深度学习100例-卷积神经网络（VGG-16）识别海贼王草帽一伙 | 第6天
生活随笔
收集整理的這篇文章主要介紹了
深度学习100例-卷积神经网络（VGG-16）识别海贼王草帽一伙 | 第6天
小編覺得挺不錯的,現(xiàn)在分享給大家,幫大家做個參考.
文章目錄
- 一、前期工作
- 1. 設(shè)置GPU
- 2. 導(dǎo)入數(shù)據(jù)
- 3. 查看數(shù)據(jù)
- 二、數(shù)據(jù)預(yù)處理
- 1. 加載數(shù)據(jù)
- 2. 可視化數(shù)據(jù)
- 3. 再次檢查數(shù)據(jù)
- 4. 配置數(shù)據(jù)集
- 5. 歸一化
- 三、構(gòu)建VGG-16網(wǎng)絡(luò)
- 1. 官方模型（已打包好）
- 2. 自建模型
- 3. 網(wǎng)絡(luò)結(jié)構(gòu)圖
- 四、編譯
- 五、訓(xùn)練模型
- 六、模型評估
一、前期工作
本文將實現(xiàn)海賊王中人物角色的識別。
🚀 我的環(huán)境：
- 語言環(huán)境：Python3.6.5
- 編譯器：jupyter notebook
- 深度學(xué)習(xí)環(huán)境：TensorFlow2.4.1
- 數(shù)據(jù)和代碼：📌【傳送門】
🚀 來自專欄：《深度學(xué)習(xí)100例》
如果你是一名深度學(xué)習(xí)小白可以先看看我這個專門為你寫的專欄：《小白入門深度學(xué)習(xí)》
🚀 往期精彩-卷積神經(jīng)網(wǎng)絡(luò)篇：
🚀 往期精彩-循環(huán)神經(jīng)網(wǎng)絡(luò)篇：
🚀 往期精彩-生成對抗網(wǎng)絡(luò)篇：
1. 設(shè)置GPU
如果使用的是CPU可以忽略這步
import tensorflow as tfgpus = tf.config.list_physical_devices("GPU")if gpus:tf.config.experimental.set_memory_growth(gpus[0], True) #設(shè)置GPU顯存用量按需使用tf.config.set_visible_devices([gpus[0]],"GPU")2. 導(dǎo)入數(shù)據(jù)
import matplotlib.pyplot as plt import os,PIL# 設(shè)置隨機(jī)種子盡可能使結(jié)果可以重現(xiàn) import numpy as np np.random.seed(1)# 設(shè)置隨機(jī)種子盡可能使結(jié)果可以重現(xiàn) import tensorflow as tf tf.random.set_seed(1)from tensorflow import keras from tensorflow.keras import layers,modelsimport pathlib data_dir = "D:\jupyter notebook\DL-100-days\datasets\hzw_photos"data_dir = pathlib.Path(data_dir)3. 查看數(shù)據(jù)
數(shù)據(jù)集中一共有路飛、索隆、娜美、烏索普、喬巴、山治、羅賓等7個人物角色
| lufei | 路飛 | 117 張 |
| suolong | 索隆 | 90 張 |
| namei | 娜美 | 84 張 |
| wusuopu | 烏索普 | 77張 |
| qiaoba | 喬巴 | 102 張 |
| shanzhi | 山治 | 47 張 |
| luobin | 羅賓 | 105張 |
二、數(shù)據(jù)預(yù)處理
1. 加載數(shù)據(jù)
使用image_dataset_from_directory方法將磁盤中的數(shù)據(jù)加載到tf.data.Dataset中
batch_size = 32 img_height = 224 img_width = 224 """ 關(guān)于image_dataset_from_directory()的詳細(xì)介紹可以參考文章：https://mtyjkh.blog.csdn.net/article/details/117018789 """ train_ds = tf.keras.preprocessing.image_dataset_from_directory(data_dir,validation_split=0.2,subset="training",seed=123,image_size=(img_height, img_width),batch_size=batch_size) Found 621 files belonging to 7 classes. Using 497 files for training. """ 關(guān)于image_dataset_from_directory()的詳細(xì)介紹可以參考文章：https://mtyjkh.blog.csdn.net/article/details/117018789 """ val_ds = tf.keras.preprocessing.image_dataset_from_directory(data_dir,validation_split=0.2,subset="validation",seed=123,image_size=(img_height, img_width),batch_size=batch_size) Found 621 files belonging to 7 classes. Using 124 files for validation.我們可以通過class_names輸出數(shù)據(jù)集的標(biāo)簽。標(biāo)簽將按字母順序?qū)?yīng)于目錄名稱。
class_names = train_ds.class_names print(class_names) ['lufei', 'luobin', 'namei', 'qiaoba', 'shanzhi', 'suolong', 'wusuopu']2. 可視化數(shù)據(jù)
