ResNet50：通过构建ResNet训练自己的数据集

先进行设备检查，tensorflow和pytorch的cuda不一样，pytorch上的cuda有效不代表tf上就可以

import tensorflow as tf  
  
# 列出所有物理GPU设备  
gpus = tf.config.list_physical_devices('GPU')  
  
if gpus:  
    print("检测到了GPU设备:")  
    for gpu in gpus:  
        try:  
            # 设置GPU显存用量按需使用  
            tf.config.experimental.set_memory_growth(gpu, True)  
            print(f"GPU设备: {gpu}")  
        except RuntimeError as e:  
            print(e)  
else:  
    print("没有检测到GPU设备，将使用CPU。")  

import matplotlib.pyplot as plt  
# 支持中文  
plt.rcParams['font.sans-serif'] = ['SimHei']  # 用来正常显示中文标签  
plt.rcParams['axes.unicode_minus'] = False  # 用来正常显示负号  
  
import os  
import numpy as np  
  
# 设置随机种子尽可能使结果可以重现  
np.random.seed(1)  
tf.random.set_seed(1)  
  
from tensorflow.keras import layers, models  
from tensorflow.keras.preprocessing.image import ImageDataGenerator  
from tensorflow.keras.applications.resnet50 import preprocess_input, decode_predictions  
  
import pathlib  

 数据集选择imagenet上的100类别，可以直接下载，不用额外传入

# 数据目录  
data_dir = "D:/pythonproject/imagenet/dataset/train"  
data_dir = pathlib.Path(data_dir)  

# 数据集参数  
batch_size = 8  
img_height = 224  
img_width = 224  

 

# 数据集参数  
batch_size = 8  
img_height = 224  
img_width = 224  

加载训练集和验证集，由于imagenet上的训练集很大，可以直接从训练集中划分，或者使用自带的验证集

# 加载训练集和验证集  
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,  
    shuffle=True  
)  
  
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,  
    shuffle=False  
)  

 进行数据增强

train_datagen = ImageDataGenerator(  
     preprocessing_function=preprocess_input,  
     # 其他数据增强参数...  
 )  

直接检查类别名称，imagenet文件夹自带标签，不用再去额外传入标签

# 类别名称  
class_names = train_ds.class_names  
print(class_names)  

#['1.10burger_2', '1.1hotdog', '1.2bread', '1.3burger', '1.4lemon', '1.5squash', '1.6mushroom', '1.7apple_2', '1.8simit', '1.9hotdog_2', '2.10ship', '2.1car_2', '2.2tank', '2.3wheel', '2.4wheel_2', '2.5scooter', '2.6forklift', '2.7bike', '2.8traffic_light', '2.9warplane', '3.10vase', '3.1sofa_2', '3.2crate', '3.3toaster', '3.4chair', '3.5bathing', '3.6sofa_3', '3.7piano', '3.8chair_2', '3.9washer', '4.10type', '4.1mic', '4.2computer', '4.3mouse', '4.4camera', '4.5control', '4.6lamp', '4.7guitar', '4.8iron', '4.9keyboard', '5.10manhole', '5.1trash_bin', '5.2bottle', '5.3shoe', '5.4basketball', '5.5shoe_2', '5.6pool_table', '5.7sign', '5.8sign_2', '5.9sign_3', '6.10volleyball', '6.11barrel', '6.12mic_2', '6.13scooter_2', '6.14tool', '6.15shoe_3', '6.16candle', '6.17coffeepot', '6.18vase_2', '6.19teapot', '6.1clock', '6.20plane', '6.21car_4', '6.22soap', '6.23ipod', '6.24cassette', '6.25sofa_5', '6.26lip', '6.27sax', '6.28cart', '6.29cart_2', '6.2clock_2', '6.30barrel_2', '6.31light', '6.32desk', '6.33coffee', '6.34cup', '6.35goblet', '6.36umbrella', '6.37helmet', '6.38syringe', '6.39swing', '6.3warplane_2', '6.40teddy', '6.41screwdriver', '6.42bank', '6.43lock', '6.44abacus', '6.45knife', '6.46hourglass', '6.47hammer', '6.48telephone', '6.49binoculars', '6.4chair_3', '6.50barbell', '6.5car_3', '6.6sofa_4', '6.7glass', '6.8phone', '6.9guitar_2']

