关于CIFAR10数据集的探索
365天深度学习训练营
第二周——彩色图片分类
- 选题：彩色图片分类
- 难度：新手入门
- 语言：Python3、TensorFlow2
- 时间：8月1-8月5日
- 要求：
- 学习如何编写一个完整的深度学习程序
- 了解分类彩色图片会灰度图片有什么区别
- 测试集accuracy到达72%
参考文章：深度学习100例-卷积神经网络（CNN）彩色图片分类 | 第2天
报名学习：365天深度学习训练营
相比于前一周（手写数字识别）
- 图片由原本色彩简单的图片（黑白图片），变为彩色图片，即由原来的单通道变为3通道（RGB）
一、导入库
import pandas as pd
import tensorflow as tf
from tensorflow import keras
from keras.layers import Dense, Input, Conv2D, MaxPool2D, Flatten, Dropout
from keras.models import Model,Sequential
from keras.optimizers import Adam
import matplotlib.pyplot as plt
2022-12-23 18:39:50.183425: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /home/usr/liangjie/soft/netcdf-fortran-4.5.4/release/lib:/home/usr/liangjie/soft/netcdf-c-4.9.0/release/lib:/home/usr/liangjie/soft/zlib-1.2.12/release/lib:/home/usr/liangjie/soft/hdf/HDF5-1.12.2-Linux/HDF_Group/HDF5/1.12.2/lib:/home/usr/liangjie/soft/gdal-3.5.0/release/gdal-release/lib64:
2022-12-23 18:39:50.183455: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
二、读取数据集
数据集说明：CIFAR-10数据集是由CIFAR(Candian Institute For Advanced Research) 收集整理的一个用于机器学习和图像识别问题的数据集。这个数据集共有60000张32 * 32的涵盖10个分类的彩色图片。
类别为：airplane,automobile,bird,cat,deer,dog,frog,horse,ship,truck
(train_images, train_labels),(test_images, test_labels) = keras.datasets.cifar10.load_data()
dict_labels = {0:'airplane',1:'automobile',2:'bird',3:'cat',4:'deer',5:'dog',6:'frog',7:'horse',8:'ship',9:'truck'}
Downloading data from https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
170498071/170498071 [==============================] - 14s 0us/step
三、训练集预览
import matplotlib.pyplot as plt
plt.figure(figsize=(10,10))
for i in range(20):
plt.subplot(5, 5, i+1)
plt.xticks([])
plt.yticks([])
plt.grid(False)
plt.imshow(train_images[i], cmap=plt.cm.binary)
plt.xlabel(dict_labels[train_labels[i][0]])
plt.show()
四、数据预处理
RGB3色图片和黑白图片的处理方式相似
train_images.shape
(50000, 32, 32, 3)
test_images.shape
(10000, 32, 32, 3)
train_images, test_images = train_images/255.0, test_images/255.0
五、构建模型
feature_input = Input(shape=(32, 32,3))
c1 = Conv2D(filters=64, kernel_size=(3,3), activation='relu', padding='same')(feature_input)
c1 = MaxPool2D(pool_size=(2,2))(c1)
c1 = Dropout(0.2)(c1)
c2 = Conv2D(filters=128, kernel_size=(3,3), activation='relu',padding='same')(c1)
c2 = MaxPool2D(pool_size=(2,2))(c2)
c2 = Dropout(0.2)(c2)
c3 = Conv2D(filters=256, kernel_size=(3,3), activation='relu',padding='same')(c2)
c3 = MaxPool2D(pool_size=(2,2))(c3)
c3 = Dropout(0.2)(c3)
f1 = Flatten()(c3)
d1 = Dense(units=128, activation='relu')(f1)
d1 = Dropout(0.2)(d1)
output = Dense(units=10, activation='softmax')(d1)
model=Model(feature_input, output)
model.compile(optimizer=Adam(0.001), loss='sparse_categorical_crossentropy', metrics=['accuracy'])
2022-12-23 18:41:05.942791: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcuda.so.1'; dlerror: libcuda.so.1: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /home/usr/liangjie/soft/netcdf-fortran-4.5.4/release/lib:/home/usr/liangjie/soft/netcdf-c-4.9.0/release/lib:/home/usr/liangjie/soft/zlib-1.2.12/release/lib:/home/usr/liangjie/soft/hdf/HDF5-1.12.2-Linux/HDF_Group/HDF5/1.12.2/lib:/home/usr/liangjie/soft/gdal-3.5.0/release/gdal-release/lib64:
2022-12-23 18:41:05.942888: W tensorflow/stream_executor/cuda/cuda_driver.cc:269] failed call to cuInit: UNKNOWN ERROR (303)
2022-12-23 18:41:05.942923: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:156] kernel driver does not appear to be running on this host (ALPHA.ITPCAS.AC.CN): /proc/driver/nvidia/version does not exist
2022-12-23 18:41:05.943166: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 AVX512F FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
model.summary()
Model: "model"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 32, 32, 3)] 0
