ResNet18和ResNet50的keras实现

已于 2023-04-07 11:15:44 修改
阅读量2.9k 收藏 32
点赞数
分类专栏： ML&DL 文章标签： ResNet Keras
于 2022-04-12 07:37:15 首次发布
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/liugan528/article/details/123719108
版权

注意低版本的keras对一些包的引用方式跟高版本有区别，注意看章节1的代码注释部分。

TensorFlow跟Keras也有版本的对应关系，https://master–floydhub-docs.netlify.app/guides/environments/。
例如：
在这里插入图片描述

1.ResNet18

ResNet18

from tensorflow import keras
from tensorflow.keras import layers
# 对于keras==2.0.8应该写成下面这句(pip install keras==2.0.8 -i https://pypi.tuna.tsinghua.edu.cn/simple)：
# from tensorflow.python.keras import layers

INPUT_SIZE = 224
CLASS_NUM = 1000

# stage_name=2,3,4,5;  block_name=a,b,c
def ConvBlock(input_tensor, num_output, stride, stage_name, block_name):
    filter1, filter2 = num_output

    x = layers.Conv2D(filter1, 3, strides=stride, padding='same', name='res'+stage_name+block_name+'_branch2a')(input_tensor)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2a')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2a_relu')(x)

    x = layers.Conv2D(filter2, 3, strides=(1, 1), padding='same', name='res'+stage_name+block_name+'_branch2b')(x)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2b')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2b_relu')(x)

    shortcut = layers.Conv2D(filter2, 1, strides=stride, padding='same', name='res'+stage_name+block_name+'_branch1')(input_tensor)
    shortcut = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch1')(shortcut)

    x = layers.add([x, shortcut], name='res'+stage_name+block_name)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_relu')(x)

    return x

def IdentityBlock(input_tensor, num_output, stage_name, block_name):
    filter1, filter2 = num_output

    x = layers.Conv2D(filter1, 3, strides=(1, 1), padding='same', name='res'+stage_name+block_name+'_branch2a')(input_tensor)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2a')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2a_relu')(x)

    x = layers.Conv2D(filter2, 3, strides=(1, 1), padding='same', name='res'+stage_name+block_name+'_branch2b')(x)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2b')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2b_relu')(x)

    shortcut = input_tensor

    x = layers.add([x, shortcut], name='res'+stage_name+block_name)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_relu')(x)

    return x

def ResNet18(input_shape, class_num):
    input = keras.Input(shape=input_shape, name='input')

    # conv1
    x = layers.Conv2D(64, 7, strides=(2, 2), padding='same', name='conv1')(input)  # 7×7, 64, stride 2
    x = layers.BatchNormalization(name='bn_conv1')(x)
    x = layers.Activation('relu', name='conv1_relu')(x)
    x = layers.MaxPooling2D((3, 3), strides=2, padding='same', name='pool1')(x)   # 3×3 max pool, stride 2

    # conv2_x
    x = ConvBlock(input_tensor=x, num_output=(64, 64), stride=(1, 1), stage_name='2', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(64, 64), stage_name='2', block_name='b')

    # conv3_x
    x = ConvBlock(input_tensor=x, num_output=(128, 128), stride=(2, 2), stage_name='3', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(128, 128), stage_name='3', block_name='b')

    # conv4_x
    x = ConvBlock(input_tensor=x, num_output=(256, 256), stride=(2, 2), stage_name='4', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(256, 256), stage_name='4', block_name='b')

    # conv5_x
    x = ConvBlock(input_tensor=x, num_output=(512, 512), stride=(2, 2), stage_name='5', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(512, 512), stage_name='5', block_name='b')

    # average pool, 1000-d fc, softmax
    x = layers.AveragePooling2D((7, 7), strides=(1, 1), name='pool5')(x)
    x = layers.Flatten(name='flatten')(x)
    x = layers.Dense(class_num, activation='softmax', name='fc1000')(x)

    model = keras.Model(input, x, name='resnet18')
    # 对于keras==2.0.8应该写成下面这句，否则会报错：AttributeError: module 'tensorflow.python.keras' has no attribute 'Model'
    # model = keras.models.Model(input, x, name='resnet18')

    model.summary()
    return model

if __name__ == '__main__':
    model = ResNet18((INPUT_SIZE, INPUT_SIZE, 3), CLASS_NUM)
    print('Done.')

