【深度学习实战】p4 猴痘病识别

最新推荐文章于 2024-07-12 16:21:48 发布

Banana0840

最新推荐文章于 2024-07-12 16:21:48 发布

阅读量825

点赞数 22

文章标签：深度学习人工智能

本文链接：https://blog.csdn.net/Banana0840/article/details/138677716

版权

🍨 本文为🔗365天深度学习训练营中的学习记录博客
🍖 原作者：K同学啊

代码


import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms, datasets
from torchsummary import summary

import os,PIL,pathlib

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

device

import os,PIL,random,pathlib

data_dir = './data/p4-data/'
data_dir = pathlib.Path(data_dir)

data_paths = list(data_dir.glob('*'))
classeNames = [str(path).split("\\")[2] for path in data_paths]
classeNames

[‘Monkeypox’, ‘Others’]

total_datadir = './data/p4-data/'

# 关于transforms.Compose的更多介绍可以参考：https://blog.csdn.net/qq_38251616/article/details/124878863
train_transforms = transforms.Compose([
    transforms.Resize([224, 224]),  # 将输入图片resize成统一尺寸
    transforms.ToTensor(),          # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
    transforms.Normalize(           # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
        mean=[0.485, 0.456, 0.406], 
        std=[0.229, 0.224, 0.225])  # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])

total_data = datasets.ImageFolder(total_datadir,transform=train_transforms)

print(total_data)
print(total_data.class_to_idx)

Dataset ImageFolder
Number of datapoints: 2142
Root location: ./data/p4-data/
StandardTransform
Transform: Compose(
Resize(size=[224, 224], interpolation=bilinear, max_size=None, antialias=None)
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)
{‘Monkeypox’: 0, ‘Others’: 1}

image_in_shape = total_data[0][0].shape
image_in_shape

image_in_shape = total_data[0][0].shape
image_in_shape

train_size = int(0.8 * len(total_data))
test_size  = len(total_data) - train_size
train_dataset, test_dataset = torch.utils.data.random_split(total_data, [train_size, test_size])
print(train_dataset, test_dataset)
print(train_size,test_size)

<torch.utils.data.dataset.Subset object at 0x000002D1F15DCD60> <torch.utils.data.dataset.Subset object at 0x000002D1F15DC1F0>
1713 429

import torch.nn.functional as F

class Network_bn(nn.Module):
    def __init__(self):
        super(Network_bn, self).__init__()
        """
        nn.Conv2d()函数：
        第一个参数（in_channels）是输入的channel数量
        第二个参数（out_channels）是输出的channel数量
        第三个参数（kernel_size）是卷积核大小
        第四个参数（stride）是步长，默认为1
        第五个参数（padding）是填充大小，默认为0
        """
        self.conv11 = nn.Conv2d(in_channels=3, out_channels=12, kernel_size=5, stride=1, padding=0)
        self.bn11 = nn.BatchNorm2d(12)
        self.conv12 = nn.Conv2d(in_channels=12, out_channels=12, kernel_size=3, stride=1, padding=0)
        self.bn12 = nn.BatchNorm2d(12)
        self.conv2 = nn.Conv2d(in_channels=12, out_channels=12, kernel_size=3, stride=1, padding=0)
        self.bn2 = nn.BatchNorm2d(12)
        self.pool = nn.MaxPool2d(2,2)
        self.conv4 = nn.Conv2d(in_channels=12, out_channels=24, kernel_size=5, stride=1, padding=0)
        self.bn4 = nn.BatchNorm2d(24)
        self.conv5 = nn.Conv2d(in_channels=24, out_channels=24, kernel_size=5, stride=1, padding=0)
        self.bn5 = nn.BatchNorm2d(24)
        self.fc1 = nn.Linear(24*50*50, len(classeNames))

    def forward(self, x):
        x = F.relu(self.bn11(self.conv11(x)))
        x = F.relu(self.bn12(self.conv12(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = self.pool(x)
        x = F.relu(self.bn4(self.conv4(x)))
        x = F.relu(self.bn5(self.conv5(x)))
        x = self.pool(x)
        x = x.view(-1, 24*50*50)
        x = self.fc1(x)

        return x

device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))

model = Network_bn().to(device)

print('Image in shape is: %s' %{image_in_shape})

print(summary(model, image_in_shape))