plt.figure(figsize=(10, 5)) # 圖形的寬為10高為5for images, labels in train_ds.take(1):for i in range(8):ax = plt.subplot(2, 4, i + 1) plt.imshow(images[i].numpy().astype("uint8"))plt.title(class_names[labels[i]])plt.axis("off") plt.imshow(images[1].numpy().astype("uint8")) <matplotlib.image.AxesImage at 0x2adcea36ee0>3. 再次檢查數(shù)據(jù)
for image_batch, labels_batch in train_ds:print(image_batch.shape)print(labels_batch.shape)break (32, 224, 224, 3) (32,)- Image_batch是形狀的張量（32,180,180,3）。這是一批形狀180x180x3的32張圖片（最后一維指的是彩色通道RGB）。
- Label_batch是形狀（32，）的張量，這些標(biāo)簽對應(yīng)32張圖片
4. 配置數(shù)據(jù)集
- shuffle()：打亂數(shù)據(jù)，關(guān)于此函數(shù)的詳細(xì)介紹可以參考：https://zhuanlan.zhihu.com/p/42417456
- prefetch()：預(yù)取數(shù)據(jù)，加速運(yùn)行，其詳細(xì)介紹可以參考我前兩篇文章，里面都有講解。
- cache()：將數(shù)據(jù)集緩存到內(nèi)存當(dāng)中，加速運(yùn)行
5. 歸一化
normalization_layer = layers.experimental.preprocessing.Rescaling(1./255)normalization_train_ds = train_ds.map(lambda x, y: (normalization_layer(x), y)) val_ds = val_ds.map(lambda x, y: (normalization_layer(x), y)) image_batch, labels_batch = next(iter(val_ds)) first_image = image_batch[0]# 查看歸一化后的數(shù)據(jù) print(np.min(first_image), np.max(first_image)) 0.0 0.9928046三、構(gòu)建VGG-16網(wǎng)絡(luò)
在官方模型與自建模型之間進(jìn)行二選一就可以啦，選著一個注釋掉另外一個，都是正版的VGG-16哈。
VGG優(yōu)缺點分析：
- VGG優(yōu)點
VGG的結(jié)構(gòu)非常簡潔，整個網(wǎng)絡(luò)都使用了同樣大小的卷積核尺寸（3x3）和最大池化尺寸（2x2）。
- VGG缺點
1)訓(xùn)練時間過長，調(diào)參難度大。2)需要的存儲容量大，不利于部署。例如存儲VGG-16權(quán)重值文件的大小為500多MB，不利于安裝到嵌入式系統(tǒng)中。
1. 官方模型（已打包好）
官網(wǎng)模型調(diào)用這塊我放到后面幾篇文章中，下面主要講一下VGG-16
# model = keras.applications.VGG16() # model.summary()2. 自建模型
from tensorflow.keras import layers, models, Input from tensorflow.keras.models import Model from tensorflow.keras.layers import Conv2D, MaxPooling2D, Dense, Flatten, Dropoutdef VGG16(nb_classes, input_shape):input_tensor = Input(shape=input_shape)# 1st blockx = Conv2D(64, (3,3), activation='relu', padding='same',name='block1_conv1')(input_tensor)x = Conv2D(64, (3,3), activation='relu', padding='same',name='block1_conv2')(x)x = MaxPooling2D((2,2), strides=(2,2), name = 'block1_pool')(x)# 2nd blockx = Conv2D(128, (3,3), activation='relu', padding='same',name='block2_conv1')(x)x = Conv2D(128, (3,3), activation='relu', padding='same',name='block2_conv2')(x)x = MaxPooling2D((2,2), strides=(2,2), name = 'block2_pool')(x)# 3rd blockx = Conv2D(256, (3,3), activation='relu', padding='same',name='block3_conv1')(x)x = Conv2D(256, (3,3), activation='relu', padding='same',name='block3_conv2')(x)x = Conv2D(256, (3,3), activation='relu', padding='same',name='block3_conv3')(x)x = MaxPooling2D((2,2), strides=(2,2), name = 'block3_pool')(x)# 4th blockx = Conv2D(512, (3,3), activation='relu', padding='same',name='block4_conv1')(x)x = Conv2D(512, (3,3), activation='relu', padding='same',name='block4_conv2')(x)x = Conv2D(512, (3,3), activation='relu', padding='same',name='block4_conv3')(x)x = MaxPooling2D((2,2), strides=(2,2), name = 'block4_pool')(x)# 5th blockx = Conv2D(512, (3,3), activation='relu', padding='same',name='block5_conv1')(x)x = Conv2D(512, (3,3), activation='relu', padding='same',name='block5_conv2')(x)x = Conv2D(512, (3,3), activation='relu', padding='same',name='block5_conv3')(x)x = MaxPooling2D((2,2), strides=(2,2), name = 'block5_pool')(x)# full connectionx = Flatten()(x)x = Dense(4096, activation='relu', name='fc1')(x)x = Dense(4096, activation='relu', name='fc2')(x)output_tensor = Dense(nb_classes, activation='softmax', name='predictions')(x)model = Model(input_tensor, output_tensor)return modelmodel=VGG16(1000, (img_width, img_height, 3)) model.summary() Model: "model" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= input_1 (InputLayer) [(None, 224, 224, 3)] 0 _________________________________________________________________ block1_conv1 (Conv2D) (None, 224, 224, 64) 1792 _________________________________________________________________ block1_conv2 (Conv2D) (None, 224, 224, 64) 36928 _________________________________________________________________ block1_pool (MaxPooling2D) (None, 112, 112, 64) 0 _________________________________________________________________ block2_conv1 (Conv2D) (None, 112, 112, 128) 73856 _________________________________________________________________ block2_conv2 (Conv2D) (None, 112, 112, 128) 147584 _________________________________________________________________ block2_pool (MaxPooling2D) (None, 56, 56, 128) 0 _________________________________________________________________ block3_conv1 (Conv2D) (None, 56, 56, 256) 295168 _________________________________________________________________ block3_conv2 (Conv2D) (None, 56, 56, 256) 590080 _________________________________________________________________ block3_conv3 (Conv2D) (None, 56, 56, 256) 590080 _________________________________________________________________ block3_pool (MaxPooling2D) (None, 28, 28, 256) 0 _________________________________________________________________ block4_conv1 (Conv2D) (None, 28, 28, 512) 1180160 _________________________________________________________________ block4_conv2 (Conv2D) (None, 28, 28, 512) 2359808 _________________________________________________________________ block4_conv3 (Conv2D) (None, 28, 28, 512) 2359808 _________________________________________________________________ block4_pool (MaxPooling2D) (None, 14, 14, 512) 0 _________________________________________________________________ block5_conv1 (Conv2D) (None, 14, 14, 512) 2359808 _________________________________________________________________ block5_conv2 (Conv2D) (None, 14, 14, 512) 2359808 _________________________________________________________________ block5_conv3 (Conv2D) (None, 14, 14, 512) 2359808 _________________________________________________________________ block5_pool (MaxPooling2D) (None, 7, 7, 512) 0 _________________________________________________________________ flatten (Flatten) (None, 25088) 0 _________________________________________________________________ fc1 (Dense) (None, 4096) 102764544 _________________________________________________________________ fc2 (Dense) (None, 4096) 16781312 _________________________________________________________________ predictions (Dense) (None, 1000) 4097000 ================================================================= Total params: 138,357,544 Trainable params: 138,357,544 Non-trainable params: 0 _________________________________________________________________3. 網(wǎng)絡(luò)結(jié)構(gòu)圖
關(guān)于卷積的相關(guān)知識可以參考文章：https://mtyjkh.blog.csdn.net/article/details/114278995
結(jié)構(gòu)說明：
- 13個卷積層（Convolutional Layer），分別用blockX_convX表示
- 3個全連接層（Fully connected Layer），分別用fcX與predictions表示
- 5個池化層（Pool layer），分別用blockX_pool表示
VGG-16包含了16個隱藏層（13個卷積層和3個全連接層），故稱為VGG-16
四、編譯
在準(zhǔn)備對模型進(jìn)行訓(xùn)練之前，還需要再對其進(jìn)行一些設(shè)置。以下內(nèi)容是在模型的編譯步驟中添加的：
- 損失函數(shù)（loss）：用于衡量模型在訓(xùn)練期間的準(zhǔn)確率。
- 優(yōu)化器（optimizer）：決定模型如何根據(jù)其看到的數(shù)據(jù)和自身的損失函數(shù)進(jìn)行更新。
- 指標(biāo)（metrics）：用于監(jiān)控訓(xùn)練和測試步驟。以下示例使用了準(zhǔn)確率，即被正確分類的圖像的比率。
五、訓(xùn)練模型
epochs = 20history = model.fit(train_ds,validation_data=val_ds,epochs=epochs ) Epoch 1/20 16/16 [==============================] - 14s 461ms/step - loss: 4.5842 - accuracy: 0.1349 - val_loss: 6.8389 - val_accuracy: 0.1129 Epoch 2/20 16/16 [==============================] - 2s 146ms/step - loss: 2.1046 - accuracy: 0.1398 - val_loss: 6.7905 - val_accuracy: 0.2016 Epoch 3/20 16/16 [==============================] - 2s 144ms/step - loss: 1.7885 - accuracy: 0.3531 - val_loss: 6.7892 - val_accuracy: 0.2903 Epoch 4/20 16/16 [==============================] - 2s 145ms/step - loss: 1.2015 - accuracy: 0.6135 - val_loss: 6.7582 - val_accuracy: 0.2742 Epoch 5/20 16/16 [==============================] - 2s 148ms/step - loss: 1.1831 - accuracy: 0.6108 - val_loss: 6.7520 - val_accuracy: 0.4113 Epoch 6/20 16/16 [==============================] - 2s 143ms/step - loss: 0.5140 - accuracy: 0.8326 - val_loss: 6.7102 - val_accuracy: 0.5806 Epoch 7/20 16/16 [==============================] - 2s 150ms/step - loss: 0.2451 - accuracy: 0.9165 - val_loss: 6.6918 - val_accuracy: 0.7823 Epoch 8/20 16/16 [==============================] - 2s 147ms/step - loss: 0.2156 - accuracy: 0.9328 - val_loss: 6.7188 - val_accuracy: 0.4113 Epoch 9/20 16/16 [==============================] - 2s 143ms/step - loss: 0.1940 - accuracy: 0.9513 - val_loss: 6.6639 - val_accuracy: 0.5968 Epoch 10/20 16/16 [==============================] - 2s 143ms/step - loss: 