图像分类是根据图像的语义信息将不同类别图像区分开来，是计算机视觉中重要的基本问题
猫狗分类属于图像分类中的粗粒度分类问题
# 使用2.0.0以上PaddlePaddle框架 import paddle print(paddle.__version__)
2.0.1
首先导入必要的包
zipfile------------->python的模块，解压缩zip文件
os------------->python的模块，可使用该模块对操作系统进行操作
paddle--->PaddlePaddle深度学习框架
os------------->python的模块，可使用该模块对操作系统进行操作
numpy---------->python第三方库，用于进行科学计算
PIL------------> Python Image Library,python第三方图像处理库
matplotlib----->python的绘图库 pyplot:matplotlib的绘图框架
os------------->提供了丰富的方法来处理文件和目录
sys------------->供对解释器使用或维护的一些变量的访问，以及与解释器强烈交互的函数。
pickle---------->模块实现了基本的数据序列和反序列化
warnings.filterwarnings("ignore")---------->忽略所有警告
cpu_count---------->获取计算机cpu核数
# 导入需要的包 import warnings warnings.filterwarnings("ignore") import tarfile import paddle import numpy as np from PIL import Image import sys import pickle from multiprocessing import cpu_count import matplotlib.pyplot as plt import os from paddle.nn import MaxPool2D,Conv2D,BatchNorm from paddle.nn import Linear print("本教程基于Paddle的版本号为："+paddle.__version__)
本教程基于Paddle的版本号为：2.0.1
''' 参数配置 ''' train_parameters = { "input_size": [1, 28, 28], #输入图片的shape "class_dim": 2, #分类数 "src_path":"data/data9154/cifar-10-python.tar.gz", #原始数据集路径 "target_path":"/home/aistudio/data/", #要解压的路径 "num_epochs": 10, #训练轮数 "train_batch_size": 100, #训练时每个批次的大小 "learning_strategy": { #优化函数相关的配置 "lr": 0.001 #超参数学习率 }, 'skip_steps': 5, #每N个批次打印一次结果 'save_steps': 5, #每N个批次保存一次模型参数 "checkpoints": "/home/aistudio/checkpoints" #保存的路径 }
Step1:准备数据
- （1）解压原始数据集
- （2）构造dataset、dataloader
数据集介绍
我们使用CIFAR10数据集。CIFAR10数据集包含60,000张32x32的彩色图片，10个类别，每个类包含6,000张。其中50,000张图片作为训练集，10000张作为验证集。这次我们只对其中的猫和狗两类进行预测。
PaddlePaddle已经内置了若干种常用的数据集，使用CIFAR10特别简单
from paddle.vision.datasets import Cifar1
1.1解压原始数据集
#解压原始数据集函数 def untar_data(src_path,target_path): ''' 解压原始数据集，将src_path路径下的tar包解压至target_path目录下 ''' if(not os.path.isdir(target_path + "cifar-10-batches-py")): tar = tarfile.open(src_path) tar.extractall(path=target_path) tar.close() print('数据集解压完成') else: print('文件已存在')
#参数初始化 src_path=train_parameters['src_path'] target_path=train_parameters['target_path'] #解压原始数据到指定路径 untar_data(src_path,target_path)
文件已存在
1.2构造dataset、dataloader
train_dataset和eval_dataset
自定义读取器处理训练集和测试集
paddle.reader.shuffle()表示每次缓存BUF_SIZE个数据项，并进行打乱
paddle.batch()表示每BATCH_SIZE组成一个batch
def unpickle(file): # data：a 10000x3072 numpy array of uint8s. Each row of the array stores a 32x32 colour image. # The first 1024 entries contain the red channel values, the next 1024 the green, # and the final 1024 the blue. The image is stored in row-major order, # so that the first 32 entries of the array are the red channel values of the first row of the image. # labels：a list of 10000 numbers in the range 0-9. # The number at index i indicates the label of the ith image in the array data. fo = open(file, 'rb') dict = pickle.load(fo,encoding = 'bytes') train_labels = dict[b'labels'] train_array = dict[b'data'] train_array=train_array.tolist() fo.close() data_len=len(train_labels) for i in range(data_len-1,-1,-1): if train_labels[i]==3: train_labels[i]=0 elif train_labels[i]==5: train_labels[i]=1 else: train_labels.pop(i) train_array.pop(i) train_array=np.array(train_array) return train_labels, train_array
import paddle.vision.transforms as T from paddle.vision.transforms import Compose, Normalize, Resize, Grayscale from PIL import Image ''' 自定义dataset数据集 ''' from paddle.io import Dataset class MyDataset(paddle.io.Dataset): """ 步骤一：继承paddle.io.Dataset类 """ def __init__(self, mode='train'): """ 步骤二：实现构造函数，定义数据集大小 """ super(MyDataset, self).__init__() # 保存标签数据 self.data = [] # 保存图像数据 self.img_datas = [] # 临时变量 xs=[] ys=[] temp_labels=[] temp_datas=[] # transform定义，转灰度图，缩放到28*28尺寸，归一化 mean = [127.5] std = [127.5] self.transforms = Compose([Resize((28,28)), Grayscale(), Normalize(mean, std, 'CHW')]) if mode == 'train': #批量读入训练数据 for i in range(1,6): temp_label,temp_data=unpickle(target_path +"cifar-10-batches-py/data_batch_%d" % (i,)) ys.append(temp_label) xs.append(temp_data) temp_labels=np.concatenate(ys) temp_datas=np.concatenate(xs) else: ##批量读入测试数据 temp_labels,temp_datas=unpickle(target_path +"cifar-10-batches-py/test_batch") temp_labels=np.array(temp_labels) temp_datas=np.array(temp_datas) # 转为3*32*32图像数据 temp_datas = temp_datas.reshape((-1,3,32,32)) self.data=temp_labels self.img_datas = temp_datas def __getitem__(self, index): """ 步骤三：实现__getitem__方法，定义指定index时如何获取数据，并返回单条数据（训练数据，对应的标签） """ #返回单一数据和标签 data_image = self.img_datas[index] # 从numpy载入Image data_image = Image.fromarray(data_image, 'RGB') # 取图像并应用transform进行resize、灰度、normalize t_data_image = self.transforms(data_image) # 取标签 label = self.data[index] return t_data_image, np.array(label, dtype='int64') def __len__(self): """ 步骤四：实现__len__方法，返回数据集总数目 """ #返回数据总数 return len(self.data)