输出：

Using cuda device
Image in shape is: {torch.Size([3, 224, 224])}
==========================================================================================
Layer (type:depth-idx)                   Output Shape              Param #
==========================================================================================
├─Conv2d: 1-1                            [-1, 12, 220, 220]        912
├─BatchNorm2d: 1-2                       [-1, 12, 220, 220]        24
├─Conv2d: 1-3                            [-1, 12, 218, 218]        1,308
├─BatchNorm2d: 1-4                       [-1, 12, 218, 218]        24
├─Conv2d: 1-5                            [-1, 12, 216, 216]        1,308
├─BatchNorm2d: 1-6                       [-1, 12, 216, 216]        24
├─MaxPool2d: 1-7                         [-1, 12, 108, 108]        --
├─Conv2d: 1-8                            [-1, 24, 104, 104]        7,224
├─BatchNorm2d: 1-9                       [-1, 24, 104, 104]        48
├─Conv2d: 1-10                           [-1, 24, 100, 100]        14,424
├─BatchNorm2d: 1-11                      [-1, 24, 100, 100]        48
├─MaxPool2d: 1-12                        [-1, 24, 50, 50]          --
├─Linear: 1-13                           [-1, 2]                   120,002
==========================================================================================
Total params: 145,346
Trainable params: 145,346
Non-trainable params: 0
Total mult-adds (M): 387.61
==========================================================================================
Input size (MB): 0.57
...
Forward/backward pass size (MB): 33.73
Params size (MB): 0.55
Estimated Total Size (MB): 34.86
==========================================================================================

loss_fn    = nn.CrossEntropyLoss() # 创建损失函数
learn_rate = 7e-5 # 学习率
opt        = torch.optim.Adam(model.parameters(),lr=learn_rate)
batch_size = 48
epochs     = 50

from torch.optim.lr_scheduler import LambdaLR
import numpy as np

lr_lambda = lambda epoch: 1.0 if epoch < 100 else np.exp(0.1 * (10 - epoch))
scheduler = LambdaLR(opt, lr_lambda, last_epoch=-1)

# 训练循环
def train(dataloader, model, loss_fn, optimizer):
    size = len(dataloader.dataset)  # 训练集的大小，一共60000张图片
    num_batches = len(dataloader)  # 批次数目，1875（60000/32）

    train_loss, train_acc = 0, 0  # 初始化训练损失和正确率

    for X, y in dataloader:  # 获取图片及其标签
        X, y = X.to(device), y.to(device)

        # 计算预测误差
        pred = model(X)  # 网络输出
        loss = loss_fn(pred, y)  # 计算网络输出和真实值之间的差距，targets为真实值，计算二者差值即为损失

        # 反向传播
        optimizer.zero_grad()  # grad属性归零
        loss.backward()  # 反向传播
        optimizer.step()  # 每一步自动更新

        # 记录acc与loss
        train_acc += (pred.argmax(1) == y).type(torch.float).sum().item()
        train_loss += loss.item()

    train_acc /= size
    train_loss /= num_batches

    return train_acc, train_loss


def test(dataloader, model, loss_fn):
    size = len(dataloader.dataset)  # 测试集的大小，一共10000张图片
    num_batches = len(dataloader)  # 批次数目，313（10000/32=312.5，向上取整）
    test_loss, test_acc = 0, 0

    # 当不进行训练时，停止梯度更新，节省计算内存消耗
    with torch.no_grad():
        for imgs, target in dataloader:
            imgs, target = imgs.to(device), target.to(device)

            # 计算loss
            target_pred = model(imgs)
            loss = loss_fn(target_pred, target)

            test_loss += loss.item()
            test_acc += (target_pred.argmax(1) == target).type(torch.float).sum().item()

    test_acc /= size
    test_loss /= num_batches

    return test_acc, test_loss


train_loss = []
train_acc = []
test_loss = []
test_acc = []


train_dl = torch.utils.data.DataLoader(train_dataset,
                                           batch_size=batch_size,
                                           shuffle=True,
                                           num_workers=1)
test_dl = torch.utils.data.DataLoader(test_dataset,
                                          batch_size=batch_size,
                                          shuffle=True,
                                          num_workers=1)

for X, y in test_dl:
    print("Shape of X [N, C, H, W]: ", X.shape)
    print("Shape of y: ", y.shape, y.dtype)
    break

def scheduler_lr(optimizer, scheduler):
    lr_history = []

    """optimizer的更新在scheduler更新的前面"""
    for epoch in range(epochs):
        optimizer.step() # 更新参数
        lr_history.append(optimizer.param_groups[0]['lr'])
        # print(optimizer.param_groups[0]['lr'])
        scheduler.step() # 调整学习率
    return lr_history

lr_history = scheduler_lr(opt, scheduler)
print(lr_history)

if __name__ == '__main__':
    for epoch in range(epochs):
        model.train()
        epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, opt)

        model.eval()
        epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)

        train_acc.append(epoch_train_acc)
        train_loss.append(epoch_train_loss)
        test_acc.append(epoch_test_acc)
        test_loss.append(epoch_test_loss)

        template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%，Test_loss:{:.3f}')
        print(
            template.format(epoch + 1, epoch_train_acc * 100, epoch_train_loss, epoch_test_acc * 100, epoch_test_loss))
    print('Done')