train_resnet18.py

from tensorflow import keras
from tensorflow.keras import layers
# 对于keras==2.0.8应该写成：from tensorflow.python.keras import layers
# pip install keras==2.0.8  -i https://pypi.tuna.tsinghua.edu.cn/simple 

INPUT_SIZE = 224
CLASS_NUM = 2

# stage_name=2,3,4,5;  block_name=a,b,c
def ConvBlock(input_tensor, num_output, stride, stage_name, block_name):
    filter1, filter2 = num_output

    x = layers.Conv2D(filter1, 3, strides=stride, padding='same', name='res'+stage_name+block_name+'_branch2a')(input_tensor)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2a')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2a_relu')(x)

    x = layers.Conv2D(filter2, 3, strides=(1, 1), padding='same', name='res'+stage_name+block_name+'_branch2b')(x)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2b')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2b_relu')(x)

    shortcut = layers.Conv2D(filter2, 1, strides=stride, padding='same', name='res'+stage_name+block_name+'_branch1')(input_tensor)
    shortcut = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch1')(shortcut)

    x = layers.add([x, shortcut], name='res'+stage_name+block_name)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_relu')(x)

    return x

def IdentityBlock(input_tensor, num_output, stage_name, block_name):
    filter1, filter2 = num_output

    x = layers.Conv2D(filter1, 3, strides=(1, 1), padding='same', name='res'+stage_name+block_name+'_branch2a')(input_tensor)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2a')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2a_relu')(x)

    x = layers.Conv2D(filter2, 3, strides=(1, 1), padding='same', name='res'+stage_name+block_name+'_branch2b')(x)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2b')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2b_relu')(x)

    shortcut = input_tensor

    x = layers.add([x, shortcut], name='res'+stage_name+block_name)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_relu')(x)

    return x

def ResNet18(input_shape, class_num):
    input = keras.Input(shape=input_shape, name='input')

    # conv1
    x = layers.Conv2D(64, 7, strides=(2, 2), padding='same', name='conv1')(input)  # 7×7, 64, stride 2
    x = layers.BatchNormalization(name='bn_conv1')(x)
    x = layers.Activation('relu', name='conv1_relu')(x)
    x = layers.MaxPooling2D((3, 3), strides=2, padding='same', name='pool1')(x)   # 3×3 max pool, stride 2

    # conv2_x
    x = ConvBlock(input_tensor=x, num_output=(64, 64), stride=(1, 1), stage_name='2', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(64, 64), stage_name='2', block_name='b')

    # conv3_x
    x = ConvBlock(input_tensor=x, num_output=(128, 128), stride=(2, 2), stage_name='3', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(128, 128), stage_name='3', block_name='b')

    # conv4_x
    x = ConvBlock(input_tensor=x, num_output=(256, 256), stride=(2, 2), stage_name='4', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(256, 256), stage_name='4', block_name='b')

    # conv5_x
    x = ConvBlock(input_tensor=x, num_output=(512, 512), stride=(2, 2), stage_name='5', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(512, 512), stage_name='5', block_name='b')

    # average pool, 1000-d fc, softmax
    x = layers.AveragePooling2D((7, 7), strides=(1, 1), name='pool5')(x)
    x = layers.Flatten(name='flatten')(x)
    x = layers.Dense(class_num, activation='softmax', name='fc1000')(x)

    model = keras.Model(input, x, name='resnet18')
    model.summary()
    return model

if __name__ == '__main__':
    model = ResNet18((INPUT_SIZE, INPUT_SIZE, 3), CLASS_NUM)
    print('Done.')

predict_resnet18.py

import matplotlib.pyplot as plt

from ResNet18 import ResNet18
import cv2
import numpy as np
from tensorflow.keras import backend as K   # K.set_image_dim_ordering('tf')
from tensorflow.keras.utils import to_categorical