# 测试定义的数据集 train_dataset = MyDataset(mode='train') eval_dataset = MyDataset(mode='val') print('=============train_dataset =============') #输出数据集的形状和标签 print('train_dataset.__getitem__(1)[0].shape',train_dataset.__getitem__(1)[0].shape) print('train_dataset.__getitem__(1)[1]', train_dataset.__getitem__(1)[1]) #输出数据集的长度 print('train_dataset.__len__()',train_dataset.__len__()) print('=============eval_dataset =============') #输出数据集的长度 print('eval_dataset.__getitem__(1)[0].shape',eval_dataset.__getitem__(1)[0].shape) print('eval_dataset.__getitem__(1)[1]', eval_dataset.__getitem__(1)[1]) print('eval_dataset.__len__()',eval_dataset.__len__())
=============train_dataset ============= train_dataset.__getitem__(1)[0].shape (1, 28, 28) train_dataset.__getitem__(1)[1] 0 train_dataset.__len__() 10000 =============eval_dataset ============= eval_dataset.__getitem__(1)[0].shape (1, 28, 28) eval_dataset.__getitem__(1)[1] 0 eval_dataset.__len__() 2000
#训练数据DataLoad加载 train_loader = paddle.io.DataLoader(train_dataset, batch_size=train_parameters['train_batch_size'], shuffle=True ) #测试数据DataLoad加载 eval_loader = paddle.io.DataLoader(eval_dataset, batch_size=train_parameters['train_batch_size'], shuffle=False )
Step2.网络配置
（1）网络搭建
*** CNN网络模型
在CNN模型中，卷积神经网络能够更好的利用图像的结构信息。下面PaddlePaddle内置的一个较简单的卷积神经网络Lenet。
LeNet-5是卷积神经网络模型的早期代表，它由LeCun在1998年提出。该模型采用顺序结构，主要包括7层（2个卷积层、2个池化层和3个全连接层），卷积层和池化层交替排列。
import paddle import paddle.nn as nn class LeNet(nn.Layer): # Lenet定义 def __init__(self, num_classes=10): # 分类数，默认10 super(LeNet, self).__init__() self.num_classes = num_classes self.features = nn.Sequential( nn.Conv2D( 1, 6, 3, stride=1, padding=1), nn.ReLU(), nn.MaxPool2D(2, 2), nn.Conv2D( 6, 16, 5, stride=1, padding=0), nn.ReLU(), nn.MaxPool2D(2, 2)) if num_classes > 0: self.fc = nn.Sequential( nn.Linear(400, 120), nn.Linear(120, 84), nn.Linear(84, num_classes)) def forward(self, inputs): x = self.features(inputs) if self.num_classes > 0: x = paddle.flatten(x, 1) x = self.fc(x) return x
# 定义网络 network=LeNet(num_classes=train_parameters['class_dim']) # 装配模型 model=paddle.Model(network) # 打印网络结构 model.summary((1, 1, 28 , 28))
--------------------------------------------------------------------------- Layer (type) Input Shape Output Shape Param # =========================================================================== Conv2D-1 [[1, 1, 28, 28]] [1, 6, 28, 28] 60 ReLU-1 [[1, 6, 28, 28]] [1, 6, 28, 28] 0 MaxPool2D-1 [[1, 6, 28, 28]] [1, 6, 14, 14] 0 Conv2D-2 [[1, 6, 14, 14]] [1, 16, 10, 10] 2,416 ReLU-2 [[1, 16, 10, 10]] [1, 16, 10, 10] 0 MaxPool2D-2 [[1, 16, 10, 10]] [1, 16, 5, 5] 0 Linear-1 [[1, 400]] [1, 120] 48,120 Linear-2 [[1, 120]] [1, 84] 10,164 Linear-3 [[1, 84]] [1, 2] 170 =========================================================================== Total params: 60,930 Trainable params: 60,930 Non-trainable params: 0 --------------------------------------------------------------------------- Input size (MB): 0.00 Forward/backward pass size (MB): 0.11 Params size (MB): 0.23 Estimated Total Size (MB): 0.35 --------------------------------------------------------------------------- {'total_params': 60930, 'trainable_params': 60930}
Step3.模型训练 and Step4.模型评估
使用 paddle.optimizer.Adam 优化器来进行优化
使用 F.cross_entropy 来计算损失值