输出：

Shape of X [N, C, H, W]:  torch.Size([48, 3, 224, 224])
Shape of y:  torch.Size([48]) torch.int64
[7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05]
Epoch: 1, Train_acc:61.9%, Train_loss:0.704, Test_acc:76.5%，Test_loss:0.542
Epoch: 2, Train_acc:78.1%, Train_loss:0.463, Test_acc:83.2%，Test_loss:0.423
Epoch: 3, Train_acc:86.3%, Train_loss:0.361, Test_acc:79.3%，Test_loss:0.464
Epoch: 4, Train_acc:86.6%, Train_loss:0.327, Test_acc:86.0%，Test_loss:0.336
Epoch: 5, Train_acc:91.8%, Train_loss:0.244, Test_acc:87.4%，Test_loss:0.301
Epoch: 6, Train_acc:94.9%, Train_loss:0.191, Test_acc:87.4%，Test_loss:0.289
Epoch: 7, Train_acc:95.5%, Train_loss:0.168, Test_acc:90.0%，Test_loss:0.274
Epoch: 8, Train_acc:96.7%, Train_loss:0.142, Test_acc:90.2%，Test_loss:0.239
Epoch: 9, Train_acc:97.5%, Train_loss:0.123, Test_acc:90.4%，Test_loss:0.244
Epoch:10, Train_acc:98.3%, Train_loss:0.104, Test_acc:90.0%，Test_loss:0.235
Epoch:11, Train_acc:98.1%, Train_loss:0.103, Test_acc:86.9%，Test_loss:0.277
Epoch:12, Train_acc:98.4%, Train_loss:0.088, Test_acc:89.3%，Test_loss:0.252
Epoch:13, Train_acc:99.5%, Train_loss:0.064, Test_acc:90.9%，Test_loss:0.221
Epoch:14, Train_acc:99.4%, Train_loss:0.062, Test_acc:90.7%，Test_loss:0.220
Epoch:15, Train_acc:99.7%, Train_loss:0.055, Test_acc:90.4%，Test_loss:0.216
Epoch:16, Train_acc:99.9%, Train_loss:0.046, Test_acc:90.9%，Test_loss:0.239
Epoch:17, Train_acc:99.9%, Train_loss:0.039, Test_acc:91.6%，Test_loss:0.210
Epoch:18, Train_acc:99.9%, Train_loss:0.038, Test_acc:91.1%，Test_loss:0.216
Epoch:19, Train_acc:99.9%, Train_loss:0.039, Test_acc:87.4%，Test_loss:0.277
Epoch:20, Train_acc:100.0%, Train_loss:0.028, Test_acc:90.2%，Test_loss:0.217
Epoch:21, Train_acc:99.9%, Train_loss:0.029, Test_acc:90.2%，Test_loss:0.232
Epoch:22, Train_acc:99.9%, Train_loss:0.024, Test_acc:90.7%，Test_loss:0.208
Epoch:23, Train_acc:99.9%, Train_loss:0.022, Test_acc:92.1%，Test_loss:0.204
Epoch:24, Train_acc:100.0%, Train_loss:0.019, Test_acc:90.9%，Test_loss:0.215
Epoch:25, Train_acc:99.9%, Train_loss:0.021, Test_acc:90.9%，Test_loss:0.215
Epoch:26, Train_acc:99.9%, Train_loss:0.020, Test_acc:90.4%，Test_loss:0.253
Epoch:27, Train_acc:99.8%, Train_loss:0.020, Test_acc:88.8%，Test_loss:0.287
Epoch:28, Train_acc:100.0%, Train_loss:0.018, Test_acc:90.9%，Test_loss:0.215
Epoch:29, Train_acc:100.0%, Train_loss:0.016, Test_acc:90.9%，Test_loss:0.212
Epoch:30, Train_acc:100.0%, Train_loss:0.013, Test_acc:90.9%，Test_loss:0.232
Epoch:31, Train_acc:100.0%, Train_loss:0.012, Test_acc:91.1%，Test_loss:0.208
Epoch:32, Train_acc:100.0%, Train_loss:0.010, Test_acc:91.4%，Test_loss:0.213
Epoch:33, Train_acc:99.9%, Train_loss:0.014, Test_acc:91.1%，Test_loss:0.222
Epoch:34, Train_acc:100.0%, Train_loss:0.010, Test_acc:90.7%，Test_loss:0.226
Epoch:35, Train_acc:100.0%, Train_loss:0.010, Test_acc:90.0%，Test_loss:0.263
Epoch:36, Train_acc:100.0%, Train_loss:0.010, Test_acc:89.7%，Test_loss:0.231
Epoch:37, Train_acc:100.0%, Train_loss:0.010, Test_acc:90.9%，Test_loss:0.223
Epoch:38, Train_acc:100.0%, Train_loss:0.009, Test_acc:90.2%，Test_loss:0.221
Epoch:39, Train_acc:100.0%, Train_loss:0.008, Test_acc:91.1%，Test_loss:0.228
Epoch:40, Train_acc:100.0%, Train_loss:0.007, Test_acc:91.4%，Test_loss:0.224
Epoch:41, Train_acc:99.9%, Train_loss:0.009, Test_acc:90.9%，Test_loss:0.232
Epoch:42, Train_acc:100.0%, Train_loss:0.007, Test_acc:89.3%，Test_loss:0.244
Epoch:43, Train_acc:100.0%, Train_loss:0.006, Test_acc:91.4%，Test_loss:0.220
Epoch:44, Train_acc:100.0%, Train_loss:0.006, Test_acc:91.1%，Test_loss:0.220
Epoch:45, Train_acc:100.0%, Train_loss:0.005, Test_acc:90.7%，Test_loss:0.220
Epoch:46, Train_acc:100.0%, Train_loss:0.006, Test_acc:90.9%，Test_loss:0.214
Epoch:47, Train_acc:100.0%, Train_loss:0.006, Test_acc:91.1%，Test_loss:0.224
Epoch:48, Train_acc:100.0%, Train_loss:0.005, Test_acc:90.2%，Test_loss:0.237
Epoch:49, Train_acc:100.0%, Train_loss:0.005, Test_acc:90.7%，Test_loss:0.223
Epoch:50, Train_acc:100.0%, Train_loss:0.005, Test_acc:90.0%，Test_loss:0.229
Done

