
下面就是尝试了一种双线性的网络架构,这里我还画了图呢!!!
该模型的整体流程是:
这里也一并给出attention机制的图吧!
# 定义一个双线性EfficientNet Attention模型def blinear_efficient__atten_model(img_rows,img_cols): K.clear_session() in_lay = Input(shape=(img_rows,img_cols,3)) base_model = EfficientNetB3(input_shape=(img_rows,img_cols,3),weights="imagenet",include_top=False) pt_depth = base_model.get_output_shape_at(0)[-1] cnn_features_a = base_model(in_lay) cnn_bn_features_a = BatchNormalization()(cnn_features_a) # attention mechanism # here we do an attention mechanism to turn pixels in the GAP on an off atten_layer = Conv2D(64,kernel_size=(1,1),padding="same",activation="relu")(Dropout(0.5)(cnn_bn_features_a)) atten_layer = Conv2D(16,kernel_size=(1,1),padding="same",activation="relu")(atten_layer) atten_layer = Conv2D(8,kernel_size=(1,1),padding="same",activation="relu")(atten_layer) atten_layer = Conv2D(1,kernel_size=(1,1),padding="valid",activation="sigmoid")(atten_layer)# H,W,1 # fan it out to all of the channels up_c2_w = np.ones((1,1,1,pt_depth)) #1,1,C up_c2 = Conv2D(pt_depth,kernel_size=(1,1),padding="same",activation="linear",use_bias=False,weights=[up_c2_w]) up_c2.trainable = True atten_layer = up_c2(atten_layer)# H,W,C cnn_atten_out_a = multiply([atten_layer,cnn_bn_features_a])# H,W,C cnn_atten_out_b = cnn_atten_out_a cnn_out_dot = multiply([cnn_atten_out_a,cnn_atten_out_b]) gap_features = GlobalAveragePooling2D()(cnn_out_dot) gap_dr = Dropout(0.25)(gap_features) dr_steps = Dropout(0.25)(Dense(1000,activation="relu")(gap_dr)) out_layer = Dense(200,activation="softmax")(dr_steps) b_eff_atten_model = Model(inputs=[in_lay],outputs=[out_layer],name="blinear_efficient_atten") return b_eff_atten_model
# 创建双线性EfficientNet Attention模型img_rows,img_cols = 256,256befficient_model = blinear_efficient__atten_model(img_rows,img_cols)
befficient_model.save("befficient_model.h5")
optimizer = optimizers.Adam(lr=0.0001)batch_size = 32epochs = 30freeze_num = 19befficient_model_history = fine_tune_model(befficient_model,optimizer,batch_size,epochs,freeze_num)
Train on 8251 samples, validate on 1768 samplesEpoch 1/38251/8251 [==============================] - 38s 5ms/step - loss: 5.3903 - acc: 0.0052 - val_loss: 14.1897 - val_acc: 0.0040Epoch 2/38251/8251 [==============================] - 33s 4ms/step - loss: 5.3926 - acc: 0.0052 - val_loss: 14.1897 - val_acc: 0.0040Epoch 3/38251/8251 [==============================] - 33s 4ms/step - loss: 5.3948 - acc: 0.0068 - val_loss: 14.1897 - val_acc: 0.0040Finish step_1Train on 8251 samples, validate on 1768 samplesEpoch 1/308251/8251 [==============================] - 193s 23ms/step - loss: 4.7127 - acc: 0.0749 - val_loss: 2.9079 - val_acc: 0.3060Epoch 00001: val_acc improved from -inf to 0.30600, saving model to blinear_efficient_atten.hdf5Epoch 2/308251/8251 [==============================] - 148s 18ms/step - loss: 2.1653 - acc: 0.4462 - val_loss: 1.3817 - val_acc: 0.6160Epoch 00002: val_acc improved from 0.30600 to 0.61595, saving model to blinear_efficient_atten.hdf5Epoch 3/308251/8251 [==============================] - 149s 18ms/step - loss: 1.1834 - acc: 0.6676 - val_loss: 1.0714 - val_acc: 0.7002Epoch 00003: val_acc improved