# 绘制损失函数图 def draw_process(title,color,iters,data,label): plt.title(title, fontsize=24) plt.xlabel("iter", fontsize=20) plt.ylabel(label, fontsize=20) plt.plot(iters, data,color=color,label=label) plt.legend() plt.grid() plt.show()
# 模型训练 # 初始化LeNet模型 model=LeNet(num_classes=train_parameters['class_dim']) # 训练模式 model.train() # 交叉熵 cross_entropy = paddle.nn.CrossEntropyLoss() # 优化器 optimizer = paddle.optimizer.Adam(learning_rate=train_parameters['learning_strategy']['lr'], parameters=model.parameters()) # 绘制loss、acc曲线图变量 steps = 0 Iters, total_loss, total_acc = [], [], [] # 开始训练 for epo in range(train_parameters['num_epochs']): for _, data in enumerate(train_loader()): steps += 1 x_data = data[0] x_data = paddle.to_tensor (x_data) y_data = paddle.to_tensor(data[1]) y_data = paddle.unsqueeze(y_data, 1) predicts = model(x_data) # 计算交叉熵 loss = cross_entropy(predicts, y_data) # 计算精确度 acc = paddle.metric.accuracy(predicts, y_data) # 反向传播 loss.backward() optimizer.step() # 梯度清零 optimizer.clear_grad() if steps % train_parameters["skip_steps"] == 0: Iters.append(steps) total_loss.append(loss.numpy()[0]) total_acc.append(acc.numpy()[0]) #打印中间过程 print('epo: {}, step: {}, loss is: {}, acc is: {}'\ .format(epo, steps, loss.numpy(), acc.numpy())) #保存模型参数 if steps % train_parameters["save_steps"] == 0: save_path = train_parameters["checkpoints"]+"/"+"save_dir_" + str(steps) + '.pdparams' print('save model to: ' + save_path) paddle.save(model.state_dict(),save_path) paddle.save(model.state_dict(),train_parameters["checkpoints"]+"/"+"save_dir_final.pdparams") draw_process("trainning loss","red",Iters,total_loss,"trainning loss") draw_process("trainning acc","green",Iters,total_acc,"trainning acc")
epo: 0, step: 5, loss is: [1.1097064], acc is: [0.49] save model to: /home/aistudio/checkpoints/save_dir_5.pdparams epo: 0, step: 10, loss is: [0.8729222], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_10.pdparams epo: 0, step: 15, loss is: [0.77851003], acc is: [0.49] save model to: /home/aistudio/checkpoints/save_dir_15.pdparams epo: 0, step: 20, loss is: [0.68859595], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_20.pdparams epo: 0, step: 25, loss is: [0.71485907], acc is: [0.48] save model to: /home/aistudio/checkpoints/save_dir_25.pdparams epo: 0, step: 30, loss is: [0.69014424], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_30.pdparams epo: 0, step: 35, loss is: [0.7331408], acc is: [0.42] save model to: /home/aistudio/checkpoints/save_dir_35.pdparams epo: 0, step: 40, loss is: [0.6923569], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_40.pdparams epo: 0, step: 45, loss is: [0.70091367], acc is: [0.49] save model to: /home/aistudio/checkpoints/save_dir_45.pdparams epo: 0, step: 50, loss is: [0.69078857], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_50.pdparams epo: 0, step: 55, loss is: [0.69088614], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_55.pdparams epo: 0, step: 60, loss is: [0.7027031], acc is: [0.46] save model to: /home/aistudio/checkpoints/save_dir_60.pdparams epo: 0, step: 65, loss is: [0.6824346], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_65.pdparams epo: 0, step: 70, loss is: [0.6795273], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_70.pdparams epo: 0, step: 75, loss is: [0.6809163], acc is: [0.59] save model to: /home/aistudio/checkpoints/save_dir_75.pdparams epo: 0, step: 80, loss is: [0.7107715], acc is: [0.43] save model to: /home/aistudio/checkpoints/save_dir_80.pdparams epo: 0, step: 85, loss is: [0.70901597], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_85.pdparams epo: 0, step: 90, loss is: [0.7054188], acc is: [0.44] save model to: /home/aistudio/checkpoints/save_dir_90.pdparams epo: 0, step: 95, loss is: [0.6982265], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_95.pdparams epo: 0, step: 100, loss is: [0.6998703], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_100.pdparams epo: 1, step: 105, loss is: [0.70260566], acc is: [0.44] save model to: /home/aistudio/checkpoints/save_dir_105.pdparams epo: 1, step: 110, loss is: [0.67828727], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_110.pdparams epo: 1, step: 115, loss is: [0.68608195], acc is: [0.49] save model to: /home/aistudio/checkpoints/save_dir_115.pdparams epo: 1, step: 120, loss is: [0.697596], acc is: [0.59] save model to: /home/aistudio/checkpoints/save_dir_120.pdparams epo: 1, step: 125, loss is: [0.7016902], acc is: [0.5] save model to: /home/aistudio/checkpoints/save_dir_125.pdparams epo: 1, step: 130, loss is: [0.6790494], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_130.pdparams epo: 1, step: 135, loss is: [0.68013227], acc is: [0.57] save model to: /home/aistudio/checkpoints/save_dir_135.pdparams epo: 1, step: 140, loss