import matplotlib.pyplot as plt
import warnings

warnings.filterwarnings("ignore")               #忽略警告信息
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
plt.rcParams['figure.dpi']         = 100        #分辨率

epochs_range = range(epochs)

plt.figure(figsize=(12, 3))
plt.subplot(1, 3, 1)

plt.axhline(y=0.9, color='r', linestyle='--', linewidth=2)
plt.plot(epochs_range, train_acc, label='Training Accuracy')
plt.plot(epochs_range, test_acc, label='Test Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')

plt.subplot(1, 3, 2)
plt.plot(epochs_range, train_loss, label='Training Loss')
plt.plot(epochs_range, test_loss, label='Test Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')

plt.subplot(1, 3, 3)
plt.plot(epochs_range, lr_history, label='Learning Rate')
plt.legend(loc='upper right')
plt.title('lr')

plt.show()

在这里插入图片描述

# 模型保存
PATH = './p4_model.pth'  # 保存的参数文件名
torch.save(model.state_dict(), PATH)

# 将参数加载到model当中
model.load_state_dict(torch.load(PATH, map_location=device))

from PIL import Image 

classes = list(total_data.class_to_idx)

def predict_one_image(image_path, model, transform, classes):
    
    test_img = Image.open(image_path).convert('RGB')
    # plt.imshow(test_img)  # 展示预测的图片

    test_img = transform(test_img)
    img = test_img.to(device).unsqueeze(0)
    
    model.eval()
    output = model(img)

    _,pred = torch.max(output,1)
    pred_class = classes[pred]
    print(f'预测结果是：{pred_class}')

# 预测训练集中的某张照片
predict_one_image(image_path='./data/p4-data/Monkeypox/M01_01_00.jpg', 
                  model=model, 
                  transform=train_transforms, 
                  classes=classes)

