解決pytorch讀取自制數(shù)據(jù)集出現(xiàn)過的問題

imgs.append(words[0], int(words[1]))

imgs.append((words[0], int(words[1])))

#是第一次運(yùn)行的話會(huì)下載數(shù)據(jù)集 現(xiàn)成的話可以使用root參數(shù)指定數(shù)據(jù)集位置
# 存放的格式如下圖
 
# 根據(jù)接口讀取默認(rèn)的CIFAR10數(shù)據(jù) 進(jìn)行訓(xùn)練和測試
#預(yù)處理
transform = transform.Compose([transform.ToTensor(), transform.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
#讀取數(shù)據(jù)集
trainset = torchvision.datasets.CIFAR10(root='./data', train=True, download=False, transform=transform)
#打包成DataLoader
trainloader = torch.utils.data.DataLoader(trainset, batch_size=4, shuffle=True, num_workers=1)
 
#同上
testset = torchvision.datasets.CIFAR10(root='./data', train=False, download=False, transform=transform)
testloader = torch.utils.data.DataLoader(trainset, batch_size=4, shuffle=False, num_workers=1)
classes = (1,2,3,4,5,6,7,8,9,10)  #類別定義
 
#使用
 for epoch in range(3):
        running_loss = 0.0 #清空loss
        for i, data in enumerate(trainloader, 0):
            # get the inputs
            inputs, labels = data #trainloader返回：id,image,labels
 
            # 將inputs與labels裝進(jìn)Variable中
            inputs, labels = Variable(inputs), Variable(labels)
            
            #使用print代替輸出
            print("epoch：", epoch, "的第", i, "個(gè)inputs", inputs.data.size(), "labels", labels.data.size())
 

# 預(yù)處理
transform = transform.Compose([transform.ToTensor(), transform.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
 
#使用torchvision.datasets.ImageFolder讀取數(shù)據(jù)集 指定train 和 test文件夾
img_data = torchvision.datasets.ImageFolder('data/cifar-10/train/', transform=transform)
data_loader = torch.utils.data.DataLoader(img_data, batch_size=4, shuffle=True, num_workers=1)
 
testset = torchvision.datasets.ImageFolder('data/cifar-10/test/', transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=4, shuffle=True, num_workers=1)
 
 for epoch in range(3):
        for i, data in enumerate(trainloader, 0):
            # get the inputs
            inputs, labels = data #trainloader返回：id,image,labels
            # 將inputs與labels裝進(jìn)Variable中
            inputs, labels = Variable(inputs), Variable(labels)
 
            #使用print代替輸出
            print("epoch：", epoch, "的第", i, "個(gè)inputs", inputs.data.size(), "labels", labels.data.size())

trainlist = []  # 保存特征的列表
 
targetpath = 'a/b/b'
filelist = os.listdir(targetpath) #列出文件夾下所有的目錄與文件
filecount = len(filelist)
# 根據(jù)根路徑 讀取所有文件名 循環(huán)讀取文件內(nèi)容 添加到list
for i in range(filecount):
     filepath = os.path.join(targetpath, filelist[j])
     with open(filepath, 'r') as f:
         line = f.readline()
         # 例如存儲(chǔ)格式為 1,2,3,4,5,6 數(shù)字之間以逗號(hào)隔開
         templist = list(map(int, line.split(',')))
         trainlist.append(templist)
 
# 數(shù)據(jù)讀取完畢 現(xiàn)在為維度為filecount的列表 我們需要轉(zhuǎn)換格式和類型
# 將數(shù)據(jù)轉(zhuǎn)換為Tensor
 
# 假如我們的兩類數(shù)據(jù)分別存在list0 和 list1中
split = len(list0) # 用于記錄標(biāo)簽的分界
 
#使用numpy.array 和 torch.from_numpy 連續(xù)將其轉(zhuǎn)換為tensor  使用torch.cat拼接
train0_numpy = numpy.array(list0)
train1_numpy = numpy.array(list1)
train_tensor = torch.cat([torch.from_numpy(train0_numpy), torch.from_numpytrain1_numpy)], 0)
#現(xiàn)在的尺寸是【樣本數(shù)，長度】 然而在使用神 經(jīng)網(wǎng)絡(luò)處理一維數(shù)據(jù)要求【樣本數(shù)，維度，長度】
# 這個(gè)維度指的像一個(gè)圖像實(shí)際上是一個(gè)二維矩陣 但是有三個(gè)RGB通道 實(shí)際就為【3，行，列】 那么需要處理三個(gè)矩陣
# 我們需要在我們的數(shù)據(jù)中加上這個(gè)維度信息
# 注意類型要一樣 可以轉(zhuǎn)換
shaper = train_tensor.shape  #獲取維度 【樣本數(shù)，長度】
aa = torch.ones((shaper[0], 1, shaper[1])) # 生成目標(biāo)矩陣
for i in range(shaper[0]):  # 將所有樣本復(fù)制到新矩陣
·    aa[i][0][:] = train_tensor[i][:]
train_tensor = aa  # 完成了數(shù)據(jù)集的轉(zhuǎn)換 【樣本數(shù)，維度，長度】
 
