jupyter notebook中使用ImageFolder来加载文件夹时把.ipynb_checkpoints文件夹忽略

from torchvision.datasets import ImageFolder
dataset = ImageFolder('data/dogcat_2/')

dataset.class_to_idx

输出：

{'.ipynb_checkpoints': 0, 'cat': 1, 'dog': 2}

将~/python3.6/site-packages/torchvision/datasets/folder.py的源代码改一下即：

将以下函数进行更改：

def find_classes(dir):
    classes = [d for d in os.listdir(dir) if os.path.isdir(os.path.join(dir, d)) ]
    classes.sort()
    class_to_idx = {classes[i]: i for i in range(len(classes))}
    return classes, class_to_idx

更改后为：

def find_classes(dir):
    classes = [d for d in os.listdir(dir) if os.path.isdir(os.path.join(dir, d)) and not d.startswith('.')]
    classes.sort()
    class_to_idx = {classes[i]: i for i in range(len(classes))}
    return classes, class_to_idx

from torchvision.datasets import ImageFolder
dataset = ImageFolder('data/dogcat_2/')

dataset.class_to_idx

输出：

{'cat': 0, 'dog': 1}