conv2d (Conv2D) (None, 32, 32, 64) 1792
max_pooling2d (MaxPooling2D (None, 16, 16, 64) 0
)
dropout (Dropout) (None, 16, 16, 64) 0
conv2d_1 (Conv2D) (None, 16, 16, 128) 73856
max_pooling2d_1 (MaxPooling (None, 8, 8, 128) 0
2D)
dropout_1 (Dropout) (None, 8, 8, 128) 0
conv2d_2 (Conv2D) (None, 8, 8, 256) 295168
max_pooling2d_2 (MaxPooling (None, 4, 4, 256) 0
2D)
dropout_2 (Dropout) (None, 4, 4, 256) 0
flatten (Flatten) (None, 4096) 0
dense (Dense) (None, 128) 524416
dropout_3 (Dropout) (None, 128) 0
dense_1 (Dense) (None, 10) 1290
=================================================================
Total params: 896,522
Trainable params: 896,522
Non-trainable params: 0
_________________________________________________________________
history = model.fit(train_images, train_labels, validation_data=(test_images, test_labels), batch_size=512, epochs=20, verbose=1)
Epoch 1/20
98/98 [==============================] - 27s 272ms/step - loss: 6.5774 - accuracy: 0.1016 - val_loss: 2.3022 - val_accuracy: 0.1106
Epoch 2/20
98/98 [==============================] - 24s 246ms/step - loss: 2.3014 - accuracy: 0.1085 - val_loss: 2.3025 - val_accuracy: 0.0987
Epoch 3/20
98/98 [==============================] - 25s 257ms/step - loss: 2.2365 - accuracy: 0.1530 - val_loss: 2.0413 - val_accuracy: 0.2585
Epoch 4/20
98/98 [==============================] - 25s 258ms/step - loss: 1.8970 - accuracy: 0.2962 - val_loss: 1.6350 - val_accuracy: 0.4292
Epoch 5/20
98/98 [==============================] - 25s 258ms/step - loss: 1.6378 - accuracy: 0.4050 - val_loss: 1.4260 - val_accuracy: 0.5028
Epoch 6/20
98/98 [==============================] - 25s 255ms/step - loss: 1.4965 - accuracy: 0.4634 - val_loss: 1.3223 - val_accuracy: 0.5333
Epoch 7/20
98/98 [==============================] - 25s 259ms/step - loss: 1.3920 - accuracy: 0.5044 - val_loss: 1.2111 - val_accuracy: 0.5832
Epoch 8/20
98/98 [==============================] - 25s 257ms/step - loss: 1.3014 - accuracy: 0.5395 - val_loss: 1.1273 - val_accuracy: 0.6059
Epoch 9/20
98/98 [==============================] - 25s 255ms/step - loss: 1.2342 - accuracy: 0.5640 - val_loss: 1.0519 - val_accuracy: 0.6363
Epoch 10/20
98/98 [==============================] - 25s 256ms/step - loss: 1.1710 - accuracy: 0.5897 - val_loss: 1.0454 - val_accuracy: 0.6392
Epoch 11/20
98/98 [==============================] - 25s 254ms/step - loss: 1.1112 - accuracy: 0.6088 - val_loss: 0.9404 - val_accuracy: 0.6826
Epoch 12/20
98/98 [==============================] - 25s 255ms/step - loss: 1.0676 - accuracy: 0.6288 - val_loss: 0.9205 - val_accuracy: 0.6856
Epoch 13/20
98/98 [==============================] - 25s 254ms/step - loss: 1.0190 - accuracy: 0.6447 - val_loss: 0.9234 - val_accuracy: 0.6849
Epoch 14/20
98/98 [==============================] - 25s 260ms/step - loss: 0.9938 - accuracy: 0.6570 - val_loss: 0.8526 - val_accuracy: 0.7130
Epoch 15/20
98/98 [==============================] - 25s 254ms/step - loss: 0.9471 - accuracy: 0.6696 - val_loss: 0.8378 - val_accuracy: 0.7157
Epoch 16/20
98/98 [==============================] - 25s 255ms/step - loss: 0.9251 - accuracy: 0.6768 - val_loss: 0.8582 - val_accuracy: 0.7045
Epoch 17/20
98/98 [==============================] - 25s 254ms/step - loss: 0.9005 - accuracy: 0.6879 - val_loss: 0.8318 - val_accuracy: 0.7130
Epoch 18/20
98/98 [==============================] - 25s 255ms/step - loss: 0.8719 - accuracy: 0.6966 - val_loss: 0.7779 - val_accuracy: 0.7311
Epoch 19/20
98/98 [==============================] - 25s 255ms/step - loss: 0.8441 - accuracy: 0.7053 - val_loss: 0.7546 - val_accuracy: 0.7404
Epoch 20/20
98/98 [==============================] - 25s 254ms/step - loss: 0.8201 - accuracy: 0.7141 - val_loss: 0.7747 - val_accuracy: 0.7337
