系统:Win10
环境:Pycharm/Vscode Python3.7
效果图:
部分代码如下:
import torch import torch.nn as nn import torch.nn.functional as F import torch.optim as optim from torchvision import datasets,transforms #定义超参数 BATCH_SIZE = 16 # 每批处理的数据 DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu") # 是否用GPU还是CPU训练 import torch # 检查是否有CUDA支持 if torch.cuda.is_available(): # 加载模型时将存储映射到CPU saved_model_weights = torch.load('trained_model.pth', map_location=torch.device('cpu')) else: # 正常加载模型 saved_model_weights = torch.load('trained_model.pth') EPOCHS = 20# 训练数据集的轮次 # 构建pipeline,对图像做处理 pipeline = transforms.Compose([ transforms.ToTensor(),# 将图片转换成tensor transforms.Normalize((0.1307,),(0.3081,)) # 正则化降低模型复杂度 ]) #下载、加载数据 from torch.utils.data import DataLoader # 下载数据集 pipeline = transforms.Compose([ transforms.Grayscale(num_output_channels=1), transforms.Resize((28, 28)), # 将图像大小调整为 28x28 transforms.ToTensor(),# 将图片转换成tensor transforms.Normalize((0.1307,),(0.3081,)) # 正则化降低模型复杂度 ]) # 加载完整的训练集 train_dataset = datasets.ImageFolder('mnist+', transform=pipeline) # 定义训练集和测试集的比例 train_ratio = 0.8 # 训练集占总体的80% test_ratio = 0.2 # 测试集占总体的20% # 计算划分的大小 train_size = int(train_ratio * len(train_dataset)) test_size = len(train_dataset) - train_size # 使用random_split函数进行划分 train_dataset, test_dataset = torch.utils.data.random_split(train_dataset, [train_size, test_size])