 配置数据集

# 配置数据集  
AUTOTUNE = tf.data.AUTOTUNE  
train_ds = train_ds.cache().prefetch(buffer_size=AUTOTUNE)  
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)  

 然后就到了最重要的构建resnet50网络的环节，如果要使用预训练的训练权重有完全不同的方法，预训练的权重在torchvision上有传入的代码，imagenet同理也有

# 只从 tensorflow.keras 导入  
from tensorflow.keras import layers, models  
from tensorflow.keras.layers import Input, Activation, BatchNormalization, Flatten
 
from keras.layers import Dense, Conv2D, MaxPooling2D, ZeroPadding2D, AveragePooling2D
 
def identity_block(input_tensor, kernel_size, filters, stage, block):
    filters1, filters2, filters3 = filters
 
    name_base = str(stage) + block + '_identity_block_block_'
 
    x = Conv2D(filters1, (1,1), name=name_base + 'conv1')(input_tensor)
    x = BatchNormalization(name=name_base + 'bn1')(x)
    x = Activation('relu', name=name_base + 'relu1')(x)
 
    x = Conv2D(filters2, kernel_size, padding='same', name=name_base + 'conv2')(x)
    x = BatchNormalization(name=name_base + 'bn2')(x)
    x = Activation('relu', name=name_base + 'relu2')(x)
 
    x = Conv2D(filters3, (1,1), name=name_base + 'conv3')(x)
    x = BatchNormalization(name=name_base + 'bn3')(x)
 
    x = layers.add([x, input_tensor], name=name_base + 'add')
    x = Activation('relu', name=name_base + 'relu4')(x)
    return x 
 
 
def conv_block(input_tensor, kernel_size, filters, stage, block, strides=(2, 2)):
 
    filters1, filters2, filters3 = filters
    res_name_base = str(stage) + block + '_conv_block_res_'
    name_base = str(stage) + block + '_conv_block_'
 
    x = Conv2D(filters1, (1,1), strides=strides, name=name_base + 'conv1')(input_tensor)
    x = BatchNormalization(name=name_base + 'bn1')(x)
    x = Activation('relu', name=name_base + 'relu1')(x)
 
 
    x = Conv2D(filters2, kernel_size, padding='same', name=name_base + 'conv2')(x)
    x = BatchNormalization(name=name_base + 'bn2')(x)
    x = Activation('relu', name=name_base + 'relu2')(x)
 
    x = Conv2D(filters3, (1,1), name=name_base + 'conv3')(x)
    x = BatchNormalization(name=name_base + 'bn3')(x)
 
    shortcut = Conv2D(filters3, (1,1), strides=strides, name=res_name_base + 'conv')(input_tensor)
    shortcut = BatchNormalization(name=name_base + 'bn')(shortcut)
 
    x = layers.add([x, shortcut], name=name_base + 'add')
    x = Activation('relu', name=name_base + 'relu4')(x)
 
    return x
 
def ResNet50(input_shape=[224,224,3], classes=1000):
    img_input = Input(shape=input_shape)
    x = ZeroPadding2D((3, 3))(img_input)
 
    x = Conv2D(64, (7,7), strides=(2,2), name='conv1')(x)
    x = BatchNormalization(name='bn_conv1')(x)
    x = Activation('relu')(x)
    x = MaxPooling2D((3, 3), strides=(2,2))(x)
 
 
    x = conv_block(x, 3, [64, 64, 256], stage=2, block='a', strides=(1, 1))
    x = identity_block(x, 3, [64, 64, 256], stage=2, block='b')
    x = identity_block(x, 3, [64, 64, 256], stage=2, block='c')
    
    x = conv_block(x, 3, [128, 128, 512], stage=3, block='a')
    x = identity_block(x, 3, [128, 128, 512], stage=3, block='b')
    x = identity_block(x, 3, [128, 128, 512], stage=3, block='c')
    x = identity_block(x, 3, [128, 128, 512], stage=3, block='d')
 
    x = conv_block(x, 3, [256, 256, 1024], stage=4, block='a')
    x = identity_block(x, 3, [256, 256, 1024], stage=4, block='b')
    x = identity_block(x, 3, [256, 256, 1024], stage=4, block='c')
    x = identity_block(x, 3, [256, 256, 1024], stage=4, block='d')
    x = identity_block(x, 3, [256, 256, 1024], stage=4, block='e')
    x = identity_block(x, 3, [256, 256, 1024], stage=4, block='f')
 
 
    x = conv_block(x, 3, [512, 512, 2048], stage=5, block='a')
    x = identity_block(x, 3, [512, 512, 2048], stage=5, block='b')
    x = identity_block(x, 3, [512, 512, 2048], stage=5, block='c')
 
    x = AveragePooling2D((7,7), name='avg_pool')(x)
 
    x = Flatten()(x)
 