INPUT_IMG_SIZE = 224
NUM_CLASSES = 2
label_dict = {0:'CAT', 1:'DOG'}

def show_predict_probability(y_gts, predictions, x_imgs, predict_probabilitys, idx):
    for i in range(len(label_dict)):
        print(label_dict[i]+', Probability:%1.9f'%(predict_probabilitys[idx][i]))
    print('label: ', label_dict[int(y_gts[idx])], ', predict: ', label_dict[predictions[idx]])
    plt.figure(figsize=(2, 2))
    plt.imshow(np.reshape(x_imgs[idx], (INPUT_IMG_SIZE, INPUT_IMG_SIZE, 3)))
    plt.show()

def plot_images_labels_prediction(images, labels, prediction, idx, num):
    fig = plt.gcf()
    fig.set_size_inches(12, 14)
    if num>25: num=25
    for i in range(0, num):
        ax = plt.subplot(2, 5, 1+i)
        ax.imshow(images[idx], cmap='binary')
        title = 'labels='+str(labels[idx])
        if len(prediction) > 0:
            title += "prediction="+str(prediction[idx])
        ax.set_title(title, fontsize=10)
        idx += 1
    plt.show()

if __name__ == '__main__':
    log_path = r"D:\02.Work\00.LearnML\003.Net\ResNet\log\\"
    model = ResNet18((224, 224, 3), NUM_CLASSES)
    model.load_weights(log_path+"resnet18.h5")

    ### cat dog dataset
    lines = []
    root_path = r"D:\03.Data\01.CatDog"
    with open(root_path + "\\test.txt") as f:
       lines = f.readlines()

    x_images_normalize = []
    y_labels_onehot = []
    y_labels = []

    for i in range(len(lines)):
        img_path = lines[i].split(";")[0]
        img = cv2.imread(img_path)
        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
        img = cv2.resize(img, (INPUT_IMG_SIZE, INPUT_IMG_SIZE))
        img = img / 255
        x_images_normalize.append(img)
        label = to_categorical(lines[i].split(";")[1], num_classes=NUM_CLASSES)
        y_labels_onehot.append(label)
        y_labels.append(lines[i].split(";")[1])
    x_images_normalize = np.array(x_images_normalize)
    # x_images_normalize = x_images_normalize.reshape(-1, INPUT_IMG_SIZE, INPUT_IMG_SIZE, 3)
    y_labels_onehot = np.array(y_labels_onehot)

    predict_probability = model.predict(x_images_normalize, verbose=1)
    predict = np.argmax(predict_probability, axis=1)

    plot_images_labels_prediction(x_images_normalize, y_labels, predict, 0, 10)
    show_predict_probability(y_labels, predict, x_images_normalize, predict_probability, 0)
    print('done')

2.ResNet50

ResNet50

from tensorflow import keras
from tensorflow.keras import layers

INPUT_SIZE = 224
CLASS_NUM = 1000

# stage_name=2,3,4,5;  block_name=a,b,c
def ConvBlock(input_tensor, num_output, stride, stage_name, block_name):
    filter1, filter2, filter3 = num_output

    x = layers.Conv2D(filter1, 1, strides=stride, padding='same', name='res'+stage_name+block_name+'_branch2a')(input_tensor)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2a')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2a_relu')(x)

    x = layers.Conv2D(filter2, 3, strides=(1, 1), padding='same', name='res'+stage_name+block_name+'_branch2b')(x)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2b')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2b_relu')(x)

    x = layers.Conv2D(filter3, 1, strides=(1, 1), padding='same', name='res'+stage_name+block_name+'_branch2c')(x)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2c')(x)

    shortcut = layers.Conv2D(filter3, 1, strides=stride, padding='same', name='res'+stage_name+block_name+'_branch1')(input_tensor)
    shortcut = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch1')(shortcut)

    x = layers.add([x, shortcut], name='res'+stage_name+block_name)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_relu')(x)

    return x

def IdentityBlock(input_tensor, num_output, stage_name, block_name):
    filter1, filter2, filter3 = num_output

    x = layers.Conv2D(filter1, 1, strides=(1, 1), padding='same', name='res'+stage_name+block_name+'_branch2a')(input_tensor)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2a')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2a_relu')(x)

    x = layers.Conv2D(filter2, 3, strides=(1, 1), padding='same', name='res'+stage_name+block_name+'_branch2b')(x)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2b')(x)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_branch2b_relu')(x)

    x = layers.Conv2D(filter3, 1, strides=(1, 1), padding='same', name='res'+stage_name+block_name+'_branch2c')(x)
    x = layers.BatchNormalization(name='bn'+stage_name+block_name+'_branch2c')(x)

    shortcut = input_tensor

    x = layers.add([x, shortcut], name='res'+stage_name+block_name)
    x = layers.Activation('relu', name='res'+stage_name+block_name+'_relu')(x)

    return x

def ResNet50(input_shape, class_num):
    input = keras.Input(shape=input_shape, name='input')