预测结果是：Monkeypox

训练记录

learn_rate = 1e-4 恒定学习率 epochs=25 batch_size = 32 SGD

learn_rate = 1e-4 恒定学习率 epochs=25 batch_size = 16 SGD

learn_rate = 1e-4 恒定学习率 epochs=50 batch_size = 16 SGD

learn_rate = 1e-4 恒定学习率 epochs=50 batch_size = 32 SGD

learn_rate = 5e-5 恒定学习率 epochs=50 batch_size = 32 SGD

learn_rate = 5e-5 恒定学习率 epochs=70 batch_size = 32 SGD

learn_rate = 5e-5 恒定学习率 epochs=50 batch_size = 64 SGD

调整网络结构，test_accuracy达到85%
conv1 55 -> conv11 33 conv12 3*3

Image in shape is: {torch.Size([3, 224, 224])}
==========================================================================================
Layer (type:depth-idx)                   Output Shape              Param #
==========================================================================================
├─Conv2d: 1-1                            [-1, 12, 222, 222]        336
├─BatchNorm2d: 1-2                       [-1, 12, 222, 222]        24
├─Conv2d: 1-3                            [-1, 12, 220, 220]        1,308
├─BatchNorm2d: 1-4                       [-1, 12, 220, 220]        24
├─Conv2d: 1-5                            [-1, 12, 218, 218]        1,308
├─BatchNorm2d: 1-6                       [-1, 12, 218, 218]        24
├─MaxPool2d: 1-7                         [-1, 12, 109, 109]        --
├─Conv2d: 1-8                            [-1, 24, 105, 105]        7,224
├─BatchNorm2d: 1-9                       [-1, 24, 105, 105]        48
├─Conv2d: 1-10                           [-1, 24, 101, 101]        14,424
├─BatchNorm2d: 1-11                      [-1, 24, 101, 101]        48
├─MaxPool2d: 1-12                        [-1, 24, 50, 50]          --
├─Linear: 1-13                           [-1, 2]                   120,002
==========================================================================================
Total params: 144,770
Trainable params: 144,770
Non-trainable params: 0
Total mult-adds (M): 366.68
==========================================================================================
Input size (MB): 0.57
...
Forward/backward pass size (MB): 34.36
Params size (MB): 0.55
Estimated Total Size (MB): 35.49
==========================================================================================