model.compile(optimizer=Adam(0.0001), loss='sparse_categorical_crossentropy', metrics=['accuracy'])
history_2 = model.fit(train_images, train_labels, validation_data=(test_images, test_labels), batch_size=512, epochs=20, verbose=1)
Epoch 1/20
98/98 [==============================] - 26s 261ms/step - loss: 0.7555 - accuracy: 0.7356 - val_loss: 0.6947 - val_accuracy: 0.7606
Epoch 2/20
98/98 [==============================] - 25s 256ms/step - loss: 0.7320 - accuracy: 0.7448 - val_loss: 0.6943 - val_accuracy: 0.7609
Epoch 3/20
98/98 [==============================] - 26s 261ms/step - loss: 0.7221 - accuracy: 0.7468 - val_loss: 0.6848 - val_accuracy: 0.7654
Epoch 4/20
98/98 [==============================] - 25s 256ms/step - loss: 0.7126 - accuracy: 0.7516 - val_loss: 0.6739 - val_accuracy: 0.7697
Epoch 5/20
98/98 [==============================] - 25s 255ms/step - loss: 0.7102 - accuracy: 0.7497 - val_loss: 0.6758 - val_accuracy: 0.7681
Epoch 6/20
98/98 [==============================] - 25s 256ms/step - loss: 0.6990 - accuracy: 0.7576 - val_loss: 0.6784 - val_accuracy: 0.7668
Epoch 7/20
98/98 [==============================] - 25s 260ms/step - loss: 0.6949 - accuracy: 0.7571 - val_loss: 0.6726 - val_accuracy: 0.7691
Epoch 8/20
98/98 [==============================] - 26s 264ms/step - loss: 0.6851 - accuracy: 0.7610 - val_loss: 0.6743 - val_accuracy: 0.7709
Epoch 9/20
98/98 [==============================] - 26s 263ms/step - loss: 0.6721 - accuracy: 0.7644 - val_loss: 0.6680 - val_accuracy: 0.7703
Epoch 10/20
98/98 [==============================] - 26s 264ms/step - loss: 0.6693 - accuracy: 0.7668 - val_loss: 0.6720 - val_accuracy: 0.7689
Epoch 11/20
98/98 [==============================] - 26s 263ms/step - loss: 0.6626 - accuracy: 0.7670 - val_loss: 0.6637 - val_accuracy: 0.7689
Epoch 12/20
98/98 [==============================] - 26s 262ms/step - loss: 0.6600 - accuracy: 0.7717 - val_loss: 0.6551 - val_accuracy: 0.7769
Epoch 13/20
98/98 [==============================] - 26s 262ms/step - loss: 0.6536 - accuracy: 0.7703 - val_loss: 0.6635 - val_accuracy: 0.7727
Epoch 14/20
98/98 [==============================] - 26s 262ms/step - loss: 0.6446 - accuracy: 0.7755 - val_loss: 0.6572 - val_accuracy: 0.7748
Epoch 15/20
98/98 [==============================] - 25s 259ms/step - loss: 0.6428 - accuracy: 0.7750 - val_loss: 0.6475 - val_accuracy: 0.7752
Epoch 16/20
98/98 [==============================] - 25s 257ms/step - loss: 0.6379 - accuracy: 0.7764 - val_loss: 0.6524 - val_accuracy: 0.7755
Epoch 17/20
98/98 [==============================] - 25s 259ms/step - loss: 0.6281 - accuracy: 0.7781 - val_loss: 0.6457 - val_accuracy: 0.7770
Epoch 18/20
98/98 [==============================] - 26s 261ms/step - loss: 0.6218 - accuracy: 0.7814 - val_loss: 0.6459 - val_accuracy: 0.7757
Epoch 19/20
98/98 [==============================] - 25s 259ms/step - loss: 0.6173 - accuracy: 0.7842 - val_loss: 0.6422 - val_accuracy: 0.7775
Epoch 20/20
98/98 [==============================] - 25s 258ms/step - loss: 0.6072 - accuracy: 0.7854 - val_loss: 0.6389 - val_accuracy: 0.7802
df_val_loss = history.history['val_loss']+history_2.history['val_loss']
df_val_acc = history.history['val_accuracy']+history_2.history['val_accuracy']
df_loss = history.history['loss']+history_2.history['loss']
df_acc = history.history['accuracy']+history_2.history['accuracy']
plt.figure(figsize=(12, 3))
plt.subplot(1, 2, 1)
size = len(df_acc)
plt.plot( df_acc, label='Training Accuracy')
plt.plot( df_val_acc, label='Test Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')
plt.subplot(1, 2, 2)
plt.plot( df_loss, label='Training Loss')
plt.plot( df_val_loss, label='Test Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