    # conv1
    x = layers.Conv2D(64, 7, strides=(2, 2), padding='same', name='conv1')(input)  # 7×7, 64, stride 2
    x = layers.BatchNormalization(name='bn_conv1')(x)
    x = layers.Activation('relu', name='conv1_relu')(x)
    x = layers.MaxPooling2D((3, 3), strides=2, padding='same', name='pool1')(x)   # 3×3 max pool, stride 2

    # conv2_x
    x = ConvBlock(input_tensor=x, num_output=(64, 64, 256), stride=(1, 1), stage_name='2', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(64, 64, 256), stage_name='2', block_name='b')
    x = IdentityBlock(input_tensor=x, num_output=(64, 64, 256), stage_name='2', block_name='c')

    # conv3_x
    x = ConvBlock(input_tensor=x, num_output=(128, 128, 512), stride=(2, 2), stage_name='3', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(128, 128, 512), stage_name='3', block_name='b')
    x = IdentityBlock(input_tensor=x, num_output=(128, 128, 512), stage_name='3', block_name='c')
    x = IdentityBlock(input_tensor=x, num_output=(128, 128, 512), stage_name='3', block_name='d')

    # conv4_x
    x = ConvBlock(input_tensor=x, num_output=(256, 256, 1024), stride=(2, 2), stage_name='4', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(256, 256, 1024), stage_name='4', block_name='b')
    x = IdentityBlock(input_tensor=x, num_output=(256, 256, 1024), stage_name='4', block_name='c')
    x = IdentityBlock(input_tensor=x, num_output=(256, 256, 1024), stage_name='4', block_name='d')
    x = IdentityBlock(input_tensor=x, num_output=(256, 256, 1024), stage_name='4', block_name='e')
    x = IdentityBlock(input_tensor=x, num_output=(256, 256, 1024), stage_name='4', block_name='f')

    # conv5_x
    x = ConvBlock(input_tensor=x, num_output=(512, 512, 2048), stride=(2, 2), stage_name='5', block_name='a')
    x = IdentityBlock(input_tensor=x, num_output=(512, 512, 2048), stage_name='5', block_name='b')
    x = IdentityBlock(input_tensor=x, num_output=(512, 512, 2048), stage_name='5', block_name='c')

    # average pool, 1000-d fc, softmax
    x = layers.AveragePooling2D((7, 7), strides=(1, 1), name='pool5')(x)
    x = layers.Flatten(name='flatten')(x)
    x = layers.Dense(class_num, activation='softmax', name='fc1000')(x)

    model = keras.Model(input, x, name='resnet50')
    model.summary()
    return model