is: [0.70905924], acc is: [0.45] save model to: /home/aistudio/checkpoints/save_dir_140.pdparams epo: 1, step: 145, loss is: [0.6931264], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_145.pdparams epo: 1, step: 150, loss is: [0.6971727], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_150.pdparams epo: 1, step: 155, loss is: [0.67896414], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_155.pdparams epo: 1, step: 160, loss is: [0.67097855], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_160.pdparams epo: 1, step: 165, loss is: [0.69235575], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_165.pdparams epo: 1, step: 170, loss is: [0.6894104], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_170.pdparams epo: 1, step: 175, loss is: [0.70366347], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_175.pdparams epo: 1, step: 180, loss is: [0.69162464], acc is: [0.48] save model to: /home/aistudio/checkpoints/save_dir_180.pdparams epo: 1, step: 185, loss is: [0.67835146], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_185.pdparams epo: 1, step: 190, loss is: [0.6919897], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_190.pdparams epo: 1, step: 195, loss is: [0.69632596], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_195.pdparams epo: 1, step: 200, loss is: [0.70401454], acc is: [0.41] save model to: /home/aistudio/checkpoints/save_dir_200.pdparams epo: 2, step: 205, loss is: [0.72231257], acc is: [0.47] save model to: /home/aistudio/checkpoints/save_dir_205.pdparams epo: 2, step: 210, loss is: [0.6722144], acc is: [0.65] save model to: /home/aistudio/checkpoints/save_dir_210.pdparams epo: 2, step: 215, loss is: [0.7005479], acc is: [0.43] save model to: /home/aistudio/checkpoints/save_dir_215.pdparams epo: 2, step: 220, loss is: [0.68955404], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_220.pdparams epo: 2, step: 225, loss is: [0.68503153], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_225.pdparams epo: 2, step: 230, loss is: [0.6742158], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_230.pdparams epo: 2, step: 235, loss is: [0.68807405], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_235.pdparams epo: 2, step: 240, loss is: [0.7038729], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_240.pdparams epo: 2, step: 245, loss is: [0.69256955], acc is: [0.49] save model to: /home/aistudio/checkpoints/save_dir_245.pdparams epo: 2, step: 250, loss is: [0.6998977], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_250.pdparams epo: 2, step: 255, loss is: [0.6635308], acc is: [0.64] save model to: /home/aistudio/checkpoints/save_dir_255.pdparams epo: 2, step: 260, loss is: [0.6831071], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_260.pdparams epo: 2, step: 265, loss is: [0.6725425], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_265.pdparams epo: 2, step: 270, loss is: [0.6881926], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_270.pdparams epo: 2, step: 275, loss is: [0.69550765], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_275.pdparams epo: 2, step: 280, loss is: [0.68708885], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_280.pdparams epo: 2, step: 285, loss is: [0.68473077], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_285.pdparams epo: 2, step: 290, loss is: [0.6903842], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_290.pdparams epo: 2, step: 295, loss is: [0.7028897], acc is: [0.48] save model to: /home/aistudio/checkpoints/save_dir_295.pdparams epo: 2, step: 300, loss is: [0.6931243], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_300.pdparams epo: 3, step: 305, loss is: [0.68098104], acc is: [0.57] save model to: /home/aistudio/checkpoints/save_dir_305.pdparams epo: 3, step: 310, loss is: [0.6757507], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_310.pdparams epo: 3, step: 315, loss is: [0.7027341], acc is: [0.44] save model to: /home/aistudio/checkpoints/save_dir_315.pdparams epo: 3, step: 320, loss is: [0.7009732], acc is: [0.49] save model to: /home/aistudio/checkpoints/save_dir_320.pdparams epo: 3, step: 325, loss is: [0.7078163], acc is: [0.47] save model to: /home/aistudio/checkpoints/save_dir_325.pdparams epo: 3, step: 330, loss is: [0.6958405], acc is: [0.44] save model to: /home/aistudio/checkpoints/save_dir_330.pdparams epo: 3, step: 335, loss is: [0.69992703], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_335.pdparams epo: 3, step: 340, loss is: [0.69363695], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_340.pdparams epo: 3, step: 345, loss is: [0.6923307], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_345.pdparams epo: 3, step: 350, loss is: [0.6739081], acc is: [0.6] save model to: /home/aistudio/checkpoints/save_dir_350.pdparams epo: 3, step: 355, loss is: [0.68306243], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_355.pdparams epo: 3, step: 360, loss is: [0.66385293], acc is: [0.64] save model to: /home/aistudio/checkpoints/save_dir_360.pdparams epo: 3, step: 365, loss is: [0.6816753], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_365.pdparams epo: 3, step: 370, loss is: [0.6921282], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_370.pdparams epo: 3, step: 375, loss is: [0.6865966], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_375.pdparams epo: 3, step: 380, loss is: [0.69338584], acc is: [0.49] save model to: /home/aistudio/checkpoints/save_dir_380.pdparams epo: 3, step: 385, loss is: [0.6800542], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_385.pdparams epo: 3, step: 390, loss is: [0.6839569], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_390.pdparams epo: 3, step: 395, loss is: [0.6774286], acc is: [0.57] save model to: /home/aistudio/checkpoints/save_dir_395.pdparams epo: 3, step: 400, loss is: [0.7004008], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_400.pdparams epo: 4, step: 405, loss is: [0.7059412], acc is: [0.44] save model to: /home/aistudio/checkpoints/save_dir_405.pdparams epo: 4, step: 410, loss is: [0.69455093], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_410.pdparams epo: 4, step: 415, loss is: [0.6933525], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_415.pdparams epo: 4, step: 420, loss is: [0.7079694], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_420.pdparams epo: 4, step: 425, loss is: [0.6937676], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_425.pdparams epo: 4, step: 430, loss is: [0.68947273], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_430.pdparams epo: 4, step: 435, loss is: [0.6781409], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_435.pdparams epo: 4, step: 440, loss is: [0.6878584], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_440.pdparams epo: 4, step: 445, loss is: [0.66629666], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_445.pdparams epo: 4, step: 450, loss is: [0.66666824], acc is: [0.62] save model to: /home/aistudio/checkpoints/save_dir_450.pdparams epo: 4, step: 455, loss is: [0.67528206], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_455.pdparams epo: 4, step: 460, loss is: [0.7015488], acc is: [0.47] save model to: /home/aistudio/checkpoints/save_dir_460.pdparams epo: 4, step: 465, loss is: [0.6915476], acc is: [0.61] save model to: /home/aistudio/checkpoints/save_dir_465.pdparams epo: 4, step: 470, loss is: [0.6868398], acc is: [0.57] save model to: /home/aistudio/checkpoints/save_dir_470.pdparams epo: 4, step: 475, loss is: [0.69640535], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_475.pdparams epo: 4, step: 480, loss is: [0.6844581], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_480.pdparams epo: 4, step: 485, loss is: [0.678205], acc is: [0.61] save model to: /home/aistudio/checkpoints/save_dir_485.pdparams epo: 4, step: 490, loss is: [0.6782288], acc is: [0.57] save model to: /home/aistudio/checkpoints/save_dir_490.pdparams epo: 4, step: 495, loss is: [0.6809789], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_495.pdparams epo: 4, step: 500, loss is: [0.6791268], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_500.pdparams epo: 5, step: 505, loss is: [0.66857773], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_505.pdparams epo: 5, step: 510, loss is: [0.68727225], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_510.pdparams epo: 5, step: 515, loss is: [0.68932843], acc is: [0.49] save model to: /home/aistudio/checkpoints/save_dir_515.pdparams epo: 5, step: 520, loss is: [0.68978363], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_520.pdparams epo: 5, step: 525, loss is: [0.69064134], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_525.pdparams epo: 5, step: 530, loss is: [0.682237], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_530.pdparams epo: 5, step: 535, loss is: [0.68976945], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_535.pdparams epo: 5, step: 540, loss is: [0.67902535], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_540.pdparams epo: 5, step: 545, loss is: [0.67134506], acc is: [0.65] save model to: /home/aistudio/checkpoints/save_dir_545.pdparams epo: 5, step: 550, loss is: [0.6688429], acc is: [0.61] save model to: /home/aistudio/checkpoints/save_dir_550.pdparams epo: 5, step: 555, loss is: [0.7254223], acc is: [0.49] save model to: /home/aistudio/checkpoints/save_dir_555.pdparams epo: 5, step: 560, loss is: [0.69241136], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_560.pdparams epo: 5, step: 565, loss is: [0.6801878], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_565.pdparams epo: 5, step: 570, loss is: [0.6906636], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_570.pdparams epo: 5, step: 575, loss is: [0.70213795], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_575.pdparams epo: 5, step: 580, loss is: [0.69319504], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_580.pdparams epo: 5, step: 585, loss is: [0.7011637], acc is: [0.48] save model to: /home/aistudio/checkpoints/save_dir_585.pdparams epo: 5, step: 590, loss is: [0.6848818], acc is: [0.59] save model to: /home/aistudio/checkpoints/save_dir_590.pdparams epo: 5, step: 595, loss is: [0.67795885], acc is: [0.6] save model to: /home/aistudio/checkpoints/save_dir_595.pdparams epo: 5, step: 600, loss is: [0.6833943], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_600.pdparams epo: 6, step: 605, loss is: [0.6795752], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_605.pdparams epo: 6, step: 610, loss is: [0.6964473], acc is: [0.5] save model to: /home/aistudio/checkpoints/save_dir_610.pdparams epo: 6, step: 615, loss is: [0.7281563], acc is: [0.43] save model to: /home/aistudio/checkpoints/save_dir_615.pdparams epo: 6, step: 620, loss is: [0.675564], acc is: [0.6] save model to: /home/aistudio/checkpoints/save_dir_620.pdparams epo: 6, step: 625, loss is: [0.6895311], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_625.pdparams epo: 6, step: 630, loss is: [0.67448664], acc is: [0.64] save model to: /home/aistudio/checkpoints/save_dir_630.pdparams epo: 6, step: 635, loss is: [0.6737503], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_635.pdparams epo: 6, step: 640, loss is: [0.70881164], acc is: [0.46] save model to: /home/aistudio/checkpoints/save_dir_640.pdparams epo: 6, step: 645, loss is: [0.68261325], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_645.pdparams epo: 6, step: 650, loss is: [0.6765045], acc is: [0.57] save model to: /home/aistudio/checkpoints/save_dir_650.pdparams epo: 6, step: 655, loss is: [0.6759614], acc is: [0.59] save model to: /home/aistudio/checkpoints/save_dir_655.pdparams epo: 6, step: 660, loss is: [0.6793112], acc is: [0.57] save model to: /home/aistudio/checkpoints/save_dir_660.pdparams epo: 6, step: 665, loss is: [0.6845392], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_665.pdparams epo: 6, step: 670, loss is: [0.6814833], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_670.pdparams epo: 6, step: 675, loss is: [0.68463284], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_675.pdparams epo: 6, step: 680, loss is: [0.6939957], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_680.pdparams epo: 6, step: 685, loss is: [0.6949662], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_685.pdparams epo: 6, step: 690, loss is: [0.6850964], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_690.pdparams epo: 6, step: 695, loss is: [0.6783737], acc is: [0.6] save model to: /home/aistudio/checkpoints/save_dir_695.pdparams epo: 6, step: 700, loss is: [0.6847418], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_700.pdparams epo: 7, step: 705, loss is: [0.6696134], acc is: [0.6] save model to: /home/aistudio/checkpoints/save_dir_705.pdparams epo: 7, step: 710, loss is: [0.699369], acc