迁移学习之Multi-Domain Adaptation多领域自适应常用数据集PACS介绍

PACS数据集

Paper：Self-supervised Domain Adaptation for Computer Vision Tasks

GitHub：https://github.com/robertofranceschi/Domain-adaptation-on-PACS-dataset

数据集下载：https://github.com/MachineLearning2020/Homework3-PACS/tree/master/PACS

• PACS数据集总共9991张图片，每张图片3x227x227
• 7 classes：Dog, Elephant, Giraffe, Guitar, Horse, House, Person
• 4 domains: Art painting, Cartoon, Photo and Sketch.
• Photo (1,670 images), Art Painting (2,048 images), Cartoon (2,344 images) and Sketch (3,929 images)

用Pytorch加载PACS数据集

PACS原始数据集目录结果：

from torch.utils.data import Dataset, DataLoader
import matplotlib.pyplot as plt
import os
import random
from PIL import Image
import torch
import numpy as np
from torchvision.transforms import transforms
from sklearn.model_selection import train_test_split
import os
import torchvision.transforms as transforms
from PIL import Image
from torch.utils.data import Dataset, DataLoader

class PACS(Dataset):
    def __init__(self, root_path, domain, train=True, transform=None, target_transform=None):
        self.root = f"{root_path}/{domain}"
        self.train = train
        self.transform = transform
        self.target_transform = target_transform

        label_name_list = os.listdir(self.root)
        self.label = []
        self.data = []

        if not os.path.exists(f"{root_path}/precessed"):
            os.makedirs(f"{root_path}/precessed")
        if os.path.exists(f"{root_path}/precessed/{domain}_data.pt") and os.path.exists(
                f"{root_path}/precessed/{domain}_label.pt"):
            print(f"Load {domain} data and label from cache.")
            self.data = torch.load(f"{root_path}/precessed/{domain}_data.pt")
            self.label = torch.load(f"{root_path}/precessed/{domain}_label.pt")
        else:
            print(f"Getting {domain} datasets")
            for index, label_name in enumerate(label_name_list):
                label_name_2_index = {
                    'dog': 0,
                    'elephant': 1,
                    'giraffe': 2,
                    'guitar': 3,
                    'horse': 4,
                    'house': 5,
                    'person': 6,
                }
                images_list = os.listdir(f"{self.root}/{label_name}")
                for img_name in images_list:
                    img = Image.open(f"{self.root}/{label_name}/{img_name}").convert('RGB')
                    img = np.array(img)
                    self.label.append(label_name_2_index[label_name])
                    if self.transform is not None:
                        img = self.transform(img)
                    self.data.append(img)
            self.data = torch.stack(self.data)
            self.label = torch.tensor(self.label, dtype=torch.long)
            torch.save(self.data, f"{root_path}/precessed/{domain}_data.pt")
            torch.save(self.label, f"{root_path}/precessed/{domain}_label.pt")

    def __getitem__(self, index):
        img, target = self.data[index], self.label[index]

        if self.transform is not None:
            img = self.transform(img)
        if self.target_transform is not None:
            target = self.target_transform(target)

        return img, target

    def __len__(self):
        return len(self.data)


def get_pacs_domain(root_path=f"{DATA_PATH}/PACS", domain='art_painting', verbose=False):
    transform = transforms.Compose([
        transforms.ToPILImage(),
        transforms.ToTensor(),
        # transforms.Resize((224, 224)),
        transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5))
    ])
    all_data = PACS(root_path, domain, transform=transform)
    # train:test=8:2
    x_train, x_test, y_train, y_test = train_test_split(all_data.data.numpy(), all_data.label.numpy(),
                                                        test_size=0.20, random_state=42)

    return x_train, y_train, x_test, y_test

其他写法：https://github.com/ValerioDiEugenio/DomainAdaptation-PACSDataset/blob/main/DomainAdaptation.ipynb

---

最后以dog类别为例，用Python代码展示四种不同风格的图片

Python可视化图片数据集的代码

修改dir_path为对应的文件夹，dir_path = f"{DATA_PATH}/PACS/art_painting/dog"

import os
import matplotlib.pyplot as plt
import random
from PIL import Image


def plotPics(data, h=3, w=3, filename="out.jpg"):
    fig, ax_array = plt.subplots(h, w, figsize=(15, 15))

    axes = ax_array.flatten()

    for i, ax in enumerate(axes):
        ri = random.randint(0, len(data) - 1)
        ax.imshow(data[ri], cmap=plt.cm.gray)

    plt.setp(axes, xticks=[], yticks=[], frame_on=False)
    fig.tight_layout()
    fig.savefig(filename)
    plt.show()


DATA_PATH = os.path.abspath(os.path.join(os.path.dirname(__file__), "../data/raw_data"))

dir_path = f"{DATA_PATH}/PACS/art_painting/dog"

data = []
for pic in os.listdir(dir_path):
    data.append(Image.open(f"{dir_path}/{pic}"))

plotPics(data, h=5, w=5)

art_painting风格

sketch风格

cartoon风格

photo风格