if __name__ == '__main__':
    model = ResNet50((INPUT_SIZE, INPUT_SIZE, 3), CLASS_NUM)
    print('Done.')
GarryLau
关注
  • 0
    点赞
  • 踩
  • 32
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
专栏目录
ResNet的Keras实现
01-22
VGGNet和GoogLeNet等网络都表明有足够的深度是模型表现良好的前提，但是在网络深度增加到一定程度时，更深的网络意味着更高的训练误差。误差升高的原因是网络越深，梯度弥散[还有梯度爆炸的可能性]的现象就越明显，所以在后向传播的时候，无法有效的把梯度更新到前面的网络层，靠前的网络层参数无法更新，导致训练和测试效果变差。所以ResNet面临的问题是怎样在增加网络深度的情况下有可以有效解决梯度消失的问题。ResNet中解决深层网络梯度消失的问题的核心结构是残差网络。
使用resnet18实现花生图片分类与目标检测
m0_63821003的博客
12-21 1732
)
resnet18实现猫狗分类
m0_54123775的博客
04-18 966
完整数据集共25000张图片。
深度学习基于Resnet18的图像多分类--训练自己的数据集(超详细 含源码)
weixin_49345320的博客
06-22 1万+
深度学习基于Resnet18的图像多分类--训练自己的数据集(超详细 含源码)
Keras实现ResNet残差网络
qq_35565669的博客
07-26 4615
# from keras.models import from keras.layers import * from keras.models import Input from keras import Model import tensorflow as tf import numpy as np def conv2d_bn(input, kernel_num, kernel_size=3, strides=(1, 1), layer_name='', padding_mode='same'...
带残差连接的ResNet18
m0_70026215的博客
11-29 2021
首先，使用带残差连接的ResNet模型相比于不带残差的模型，在训练过程中表现出更好的性能。带残差的模型具有更快的收敛速度、更低的损失和更高的准确率。这证明了残差块确实能够为网络带来性能提升，而无脑堆砌网络层并不能有效地提高模型的性能。这个结果也打破了我一直都认为神经网络越深性能越好的理论认知，同时通过学长的博客我认识到残差连接能够有效地缓解梯度消失问题，减少训练难度，并提高了网络的深度和表达能力。这也算一个小小的收获吧(那一大堆推导我真没看懂！！哭）
【计算机视觉+CNN】keras+ResNet残差网络实现图像识别分类实战（附源码和数据集 超详细）
showswoller的博客
12-07 2750
【计算机视觉】keras+ResNet残差网络实现图像识别分类实战（附源码和数据集 超详细）
keras复现ResNet_18
qq_36346625的博客
05-11 4788
原来复现了ResNet_50。今天再复现ResNet_18的时候才发现原来ResNet_50的复现并不正确。错误地方是论文中的卷积核大小都是（3,3）。而使用keras复现的时候由于卷积层输入输出大小的计算公式不熟悉，误打误撞，虽然复现后的模型plot_model、甚至是model.summary()出来的结果都是一模一样的，但是却不是真正的ResNet_50。今天先正确复现ResNet_18，改天抽空再复现ResNet_50。 ResNet论文获取连接：《Deep Residual Learning fo
基于Keras的ResNet实现
u013093426的博客
07-24 2万+
本文是吴恩达《深度学习》第四课《卷积神经网络》第二周课后题第二部分的实现。 0.ResNet简介 目前神经网络变得越来越复杂，从几层到几十层甚至一百多层的网络都有。深层网络的主要的优势是可以表达非常复杂的函数，它可以从不同抽象程度的层里学到特征，比如在比较低层次里学习到边缘特征而在较高层里可以学到复杂的特征。然而使用深层网络并非总是奏效，因为存在一个非常大的障碍——梯度消失：在非常深层的网络中...
keras的resnet50模型介绍
shuijinghua的博客
07-26 9522
网络的完整结构,可通过下面代码显示在控制台上 import keras_resnet.models import keras shape, classes = (224, 224, 3), 1000 x = keras.layers.Input(shape) model = keras_resnet.models.ResNet50(x) model.summary() 网络源码实现: 下面...
Keras入门与残差网络的搭建
青山渺渺
09-11 8992
发现草稿箱里还有一篇很早之前的学习笔记，希望可以帮助到有需要的童鞋~
使用Keras预训练模型ResNet50进行图像分类方式
09-16
主要介绍了使用Keras预训练模型ResNet50进行图像分类方式，具有很好的参考价值，希望对大家有所帮助。一起跟随小编过来看看吧
基于Keras+Resnet_v1对cifar-10进行图像分类
05-17
利用CNN处理CIFAR-10的测试精度没达到0.9，所以来试试Rsenet~通过数据增强等处理方式，利用20层的Resnet对其进行测试，精度达到0.9139.
深度学习入门示例之使用Keras实现ResNet算法.zip
最新发布
05-10
深度学习入门示例之使用Keras实现ResNet算法
ResNet-50-model.keras.h5
04-22
深度学习的Keras 数据集，因为一些不可描述的原因不能直接下载了。现在分享出来 君可自取之，岁月安好。共同努力~
实验六 卷积神经网络（5）使用预训练resnet18实现CIFAR-10分类
m0_64715292的博客
11-11 3355
预训练resnet18实现CIFAR-10分类
基于ResNet18网络完成图像分类任务
m0_70026215的博客