learn_rate = 5e-5 恒定学习率 epochs=50 batch_size = 64 SGD

conv1 55 -> conv11 33 conv12 33
conv2 55 -> conv21 33 conv22 33

learn_rate = 1e-4 恒定学习率 epochs=50 batch_size = 64 SGD

conv1 55 -> conv11 33 conv12 33
conv2 55 -> conv21 33 conv22 33

Shape of X [N, C, H, W]:  torch.Size([48, 3, 224, 224])
Shape of y:  torch.Size([48]) torch.int64
[0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001]
Epoch: 1, Train_acc:59.0%, Train_loss:0.689, Test_acc:63.2%，Test_loss:0.641
Epoch: 2, Train_acc:65.1%, Train_loss:0.627, Test_acc:67.6%，Test_loss:0.619
Epoch: 3, Train_acc:70.1%, Train_loss:0.578, Test_acc:70.2%，Test_loss:0.575
Epoch: 4, Train_acc:71.8%, Train_loss:0.539, Test_acc:74.1%，Test_loss:0.557
Epoch: 5, Train_acc:73.5%, Train_loss:0.531, Test_acc:72.0%，Test_loss:0.564
Epoch: 6, Train_acc:76.6%, Train_loss:0.493, Test_acc:76.7%，Test_loss:0.522
Epoch: 7, Train_acc:78.8%, Train_loss:0.476, Test_acc:69.0%，Test_loss:0.604
Epoch: 8, Train_acc:79.7%, Train_loss:0.449, Test_acc:77.9%，Test_loss:0.494
Epoch: 9, Train_acc:82.0%, Train_loss:0.429, Test_acc:80.7%，Test_loss:0.477
Epoch:10, Train_acc:84.5%, Train_loss:0.409, Test_acc:81.4%，Test_loss:0.468
Epoch:11, Train_acc:83.6%, Train_loss:0.407, Test_acc:80.7%，Test_loss:0.456
Epoch:12, Train_acc:85.9%, Train_loss:0.386, Test_acc:81.6%，Test_loss:0.452
Epoch:13, Train_acc:86.7%, Train_loss:0.376, Test_acc:82.1%，Test_loss:0.442
Epoch:14, Train_acc:87.4%, Train_loss:0.361, Test_acc:80.9%，Test_loss:0.451
Epoch:15, Train_acc:88.0%, Train_loss:0.354, Test_acc:81.6%，Test_loss:0.430
Epoch:16, Train_acc:89.6%, Train_loss:0.341, Test_acc:83.7%，Test_loss:0.422
Epoch:17, Train_acc:89.8%, Train_loss:0.329, Test_acc:84.4%，Test_loss:0.413
Epoch:18, Train_acc:89.8%, Train_loss:0.325, Test_acc:83.4%，Test_loss:0.411
Epoch:19, Train_acc:89.1%, Train_loss:0.319, Test_acc:84.1%，Test_loss:0.405
Epoch:20, Train_acc:90.3%, Train_loss:0.311, Test_acc:84.6%，Test_loss:0.403
Epoch:21, Train_acc:90.1%, Train_loss:0.309, Test_acc:82.8%，Test_loss:0.412
Epoch:22, Train_acc:90.0%, Train_loss:0.297, Test_acc:85.8%，Test_loss:0.384
Epoch:23, Train_acc:92.2%, Train_loss:0.290, Test_acc:85.3%，Test_loss:0.387
Epoch:24, Train_acc:92.1%, Train_loss:0.283, Test_acc:85.8%，Test_loss:0.378
Epoch:25, Train_acc:91.9%, Train_loss:0.279, Test_acc:86.2%，Test_loss:0.374
Epoch:26, Train_acc:92.6%, Train_loss:0.269, Test_acc:86.9%，Test_loss:0.370
Epoch:27, Train_acc:92.4%, Train_loss:0.266, Test_acc:86.5%，Test_loss:0.369
Epoch:28, Train_acc:93.1%, Train_loss:0.257, Test_acc:86.2%，Test_loss:0.360
Epoch:29, Train_acc:92.5%, Train_loss:0.259, Test_acc:86.2%，Test_loss:0.357
Epoch:30, Train_acc:93.3%, Train_loss:0.247, Test_acc:86.0%，Test_loss:0.366
Epoch:31, Train_acc:93.6%, Train_loss:0.246, Test_acc:86.9%，Test_loss:0.361
Epoch:32, Train_acc:93.9%, Train_loss:0.242, Test_acc:86.5%，Test_loss:0.349
Epoch:33, Train_acc:93.4%, Train_loss:0.238, Test_acc:88.1%，Test_loss:0.352
Epoch:34, Train_acc:94.2%, Train_loss:0.231, Test_acc:87.9%，Test_loss:0.340
Epoch:35, Train_acc:93.8%, Train_loss:0.232, Test_acc:87.9%，Test_loss:0.340
Epoch:36, Train_acc:94.0%, Train_loss:0.228, Test_acc:86.9%，Test_loss:0.340
Epoch:37, Train_acc:94.2%, Train_loss:0.221, Test_acc:86.9%，Test_loss:0.335
Epoch:38, Train_acc:94.5%, Train_loss:0.217, Test_acc:88.8%，Test_loss:0.333
Epoch:39, Train_acc:95.0%, Train_loss:0.209, Test_acc:88.8%，Test_loss:0.326
Epoch:40, Train_acc:95.3%, Train_loss:0.206, Test_acc:90.0%，Test_loss:0.325
Epoch:41, Train_acc:94.3%, Train_loss:0.208, Test_acc:89.5%，Test_loss:0.329
Epoch:42, Train_acc:94.6%, Train_loss:0.204, Test_acc:88.3%，Test_loss:0.321
Epoch:43, Train_acc:95.5%, Train_loss:0.203, Test_acc:89.3%，Test_loss:0.324
Epoch:44, Train_acc:95.6%, Train_loss:0.197, Test_acc:88.1%，Test_loss:0.322
Epoch:45, Train_acc:94.9%, Train_loss:0.199, Test_acc:88.3%，Test_loss:0.314
Epoch:46, Train_acc:95.2%, Train_loss:0.198, Test_acc:88.8%，Test_loss:0.318
Epoch:47, Train_acc:95.2%, Train_loss:0.189, Test_acc:88.3%，Test_loss:0.310
Epoch:48, Train_acc:96.0%, Train_loss:0.186, Test_acc:88.6%，Test_loss:0.308
Epoch:49, Train_acc:95.7%, Train_loss:0.186, Test_acc:89.5%，Test_loss:0.313
Epoch:50, Train_acc:95.7%, Train_loss:0.184, Test_acc:88.6%，Test_loss:0.306
Done

learn_rate = 1e-4 恒定学习率 epochs=50 batch_size = 32 SGD

conv1 55 -> conv11 33 conv12 33
conv2 55 -> conv21 33 conv22 33

learn_rate = 1e-4 恒定学习率 epochs=70 batch_size = 64 SGD

conv1 55 -> conv11 33 conv12 33
conv2 55 -> conv21 33 conv22 33

learn_rate = 1e-4 恒定学习率 epochs=50 batch_size = 48 SGD

conv1 55 -> conv11 33 conv12 33
conv2 55 -> conv21 33 conv22 33
conv4 55 -> conv41 33 conv42 3*3