is: [0.59] save model to: /home/aistudio/checkpoints/save_dir_710.pdparams epo: 7, step: 715, loss is: [0.6834408], acc is: [0.61] save model to: /home/aistudio/checkpoints/save_dir_715.pdparams epo: 7, step: 720, loss is: [0.6834759], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_720.pdparams epo: 7, step: 725, loss is: [0.68610823], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_725.pdparams epo: 7, step: 730, loss is: [0.667547], acc is: [0.6] save model to: /home/aistudio/checkpoints/save_dir_730.pdparams epo: 7, step: 735, loss is: [0.70002645], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_735.pdparams epo: 7, step: 740, loss is: [0.6882743], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_740.pdparams epo: 7, step: 745, loss is: [0.6829937], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_745.pdparams epo: 7, step: 750, loss is: [0.6799063], acc is: [0.57] save model to: /home/aistudio/checkpoints/save_dir_750.pdparams epo: 7, step: 755, loss is: [0.6759838], acc is: [0.59] save model to: /home/aistudio/checkpoints/save_dir_755.pdparams epo: 7, step: 760, loss is: [0.7013712], acc is: [0.47] save model to: /home/aistudio/checkpoints/save_dir_760.pdparams epo: 7, step: 765, loss is: [0.67678285], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_765.pdparams epo: 7, step: 770, loss is: [0.6903254], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_770.pdparams epo: 7, step: 775, loss is: [0.71212935], acc is: [0.45] save model to: /home/aistudio/checkpoints/save_dir_775.pdparams epo: 7, step: 780, loss is: [0.66622734], acc is: [0.62] save model to: /home/aistudio/checkpoints/save_dir_780.pdparams epo: 7, step: 785, loss is: [0.6900478], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_785.pdparams epo: 7, step: 790, loss is: [0.6736644], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_790.pdparams epo: 7, step: 795, loss is: [0.70363], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_795.pdparams epo: 7, step: 800, loss is: [0.6934418], acc is: [0.49] save model to: /home/aistudio/checkpoints/save_dir_800.pdparams epo: 8, step: 805, loss is: [0.6693335], acc is: [0.59] save model to: /home/aistudio/checkpoints/save_dir_805.pdparams epo: 8, step: 810, loss is: [0.694484], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_810.pdparams epo: 8, step: 815, loss is: [0.7255772], acc is: [0.37] save model to: /home/aistudio/checkpoints/save_dir_815.pdparams epo: 8, step: 820, loss is: [0.6915439], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_820.pdparams epo: 8, step: 825, loss is: [0.6881697], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_825.pdparams epo: 8, step: 830, loss is: [0.68885416], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_830.pdparams epo: 8, step: 835, loss is: [0.6819633], acc is: [0.52] save model to: /home/aistudio/checkpoints/save_dir_835.pdparams epo: 8, step: 840, loss is: [0.68416965], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_840.pdparams epo: 8, step: 845, loss is: [0.6801962], acc is: [0.5] save model to: /home/aistudio/checkpoints/save_dir_845.pdparams epo: 8, step: 850, loss is: [0.67312473], acc is: [0.62] save model to: /home/aistudio/checkpoints/save_dir_850.pdparams epo: 8, step: 855, loss is: [0.6651606], acc is: [0.64] save model to: /home/aistudio/checkpoints/save_dir_855.pdparams epo: 8, step: 860, loss is: [0.66604716], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_860.pdparams epo: 8, step: 865, loss is: [0.6775603], acc is: [0.61] save model to: /home/aistudio/checkpoints/save_dir_865.pdparams epo: 8, step: 870, loss is: [0.6985699], acc is: [0.5] save model to: /home/aistudio/checkpoints/save_dir_870.pdparams epo: 8, step: 875, loss is: [0.6906618], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_875.pdparams epo: 8, step: 880, loss is: [0.6838692], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_880.pdparams epo: 8, step: 885, loss is: [0.6818925], acc is: [0.63] save model to: /home/aistudio/checkpoints/save_dir_885.pdparams epo: 8, step: 890, loss is: [0.7003258], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_890.pdparams epo: 8, step: 895, loss is: [0.7080064], acc is: [0.5] save model to: /home/aistudio/checkpoints/save_dir_895.pdparams