12-02 2831
CIFAR-10数据集包含了10种不同的类别、共60,000张图像，其中每个类别的图像都是6000张，图像大小均为像素。CIFAR-10数据集的示例如图所示。什么是预训练呢？举例子进行简单的介绍。
Resnet 18网络模型
热门推荐
Chenzhinan1219的博客
12-20 5万+
1. 残差网络：（Resnet） 残差块： 让我们聚焦于神经网络局部：如图左侧所示，假设我们的原始输入为x，而希望学出的理想映射为f(x)（作为上方激活函数的输入）。左图虚线框中的部分需要直接拟合出该映射f(x)，而右图虚线框中的部分则需要拟合出残差映射f(x)−x。 残差映射在现实中往往更容易优化。 以本节开头提到的恒等映射作为我们希望学出的理想映射f(x)，我们只需将右图虚线框内上方的加权运算（如仿射）的权重和偏置参数设成0，那么f(x)即为恒等映射。 实际中，当理想映射f(x)极接近于恒等映..
Keras---使用Keras实现Resnet网络
weixin_42206075的博客
12-14 3675
Keras—使用Keras实现Resnet网络 网络结构 图片引用：https://blog.csdn.net/yao1249736473/article/details/89330791 说明： 1.所谓的“残差网络”，在我个人理解结合参考网上大佬的blog，我大概总结了下：其实说白了，当我们在做简单的维度叠加，使其网络深度足够深的时候，其实越往后作用就越小，反而还会使模型难训练。因为这是属于一种叫做“恒等映射”，但是当我们做残差的时候，就相当于打破了这个“恒等”，使模型可以更好的训练。 2.残差的
resnet 用keras实现 输入50*50
01-19
以下是使用Keras实现ResNet50网络并输入大小为50x50的示例代码： ```python import tensorflow as tf from tensorflow.keras.layers import Input, Conv2D, BatchNormalization, Activation, MaxPooling2D, Add, GlobalAveragePooling2D, Dense from tensorflow.keras.models import Model def resnet_block(inputs, filters, strides=1): x = Conv2D(filters, kernel_size=3, strides=strides, padding='same')(inputs) x = BatchNormalization()(x) x = Activation('relu')(x) x = Conv2D(filters, kernel_size=3, padding='same')(x) x = BatchNormalization()(x) if strides != 1 or inputs.shape[-1] != filters: shortcut = Conv2D(filters, kernel_size=1, strides=strides, padding='same')(inputs) shortcut = BatchNormalization()(shortcut) else: shortcut = inputs x = Add()([x, shortcut]) x = Activation('relu')(x) return x def ResNet50(input_shape=(50,50, 3), num_classes=1000): inputs = Input(shape=input_shape) x = Conv2D(64, kernel_size=7, strides=2, padding='same')(inputs) x = BatchNormalization()(x) x = Activation('relu')(x) x = MaxPooling2D(pool_size=3, strides=2, padding='same')(x) x = resnet_block(x, filters=64) x = resnet_block(x, filters=64) x = resnet_block(x, filters=64) x = resnet_block(x, filters=128, strides=2) x = resnet_block(x, filters=128) x = resnet_block(x, filters=128) x = resnet_block(x, filters=128) x = resnet_block(x, filters=256, strides=2) x = resnet_block(x, filters=256) x = resnet_block(x, filters=256) x = resnet_block(x, filters=256) x = resnet_block(x, filters=256) x = resnet_block(x, filters=256) x = resnet_block(x, filters=512, strides=2) x = resnet_block(x, filters=512) x = resnet_block(x, filters=512) x = GlobalAveragePooling2D()(x) x = Dense(num_classes, activation='softmax')(x) model = Model(inputs=inputs, outputs=x) return model model = ResNet50(input_shape=(50, 50, 3), num_classes=1000) model.summary() ``` 这段代码定义了一个ResNet50网络模型，输入大小为50x50的彩色图像。模型的结构包括卷积层、批归一化层、激活函数层、池化层、残差块、全局平均池化层和全连接层。最后使用`model.summary()`打印出模型的结构信息。

“相关推荐”对你有帮助么？

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值