learn_rate = 7e-5 恒定学习率 epochs=50 batch_size = 48 Adam

conv1 55 -> conv11 33 conv12 3*3

Shape of X [N, C, H, W]:  torch.Size([48, 3, 224, 224])
Shape of y:  torch.Size([48]) torch.int64
[7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05, 7e-05]
Epoch: 1, Train_acc:61.9%, Train_loss:0.704, Test_acc:76.5%，Test_loss:0.542
Epoch: 2, Train_acc:78.1%, Train_loss:0.463, Test_acc:83.2%，Test_loss:0.423
Epoch: 3, Train_acc:86.3%, Train_loss:0.361, Test_acc:79.3%，Test_loss:0.464
Epoch: 4, Train_acc:86.6%, Train_loss:0.327, Test_acc:86.0%，Test_loss:0.336
Epoch: 5, Train_acc:91.8%, Train_loss:0.244, Test_acc:87.4%，Test_loss:0.301
Epoch: 6, Train_acc:94.9%, Train_loss:0.191, Test_acc:87.4%，Test_loss:0.289
Epoch: 7, Train_acc:95.5%, Train_loss:0.168, Test_acc:90.0%，Test_loss:0.274
Epoch: 8, Train_acc:96.7%, Train_loss:0.142, Test_acc:90.2%，Test_loss:0.239
Epoch: 9, Train_acc:97.5%, Train_loss:0.123, Test_acc:90.4%，Test_loss:0.244
Epoch:10, Train_acc:98.3%, Train_loss:0.104, Test_acc:90.0%，Test_loss:0.235
Epoch:11, Train_acc:98.1%, Train_loss:0.103, Test_acc:86.9%，Test_loss:0.277
Epoch:12, Train_acc:98.4%, Train_loss:0.088, Test_acc:89.3%，Test_loss:0.252
Epoch:13, Train_acc:99.5%, Train_loss:0.064, Test_acc:90.9%，Test_loss:0.221
Epoch:14, Train_acc:99.4%, Train_loss:0.062, Test_acc:90.7%，Test_loss:0.220
Epoch:15, Train_acc:99.7%, Train_loss:0.055, Test_acc:90.4%，Test_loss:0.216
Epoch:16, Train_acc:99.9%, Train_loss:0.046, Test_acc:90.9%，Test_loss:0.239
Epoch:17, Train_acc:99.9%, Train_loss:0.039, Test_acc:91.6%，Test_loss:0.210
Epoch:18, Train_acc:99.9%, Train_loss:0.038, Test_acc:91.1%，Test_loss:0.216
Epoch:19, Train_acc:99.9%, Train_loss:0.039, Test_acc:87.4%，Test_loss:0.277
Epoch:20, Train_acc:100.0%, Train_loss:0.028, Test_acc:90.2%，Test_loss:0.217
Epoch:21, Train_acc:99.9%, Train_loss:0.029, Test_acc:90.2%，Test_loss:0.232
Epoch:22, Train_acc:99.9%, Train_loss:0.024, Test_acc:90.7%，Test_loss:0.208
Epoch:23, Train_acc:99.9%, Train_loss:0.022, Test_acc:92.1%，Test_loss:0.204
Epoch:24, Train_acc:100.0%, Train_loss:0.019, Test_acc:90.9%，Test_loss:0.215
Epoch:25, Train_acc:99.9%, Train_loss:0.021, Test_acc:90.9%，Test_loss:0.215
Epoch:26, Train_acc:99.9%, Train_loss:0.020, Test_acc:90.4%，Test_loss:0.253
Epoch:27, Train_acc:99.8%, Train_loss:0.020, Test_acc:88.8%，Test_loss:0.287
Epoch:28, Train_acc:100.0%, Train_loss:0.018, Test_acc:90.9%，Test_loss:0.215
Epoch:29, Train_acc:100.0%, Train_loss:0.016, Test_acc:90.9%，Test_loss:0.212
Epoch:30, Train_acc:100.0%, Train_loss:0.013, Test_acc:90.9%，Test_loss:0.232
Epoch:31, Train_acc:100.0%, Train_loss:0.012, Test_acc:91.1%，Test_loss:0.208
Epoch:32, Train_acc:100.0%, Train_loss:0.010, Test_acc:91.4%，Test_loss:0.213
Epoch:33, Train_acc:99.9%, Train_loss:0.014, Test_acc:91.1%，Test_loss:0.222
Epoch:34, Train_acc:100.0%, Train_loss:0.010, Test_acc:90.7%，Test_loss:0.226
Epoch:35, Train_acc:100.0%, Train_loss:0.010, Test_acc:90.0%，Test_loss:0.263
Epoch:36, Train_acc:100.0%, Train_loss:0.010, Test_acc:89.7%，Test_loss:0.231
Epoch:37, Train_acc:100.0%, Train_loss:0.010, Test_acc:90.9%，Test_loss:0.223
Epoch:38, Train_acc:100.0%, Train_loss:0.009, Test_acc:90.2%，Test_loss:0.221
Epoch:39, Train_acc:100.0%, Train_loss:0.008, Test_acc:91.1%，Test_loss:0.228
Epoch:40, Train_acc:100.0%, Train_loss:0.007, Test_acc:91.4%，Test_loss:0.224
Epoch:41, Train_acc:99.9%, Train_loss:0.009, Test_acc:90.9%，Test_loss:0.232
Epoch:42, Train_acc:100.0%, Train_loss:0.007, Test_acc:89.3%，Test_loss:0.244
Epoch:43, Train_acc:100.0%, Train_loss:0.006, Test_acc:91.4%，Test_loss:0.220
Epoch:44, Train_acc:100.0%, Train_loss:0.006, Test_acc:91.1%，Test_loss:0.220
Epoch:45, Train_acc:100.0%, Train_loss:0.005, Test_acc:90.7%，Test_loss:0.220
Epoch:46, Train_acc:100.0%, Train_loss:0.006, Test_acc:90.9%，Test_loss:0.214
Epoch:47, Train_acc:100.0%, Train_loss:0.006, Test_acc:91.1%，Test_loss:0.224
Epoch:48, Train_acc:100.0%, Train_loss:0.005, Test_acc:90.2%，Test_loss:0.237
Epoch:49, Train_acc:100.0%, Train_loss:0.005, Test_acc:90.7%，Test_loss:0.223
Epoch:50, Train_acc:100.0%, Train_loss:0.005, Test_acc:90.0%，Test_loss:0.229
Done