    # 修改输出层以匹配您的类别数  
    x = Dense(classes, activation='softmax', name='fc' + str(classes))(x)  
  
    model = models.Model(inputs=img_input, outputs=x, name='resnet50')  
    return model  
 
model = ResNet50()
 
model.summary()

然后进行编译与训练，训练轮次其实很小就能取得不错的效果，在如此臃肿的架构下太高的训练轮次很容易造成过拟合：

# 编译模型  
opt = tf.keras.optimizers.Adam(learning_rate=1e-5)  # 调整学习率  
model.compile(optimizer=opt, loss='sparse_categorical_crossentropy', metrics=['accuracy'])  

# 训练模型  
epochs = 5  
history = model.fit(train_ds, validation_data=val_ds, epochs=epochs)

保存模型，下次可直接调用预训练的权重：

# 指定保存模型的路径  
model_save_path = 'path_to_save_model/resnet50_ten_epochs'  
  
# 保存整个模型，包括架构、权重和编译信息  
tf.keras.models.save_model(model, model_save_path, save_format='tf')  
  
print(f"模型已保存到 {model_save_path}")

可视化模型训练效果：

# 评估模型并绘制结果  
acc = history.history['accuracy']  
val_acc = history.history['val_accuracy']  
loss = history.history['loss']  
val_loss = history.history['val_loss']  
  
# 绘制平滑曲线（可选，这里简单使用移动平均）  
import numpy as np  
smoothed_acc = np.convolve(acc, np.ones(5)/5, mode='valid')  
smoothed_val_acc = np.convolve(val_acc, np.ones(5)/5, mode='valid')  
smoothed_loss = np.convolve(loss, np.ones(5)/5, mode='valid')  
smoothed_val_loss = np.convolve(val_loss, np.ones(5)/5, mode='valid')  
  
# 绘制训练和验证的准确率和损失  
plt.figure(figsize=(14, 8))  
plt.subplot(1, 2, 1)  
plt.plot(range(len(smoothed_acc)), smoothed_acc, label='Training Accuracy (Smoothed)')  
plt.plot(range(len(smoothed_val_acc)), smoothed_val_acc, label='Validation Accuracy (Smoothed)')  
plt.plot(range(len(acc)), acc, alpha=0.5, label='Training Accuracy')  
plt.plot(range(len(val_acc)), val_acc, alpha=0.5, label='Validation Accuracy')  
plt.legend(loc='lower right')  
plt.title('Training and Validation Accuracy')  
  
plt.subplot(1, 2, 2)  
plt.plot(range(len(smoothed_loss)), smoothed_loss, label='Training Loss (Smoothed)')  
plt.plot(range(len(smoothed_val_loss)), smoothed_val_loss, label='Validation Loss (Smoothed)')  
plt.plot(range(len(loss)), loss, alpha=0.5, label='Training Loss')  
plt.plot(range(len(val_loss)), val_loss, alpha=0.5, label='Validation Loss')  
plt.legend(loc='upper right')  
plt.title('Training and Validation Loss')  
plt.show()  
  
# 打印每个epoch的详细信息  
for epoch, (train_acc, val_acc) in enumerate(zip(acc, val_acc)):  
    print(f'Epoch {epoch+1}, Train Acc: {train_acc:.4f}, Val Acc: {val_acc:.4f}')  

最后绘制验证的图形：

# 绘制验证图像和预测  
plt.figure(figsize=(10, 5))  
for images, labels in val_ds.take(1):  
    for i in range(8):  
        ax = plt.subplot(2, 4, i + 1)  
        plt.imshow(images[i].numpy().astype("uint8"))  
        img_array = tf.expand_dims(images[i], 0)  
        predictions = model.predict(img_array)  
        plt.title(class_names[np.argmax(predictions)])  
        plt.axis('off')  
plt.show()