module 'torchvision.datasets' has no attribute 'VOCDetection'

module ‘torchvision.datasets’ has no attribute ‘VOCDetection’

这是因为 VOCDetection 还没有添加到最新的 release 版本的导致的错误, 我们可以通过源码的方式重新安装 torchvision. 方法如下:

首先查看当前虚拟环境的 torchvision 的安装位置:

import torchvision as tv
print(tv.__file__)
# /home/zerozone/.pyenv/versions/a3py3.5/lib/python3.5/site-packages/torchvision/__init__.py

然后进入上面的文件夹, 删除旧的 torchvision

cd /home/zerozone/.pyenv/versions/a3py3.5/lib/python3.5/site-packages/
rm -rf torchvision*

然后下载最新版本的 torchvision 并安装(注意不要更换安装路径)

git clone https://github.com/pytorch/vision.git
python setup.py install

最后查看新安装的 torchvision 中是否包含 VOCDetection:

>>> import torchvision as tv
>>> print(dir(tv.datasets))
# ['CIFAR10', 'CIFAR100', 'CocoCaptions', 'CocoDetection', 'DatasetFolder', 'EMNIST', 'FakeData', 'FashionMNIST', 'Flickr30k', 'Flickr8k', 'ImageFolder', 'LSUN', 'LSUNClass', 'MNIST', 'Omniglot', 'PhotoTour', 'SBU', 'SEMEION', 'STL10', 'SVHN', 'VOCDetection', 'VOCSegmentation', '__all__', '__builtins__', '__cached__', '__doc__', '__file__', '__loader__', '__name__', '__package__', '__path__', '__spec__', 'cifar', 'coco', 'fakedata', 'flickr', 'folder', 'lsun', 'mnist', 'omniglot', 'phototour', 'sbu', 'semeion', 'stl10', 'svhn', 'utils', 'voc']