知识点记录

BatchNorm2d

用于进行批归一化（Batch Normalization）操作。它通常应用于卷积神经网络（CNN）的卷积层之后或全连接层之前，用于规范化输入数据并加速网络的训练过程。
批归一化是一种常用的技术，旨在解决深度神经网络训练中的梯度消失和梯度爆炸问题，并提高网络的收敛速度和稳定性。批归一化通过对每个小批量样本的特征进行规范化，使得特征的均值接近0，方差接近1。这样可以使得输入数据分布更加稳定，有助于缓解梯度问题，提高网络的泛化能力。
BatchNorm2d层是应用于二维数据（例如图像）的批归一化操作。它在每个通道的特征图上进行归一化，并通过学习可学习的参数来调整规范化的结果。具体而言，对于每个通道，BatchNorm2d计算该通道上所有样本的均值和方差，并使用这些统计量来对该通道上的特征进行规范化。这样可以保持不同样本之间的特征分布的一致性。
例：

import torch
import torch.nn as nn

batchnorm = nn.BatchNorm2d(2)
input_data = torch.Tensor([
    [[[1,2],[3,4]],[[100,102],[4,3]]],
    [[[2,3],[4,5]],[[101,103],[5,3]]]
    ]
)
print (input_data)
output = batchnorm(input_data)
print (output)

tensor([[[[  1.,   2.],
          [  3.,   4.]],

         [[100., 102.],
          [  4.,   3.]]],


        [[[  2.,   3.],
          [  4.,   5.]],

         [[101., 103.],
          [  5.,   3.]]]])
tensor([[[[-1.6330, -0.8165],
          [ 0.0000,  0.8165]],

         [[ 0.9691,  1.0100],
          [-0.9947, -1.0151]]],


        [[[-0.8165,  0.0000],
          [ 0.8165,  1.6330]],

         [[ 0.9896,  1.0305],
          [-0.9742, -1.0151]]]], grad_fn=<NativeBatchNormBackward0>)

关于卷积核的思考

一直在思考这个问题，我们只设置了卷积核的size和channel，但是卷积核里面的具体值是多少是我们没有设置过的。
我们都知道有不同的卷积核可以用来识别不同的特征，但是在cnn的过程中，这个卷积核的变化是我们看不到的。
暂时没找到太好的解释，引用一下gpt的回答：

结论

batch_size对于test_accuracy有什么影响
batch_size越小，容易震荡越大，其实理解也很容易，小批量数据更容易会不均匀，所以适当增大较好
如何分阶段配置学习率
下次一定

参考：

https://blog.csdn.net/weixin_44943389/article/details/131281942

一文弄懂BatchNorm1d和BatchNorm2d
https://blog.csdn.net/bigkaimyc/article/details/136648815

Banana0840

关注

22
点赞
踩
30

收藏

觉得还不错? 一键收藏
0
评论
【深度学习实战】p4 猴痘病识别

具体而言，对于每个通道，BatchNorm2d计算该通道上所有样本的均值和方差，并使用这些统计量来对该通道上的特征进行规范化。这样可以保持不同样本之间的特征分布的一致性。是一种常用的技术，旨在解决深度神经网络训练中的梯度消失和梯度爆炸问题，并提高网络的收敛速度和稳定性。一直在思考这个问题，我们只设置了卷积核的size和channel，但是卷积核里面的具体值是多少是我们没有设置过的。我们都知道有不同的卷积核可以用来识别不同的特征，但是在cnn的过程中，这个卷积核的变化是我们看不到的。
复制链接

扫一扫