# 注 意 如果是讀取的圖像 我們需要的目標(biāo)維度是【樣本數(shù)，維度，size_w,size_h】
# 卷積接受的輸入是這樣的四維度 最后的兩個(gè)是圖像的尺寸 維度表示是通道數(shù)量 
  
# 下面是生成標(biāo)簽 標(biāo)簽注意類別之間的分界 split已經(jīng)在上文計(jì)算出來
# 訓(xùn)練標(biāo)簽的
total = len(list0) + len(list1)
train_label = numpy.zeros(total)
train_label[split+1:total] = 1
train_label_tensor = torch.from_numpy(train_label).int()
# print(train_tensor.size(),train_label_tensor.size())
 
# 搭建dataloader完畢
train_dataset = TensorDataset(train_tensor, train_label_tensor)
train_loader = DataLoader(dataset=train_dataset, batch_size=4, shuffle=True)
 
for epoch in range(3):
    for i, data in enumerate(trainloader, 0):
        # get the inputs
        inputs, labels = data #trainloader返回：id,image,labels
        # 將inputs與labels裝進(jìn)Variable中
        inputs, labels = Variable(inputs), Variable(labels)
 
        #使用print代替輸出
        print("epoch：", epoch, "的第", i, "個(gè)inputs", inputs.data.size(), "labels", labels.data.size())

class Dataset(object):
"""An abstract class representing a Dataset.
All other datasets should subclass it. All subclasses should override
``__len__``, that provides the size of the dataset, and ``__getitem__``,
supporting integer indexing in range from 0 to len(self) exclusive.
"""
def __getitem__(self, index):
	raise NotImplementedError
def __len__(self):
	raise NotImplementedError
def __add__(self, other):
	return ConcatDataset([self, other])

# coding: utf-8
from PIL import Image
from torch.utils.data import Dataset
class MyDataset(Dataset):
# 初始化讀取txt 可以設(shè)定變換
def __init__(self, txt_path, transform = None, target_transform = None):
	fh = open(txt_path, 'r')
	imgs = []
	for line in fh:
		line = line.rstrip()
		words = line.split()
         # 保存列表 其中有圖像的數(shù)據(jù) 和標(biāo)簽
		imgs.append((words[0], int(words[1])))
		self.imgs = imgs 
		self.transform = transform
		self.target_transform = target_transform
def __getitem__(self, index):
	fn, label = self.imgs[index]
	img = Image.open(fn).convert('RGB') 
	if self.transform is not None:
		img = self.transform(img) 
    # 返回圖像和標(biāo)簽
    
	return img, label
def __len__(self):
	return len(self.imgs)
 
# 當(dāng)然也可以創(chuàng)建myImageFloder 其txt格式在下圖顯示 
import os
import torch
import torch.utils.data as data
from PIL import Image 
def default_loader(path):
    return Image.open(path).convert('RGB')
 
class myImageFloder(data.Dataset):
    def __init__(self, root, label, transform = None, target_transform=None, loader=default_loader):
        fh = open(label) #打開label文件
        c=0
        imgs=[]  # 保存圖像的列表
        class_names=[]
        for line in  fh.readlines(): #讀取每一行數(shù)據(jù)
            if c==0:
                class_names=[n.strip() for n in line.rstrip().split('	')] 
            else:
                cls = line.split() #分割為列表
                fn = cls.pop(0)  #彈出最上的一個(gè)
                if os.path.isfile(os.path.join(root, fn)):  # 組合路徑名 讀取圖像
                    imgs.append((fn, tuple([float(v) for v in cls])))  #添加到列表
            c=c+1
 
        # 設(shè)置信息
        self.root = root
        self.imgs = imgs
        self.classes = class_names
        self.transform = transform
        self.target_transform = target_transform
        self.loader = loader
 
    def __getitem__(self, index):  # 獲取圖像 給定序號(hào)
        fn, label = self.imgs[index]  #讀取圖像的內(nèi)容和對(duì)應(yīng)的label
        img = self.loader(os.path.join(self.root, fn))
        if self.transform is not None:  # 是否變換
            img = self.transform(img)
        return img, torch.Tensor(label) # 返回圖像和label
 
    def __len__(self):
        return len(self.imgs)
    
    def getName(self):
        return self.classes
# 

# 而后使用的時(shí)候就可以正常的使用
trainset = MyDataset(txt_path=pathFile,transform = None, target_transform = None)
# trainset = torch.utils.data.DataLoader(myFloder.myImageFloder(root = "../data/testImages/images", label = "../data/testImages/test_images.txt", transform = mytransform ), batch_size= 2, shuffle= False, num_workers= 2)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=32, shuffle=True, num_workers=8)