在pytorch中自定义dataset读取数据2021-1-8学习笔记

在pytorch中自定义dataset读取数据

utils

import os import json import pickle import randomimport matplotlib.pyplot as pltdef read_split_data(root: str, val_rate: float = 0.2):# val_rate划分验证集的比例random.seed(0) # 保证随机结果可复现 #随机种子设置为0，大家划分的是一样的assert os.path.exists(root), "dataset root: {} does not exist.".format(root) #不存在路径报错# 遍历文件夹，一个文件夹对应一个类别flower_class = [cla for cla in os.listdir(root) if os.path.isdir(os.path.join(root, cla))]#不是文件夹丢弃# 排序，保证顺序一致flower_class.sort()# 生成类别名称以及对应的数字索引class_indices = dict((k, v) for v, k in enumerate(flower_class))json_str = json.dumps(dict((val, key) for key, val in class_indices.items()), indent=4)with open('class_indices.json', 'w') as json_file:json_file.write(json_str)train_images_path = [] # 存储训练集的所有图片路径train_images_label = [] # 存储训练集图片对应索引信息val_images_path = [] # 存储验证集的所有图片路径val_images_label = [] # 存储验证集图片对应索引信息every_class_num = [] # 存储每个类别的样本总数supported = [".jpg", ".JPG", ".png", ".PNG"] # 支持的文件后缀类型# 遍历每个文件夹下的文件for cla in flower_class:cla_path = os.path.join(root, cla) #获得该类别的路径# 遍历获取supported支持的所有文件路径images = [os.path.join(root, cla, i) for i in os.listdir(cla_path)if os.path.splitext(i)[-1] in supported]#splitext(i)[-1]分割出文件名称和后缀名 然后用in判断是否在supported里# 获取该类别对应的索引image_class = class_indices[cla]# 记录该类别的样本数量every_class_num.append(len(images))# 按比例随机采样验证样本val_path = random.sample(images, k=int(len(images) * val_rate))for img_path in images:if img_path in val_path: # 如果该路径在采样的验证集样本中则存入验证集val_images_path.append(img_path)val_images_label.append(image_class)else: # 否则存入训练集train_images_path.append(img_path)train_images_label.append(image_class)print("{} images were found in the dataset.".format(sum(every_class_num)))plot_image = Falseif plot_image:# 绘制每种类别个数柱状图plt.bar(range(len(flower_class)), every_class_num, align='center')# 将横坐标0,1,2,3,4替换为相应的类别名称plt.xticks(range(len(flower_class)), flower_class)# 在柱状图上添加数值标签for i, v in enumerate(every_class_num):plt.text(x=i, y=v + 5, s=str(v), ha='center')# 设置x坐标plt.xlabel('image class')# 设置y坐标plt.ylabel('number of images')# 设置柱状图的标题plt.title('flower class distribution')plt.show()return train_images_path, train_images_label, val_images_path, val_images_labeldef plot_data_loader_image(data_loader):batch_size = data_loader.batch_sizeplot_num = min(batch_size, 4)json_path = './class_indices.json'assert os.path.exists(json_path), json_path + " does not exist."json_file = open(json_path, 'r')class_indices = json.load(json_file)for data in data_loader:images, labels = datafor i in range(plot_num):# [C, H, W] -> [H, W, C] transpose调整顺序img = images[i].numpy().transpose(1, 2, 0)# 反Normalize操作img = (img * [0.229, 0.224, 0.225] + [0.485, 0.456, 0.406]) * 255label = labels[i].item()plt.subplot(1, plot_num, i+1)plt.xlabel(class_indices[str(label)])plt.xticks([]) # 去掉x轴的刻度plt.yticks([]) # 去掉y轴的刻度plt.imshow(img.astype('uint8'))plt.show()def write_pickle(list_info: list, file_name: str):with open(file_name, 'wb') as f:pickle.dump(list_info, f)def read_pickle(file_name: str) -> list:with open(file_name, 'rb') as f:info_list = pickle.load(f)return info_list

mydataset

from PIL import Image import torch from torch.utils.data import Datasetclass MyDataSet(Dataset):"""自定义数据集"""def __init__(self, images_path: list, images_class: list, transform=None):#初始化函数self.images_path = images_pathself.images_class = images_classself.transform = transformdef __len__(self):#计算该数据集下所有的样本个数return len(self.images_path)def __getitem__(self, item):#每次传入一个索引，就返回该索引对应的图片以及标签信息img = Image.open(self.images_path[item])#获得img的路径，然后得到PIL格式图像，pytorch用PIL比openCV好# RGB为彩色图片，L为灰度图片if img.mode != 'RGB':raise ValueError("image: {} isn't RGB mode.".format(self.images_path[item]))#报错，如果是灰度，就把上一行改成Llabel = self.images_class[item]if self.transform is not None:img = self.transform(img)#对图像进行预处理return img, label@staticmethod#是个静态方法def collate_fn(batch):#dataloader会使用# 官方实现的default_collate可以参考# https://github.com/pytorch/pytorch/blob/67b7e751e6b5931a9f45274653f4f653a4e6cdf6/torch/utils/data/_utils/collate.pyimages, labels = tuple(zip(*batch))#zip将图片和图片放一起，标签和标签放一起images = torch.stack(images, dim=0)#拼接，并会在dim=0的维度上进行拼接（就是拼成一个矩阵）labels = torch.as_tensor(labels)#标签也转换成tensorreturn images, labels

main

import osimport torch from torchvision import transformsfrom my_dataset import MyDataSet from utils import read_split_data, plot_data_loader_image# http://download.tensorflow.org/example_images/flower_photos.tgz root = "/home/wz/my_github/data_set/flower_data/flower_photos" # 数据集所在根目录def main():device = torch.device("cuda" if torch.cuda.is_available() else "cpu")print("using {} device.".format(device))train_images_path, train_images_label, val_images_path, val_images_label = read_split_data(root)data_transform = {"train": transforms.Compose([transforms.RandomResizedCrop(224),#随机裁剪transforms.RandomHorizontalFlip(),#水平翻转transforms.ToTensor(),#转化成tensor格式transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]),"val": transforms.Compose([transforms.Resize(256),transforms.CenterCrop(224),transforms.ToTensor(),transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])}##这个很重要，可以自己实现#实例化datasettrain_data_set = MyDataSet(images_path=train_images_path,#训练集图像列表images_class=train_images_label,#训练集所有图像对应的标签信息transform=data_transform["train"])#预处理方法batch_size = 8nw = min([os.cpu_count(), batch_size if batch_size > 1 else 0, 8]) # number of workersprint('Using {} dataloader workers'.format(nw))train_loader = torch.utils.data.DataLoader(train_data_set,#从实例化的dataset当中取得图片，然后打包成一个一个batch，然后输入网络进行训练batch_size=batch_size,shuffle=True,#打乱数据集num_workers=nw,#训练时建议nw，调试时建议0collate_fn=train_data_set.collate_fn)# plot_data_loader_image(train_loader)for step, data in enumerate(train_loader):images, labels = dataif __name__ == '__main__':main()

总结

以上是生活随笔为你收集整理的在pytorch中自定义dataset读取数据2021-1-8学习笔记的全部内容，希望文章能够帮你解决所遇到的问题。

如果觉得生活随笔网站内容还不错，欢迎将生活随笔推荐给好友。