Pytorch中使用ImageFolder读取数据集时忽略特定文件

import platform
from torchvision.datasets import ImageFolder


class Check(object):
    def __init__(self,
                 key_word: str):
        self.key_word = key_word
        self.separator = '\\' if platform.system() == 'Windows' else '/'

    def __call__(self, 
                 file_name: str) -> bool:
        folders = file_name.split(self.separator)
        return folders[-1].find(self.key_word) < 0

dataset = ImageFolder('./data', is_valid_file=Check('invalid'))

from torchvision.datasets.folder import *
from typing import *


class FilterableImageFolder(ImageFolder):
    def __init__(
            self,
            root: str,
            transform: Optional[Callable] = None,
            target_transform: Optional[Callable] = None,
            loader: Callable[[str], Any] = default_loader,
            is_valid_file: Optional[Callable[[str], bool]] = None,
            valid_classes: List = None
    ):
        self.valid_classes = valid_classes
        super(FilterableImageFolder, self).__init__(root, transform, target_transform, loader, is_valid_file)

    def find_classes(self, directory: str) -> Tuple[List[str], Dict[str, int]]:
        classes = sorted(entry.name for entry in os.scandir(directory) if entry.is_dir())
        #增加了这下面这句
        classes = [valid_class for valid_class in classes if valid_class in self.valid_classes]
        if not classes:
            raise FileNotFoundError(f"Couldn't find any class folder in {directory}.")

        class_to_idx = {cls_name: i for i, cls_name in enumerate(classes)}
        return classes, class_to_idx

dataset = FilterableImageFolder('./data', valid_classes=['cat', 'dog'])