from 0.61595 to 0.70023, saving model to blinear_efficient_atten.hdf5Epoch 4/308251/8251 [==============================] - 149s 18ms/step - loss: 0.8070 - acc: 0.7666 - val_loss: 0.9743 - val_acc: 0.7342Epoch 00004: val_acc improved from 0.70023 to 0.73416, saving model to blinear_efficient_atten.hdf5Epoch 5/30Epoch 00007: val_acc improved from 0.74830 to 0.75735, saving model to blinear_efficient_atten.hdf5Epoch 00010: val_acc did not improve from 0.76867Epoch 11/308251/8251 [==============================] - 149s 18ms/step - loss: 0.1421 - acc: 0.9547 - val_loss: 1.1319 - val_acc: 0.7692Epoch 00011: val_acc improved from 0.76867 to 0.76923, saving model to blinear_efficient_atten.hdf5Epoch 12/308251/8251 [==============================] - 149s 18ms/step - loss: 0.1232 - acc: 0.9622 - val_loss: 1.0809 - val_acc: 0.7704Epoch 00018: val_acc improved from 0.77489 to 0.78224, saving model to blinear_efficient_atten.hdf5Epoch 19/308251/8251 [==============================] - 149s 18ms/step - loss: 0.0880 - acc: 0.9714 - val_loss: 1.2171 - val_acc: 0.7721Epoch 00022: ReduceLROnPlateau reducing learning rate to 1.9999999494757503e-05.Epoch 23/308251/8251 [==============================] - 148s 18ms/step - loss: 0.0465 - acc: 0.9859 - val_loss: 1.1591 - val_acc: 0.7930Epoch 00023: val_acc improved from 0.78224 to 0.79299, saving model to blinear_efficient_atten.hdf5Epoch 24/308251/8251 [==============================] - 148s 18ms/step - loss: 0.0360 - acc: 0.9893 - val_loss: 1.1312 - val_acc: 0.7969Epoch 00024: val_acc improved from 0.79299 to 0.79695, saving model to blinear_efficient_atten.hdf5Epoch 25/308251/8251 [==============================] - 148s 18ms/step - loss: 0.0275 - acc: 0.9920 - val_loss: 1.1477 - val_acc: 0.8015Epoch 00028: val_acc did not improve from 0.80147Epoch 29/308251/8251 [==============================] - 148s 18ms/step - loss: 0.0248 - acc: 0.9922 - val_loss: 1.1467 - val_acc: 0.8020Epoch 00029: val_acc improved from 0.80147 to 0.80204, saving model to blinear_efficient_atten.hdf5Epoch 30/308251/8251 [==============================] - 148s 18ms/step - loss: 0.0232 - acc: 0.9919 - val_loss: 1.1427 - val_acc: 0.8003Epoch 00030: val_acc did not improve from 0.80204Finish fine-tune
history_plot(befficient_model_history)
可以从图中看到,双线性的结构,准确率还会提升一些。
终于来到故事的结尾处了,最后在尝试一些双线性的VGG16。
6.双线性VGG16模型
# 定义双线性VGG16模型from keras import backend as Kdef batch_dot(cnn_ab): return K.batch_dot(cnn_ab[0], cnn_ab[1], axes=[1, 1])def sign_sqrt(x): return K.sign(x) * K.sqrt(K.abs(x) + 1e-10)def l2_norm(x): return K.l2_normalize(x, axis=-1) def bilinear_vgg16(img_rows,img_cols): input_tensor = Input(shape=(img_rows,img_cols,3)) input_tensor = Lambda(imagenet_utils.preprocess_input)(input_tensor) model_vgg16 = VGG16(include_top=False, weights="imagenet", input_tensor=input_tensor,pooling="avg") cnn_out_a = model_vgg16.layers[-2].output cnn_out_shape = model_vgg16.layers[-2].output_shape cnn_out_a = Reshape([cnn_out_shape[1]*cnn_out_shape[2], cnn_out_shape[-1]])(cnn_out_a) cnn_out_b = cnn_out_a cnn_out_dot = Lambda(batch_dot)([cnn_out_a, cnn_out_b]) cnn_out_dot = Reshape([cnn_out_shape[-1]*cnn_out_shape[-1]])(cnn_out_dot) sign_sqrt_out = Lambda(sign_sqrt)(cnn_out_dot) l2_norm_out = Lambda(l2_norm)(sign_sqrt_out) fc1 = Dense(1024,activation="relu",name="fc1")(l2_norm_out) dropout = Dropout(0.5)(fc1) output = Dense(n_classes, activation="softmax",name="output")(dropout) bvgg16_model = Model(inputs=model_vgg16.input, outputs=output,name="bvgg16") return bvgg16_model
# 创建双线性VGG16模型img_rows,img_cols = 300,300bvgg16_model = bilinear_vgg16(img_rows,img_cols)
for i,layer in enumerate(bvgg16_model.layers): print(i,layer.name)
0 input_11 lambda_12 block1_conv13 block1_conv24 block1_pool5 block2_conv16 block2_conv27 block2_pool8 block3_conv19 block3_conv210 block3_conv311 block3_pool12 block4_conv113 block4_conv214 block4_conv315 block4_pool16 block5_conv117 block5_conv218 block5_conv319 block5_pool20 reshape_121 lambda_222 reshape_223 lambda_324 lambda_425 fc126 dropout_127 output
optimizer = optimizers.Adam(lr=0.0001)batch_size = 32epochs = 100freeze_num = 25bvgg16_history = fine_tune_model(bvgg16_model,optimizer,batch_size,epochs,freeze_num)
WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/optimizers.py:793: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.Train on 8251 samples, validate on 1768 samplesEpoch 1/38251/8251 [==============================] - 80s 10ms/step - loss: 5.1197 - acc: 0.0572 - val_loss: 4.8534 - val_acc: 0.2002Epoch 2/38251/8251 [==============================] - 71s 9ms/step - loss: 4.4758 - acc: 0.1863 - val_loss: 4.1177 - val_acc: 0.3569Epoch 3/38251/8251 [==============================] - 71s 9ms/step - loss: 3.7386 - acc: 0.2743 - val_loss: 3.4439 - val_acc: 0.4378Finish step_1Train on 8251 samples, validate on 1768 samplesEpoch 1/1008251/8251 [==============================] - 76s 9ms/step - loss: 2.9186 - acc: 0.3475 - val_loss: 2.5064 - val_acc: 0.5334Epoch 00001: val_loss improved from inf to 2.50638, saving model to bvgg16.hdf5Epoch 2/1008251/8251 [==============================] - 70s 9ms/step - loss: 2.3073 - acc: 0.4696 - val_loss: 2.1717 - val_acc: 0.5888Epoch 00002: val_loss improved from 2.50638 to 2.17170, saving model to bvgg16.hdf5Epoch 3/1008251/8251 [==============================] - 70s 9ms/step - loss: 2.0086 - acc: 0.5355 - val_loss: 1.9604 - val_acc: 0.6222Epoch 00067: val_loss did not improve from 0.89483Epoch 68/1008251/8251 [==============================] - 71s 9ms/step - loss: 0.0539 - acc: 0.9971 - val_loss: 0.8984 - val_acc: 0.7590Epoch 00068: val_loss did not improve from 0.89483Epoch 00068: ReduceLROnPlateau reducing learning rate to 3.999999898951501e-06.Epoch 69/1008251/8251 [==============================] - 71s 9ms/step - loss: 0.0536 - acc: 0.9972 - val_loss: 0.8972 - val_acc: 0.7602Epoch 00069: val_loss did not improve from 0.89483Epoch 70/1008251/8251 [==============================] - 71s 9ms/step - loss: 0.0517 - acc: 0.9973 - val_loss: 0.8968 - val_acc: 0.7630Epoch 00070: val_loss did not improve from 0.89483Restoring model weights from the end of the best epochEpoch 00070: early stoppingFinish fine-tune
history_plot(bvgg16_history)
终于,完成了。至于效果的话,大家就看图感受吧。效果肯定是不如EfficientNet了。
当然大家可以调调图片的分辨率,学习率,batch_size等等,好好练丹吧!
转载地址:http://degbf.baihongyu.com/