0.0767 - accuracy: 0.9812 - val_loss: 6.6101 - val_accuracy: 0.7419 Epoch 11/20 16/16 [==============================] - 2s 146ms/step - loss: 0.0245 - accuracy: 0.9894 - val_loss: 6.5526 - val_accuracy: 0.8226 Epoch 12/20 16/16 [==============================] - 2s 149ms/step - loss: 0.0387 - accuracy: 0.9861 - val_loss: 6.5636 - val_accuracy: 0.6210 Epoch 13/20 16/16 [==============================] - 2s 152ms/step - loss: 0.2146 - accuracy: 0.9289 - val_loss: 6.7039 - val_accuracy: 0.4839 Epoch 14/20 16/16 [==============================] - 2s 152ms/step - loss: 0.2566 - accuracy: 0.9087 - val_loss: 6.6852 - val_accuracy: 0.6532 Epoch 15/20 16/16 [==============================] - 2s 149ms/step - loss: 0.0579 - accuracy: 0.9840 - val_loss: 6.5971 - val_accuracy: 0.6935 Epoch 16/20 16/16 [==============================] - 2s 152ms/step - loss: 0.0414 - accuracy: 0.9866 - val_loss: 6.6049 - val_accuracy: 0.7581 Epoch 17/20 16/16 [==============================] - 2s 146ms/step - loss: 0.0907 - accuracy: 0.9689 - val_loss: 6.6476 - val_accuracy: 0.6452 Epoch 18/20 16/16 [==============================] - 2s 147ms/step - loss: 0.0929 - accuracy: 0.9685 - val_loss: 6.6590 - val_accuracy: 0.7903 Epoch 19/20 16/16 [==============================] - 2s 146ms/step - loss: 0.0364 - accuracy: 0.9935 - val_loss: 6.5915 - val_accuracy: 0.6290 Epoch 20/20 16/16 [==============================] - 2s 151ms/step - loss: 0.1081 - accuracy: 0.9662 - val_loss: 6.6541 - val_accuracy: 0.6613六、模型評估
acc = history.history['accuracy'] val_acc = history.history['val_accuracy']loss = history.history['loss'] val_loss = history.history['val_loss']epochs_range = range(epochs)plt.figure(figsize=(12, 4)) plt.subplot(1, 2, 1) plt.plot(epochs_range, acc, label='Training Accuracy') plt.plot(epochs_range, val_acc, label='Validation Accuracy') plt.legend(loc='lower right') plt.title('Training and Validation Accuracy')plt.subplot(1, 2, 2) plt.plot(epochs_range, loss, label='Training Loss') plt.plot(epochs_range, val_loss, label='Validation Loss') plt.legend(loc='upper right') plt.title('Training and Validation Loss') plt.show()為體現(xiàn)原汁原味的VGG-16，本文并未對模型參數(shù)進(jìn)行修改，可依據(jù)實際情況修改模型中的相關(guān)性參數(shù)，適應(yīng)實際情況以便提升分類效果。
其他精彩內(nèi)容：
- 深度學(xué)習(xí)100例-卷積神經(jīng)網(wǎng)絡(luò)（CNN）實現(xiàn)mnist手寫數(shù)字識別 | 第1天
- 深度學(xué)習(xí)100例-卷積神經(jīng)網(wǎng)絡(luò)（CNN）彩色圖片分類 | 第2天
- 深度學(xué)習(xí)100例-卷積神經(jīng)網(wǎng)絡(luò)（CNN）服裝圖像分類 | 第3天
- 深度學(xué)習(xí)100例-卷積神經(jīng)網(wǎng)絡(luò)（CNN）花朵識別 | 第4天
- 深度學(xué)習(xí)100例-卷積神經(jīng)網(wǎng)絡(luò)（CNN）天氣識別 | 第5天
《深度學(xué)習(xí)100例》專欄直達(dá)：【傳送門】
如果覺得本文對你有幫助記得 點個關(guān)注，給個贊，加個收藏
總結(jié)
以上是生活随笔為你收集整理的深度学习100例-卷积神经网络（VGG-16）识别海贼王草帽一伙 | 第6天的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。
- 上一篇： 递归思想如何理解？
- 下一篇： 采用keras深度学习框架搭建卷积神经网