epo: 8, step: 900, loss is: [0.67341954], acc is: [0.61] save model to: /home/aistudio/checkpoints/save_dir_900.pdparams epo: 9, step: 905, loss is: [0.68930835], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_905.pdparams epo: 9, step: 910, loss is: [0.7058289], acc is: [0.46] save model to: /home/aistudio/checkpoints/save_dir_910.pdparams epo: 9, step: 915, loss is: [0.67915636], acc is: [0.6] save model to: /home/aistudio/checkpoints/save_dir_915.pdparams epo: 9, step: 920, loss is: [0.687831], acc is: [0.58] save model to: /home/aistudio/checkpoints/save_dir_920.pdparams epo: 9, step: 925, loss is: [0.6957987], acc is: [0.46] save model to: /home/aistudio/checkpoints/save_dir_925.pdparams epo: 9, step: 930, loss is: [0.6923476], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_930.pdparams epo: 9, step: 935, loss is: [0.70298016], acc is: [0.47] save model to: /home/aistudio/checkpoints/save_dir_935.pdparams epo: 9, step: 940, loss is: [0.69297534], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_940.pdparams epo: 9, step: 945, loss is: [0.6787053], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_945.pdparams epo: 9, step: 950, loss is: [0.6894692], acc is: [0.57] save model to: /home/aistudio/checkpoints/save_dir_950.pdparams epo: 9, step: 955, loss is: [0.70166737], acc is: [0.51] save model to: /home/aistudio/checkpoints/save_dir_955.pdparams epo: 9, step: 960, loss is: [0.69754714], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_960.pdparams epo: 9, step: 965, loss is: [0.6867398], acc is: [0.54] save model to: /home/aistudio/checkpoints/save_dir_965.pdparams epo: 9, step: 970, loss is: [0.6726653], acc is: [0.59] save model to: /home/aistudio/checkpoints/save_dir_970.pdparams epo: 9, step: 975, loss is: [0.6738178], acc is: [0.56] save model to: /home/aistudio/checkpoints/save_dir_975.pdparams epo: 9, step: 980, loss is: [0.6699579], acc is: [0.55] save model to: /home/aistudio/checkpoints/save_dir_980.pdparams epo: 9, step: 985, loss is: [0.6805726], acc is: [0.57] save model to: /home/aistudio/checkpoints/save_dir_985.pdparams epo: 9, step: 990, loss is: [0.68973434], acc is: [0.53] save model to: /home/aistudio/checkpoints/save_dir_990.pdparams epo: 9, step: 995, loss is: [0.66840816], acc is: [0.6] save model to: /home/aistudio/checkpoints/save_dir_995.pdparams epo: 9, step: 1000, loss is: [0.7072126], acc is: [0.49] save model to: /home/aistudio/checkpoints/save_dir_1000.pdparams
模型验证
训练完成后，需要验证模型的效果，此时，加载测试数据集，然后用训练好的模对测试集进行预测，计算损失与精度。
''' 模型评估 ''' model__state_dict = paddle.load(train_parameters["checkpoints"]+"/"+"save_dir_final.pdparams") model_eval = LeNet( num_classes=train_parameters['class_dim']) model_eval.set_state_dict(model__state_dict) model_eval.eval() accs = [] for _, data in enumerate(eval_loader()): x_data = data[0] y_data = paddle.to_tensor(data[1]) y_data = paddle.unsqueeze(y_data, 1) predicts = model_eval(x_data) # 计算acc acc = paddle.metric.accuracy(predicts, y_data) accs.append(acc.numpy()[0]) print('模型在验证集上的准确率为：',np.mean(accs))
模型在验证集上的准确率为： 0.5365
Step5.模型预测
# 图片预处理 def load_image(file): ''' 预测图片预处理 ''' #打开图片 im = Image.open(file) #将图片调整为跟训练数据一样的大小 28*28，设定ANTIALIAS，即抗锯齿.resize是缩放 im = im.resize((28, 28), Image.ANTIALIAS) # 转灰度图 im = im.convert('1') #建立图片矩阵 类型为float32 im = np.array(im).astype(np.float32) #矩阵转置 #将像素值从【0-255】转换为【0-1】 im = im / 255.0 #print(im) im = np.expand_dims(im, axis=0) # 保持和之前输入image维度一致 print('im_shape的维度：',im.shape) return im
''' 模型预测 ''' # 载入模型 model__state_dict = paddle.load(train_parameters["checkpoints"]+"/"+"save_dir_final.pdparams") model_eval = LeNet( num_classes=train_parameters['class_dim']) model_eval.set_state_dict(model__state_dict) #训练模式 model.eval() #展示预测图片 infer_path='/home/aistudio/data/data7940/dog.png' img = Image.open(infer_path) plt.imshow(img) #根据数组绘制图像 plt.show() #显示图像 #对预测图片进行预处理 infer_img = load_image(infer_path) infer_img = infer_img.reshape(1,28,28) # infer_img = infer_img.toGra #定义标签列表 label_list = [ "cat", "dog"] data = infer_img dy_x_data = np.array(data).astype('float32') dy_x_data=dy_x_data[np.newaxis,:, : ,:] img = paddle.to_tensor(dy_x_data) out = model(img) lab = np.argmax(out.numpy()) #argmax():返回最大数的索引 print(label_list[lab])
im_shape的维度： (1, 28, 28) dog
