**VOCtrainval_25-May-2011.tar** downloaded with `VOCSegmentation()` and `VOCDetection()` is extracted to `data/` as **TrainVal/VOCdevkit/VOC2011/** but not to `data/VOCdevkit/` as **VOC2011/**

🐛 Describe the bug

VOCtrainval_25-May-2011.tar downloaded with VOCSegmentation() and VOCDetection() is extracted to data/ as TrainVal/VOCdevkit/VOC2011/ but not to data/VOCdevkit/ as VOC2011/ as shown below, then TrainVal/VOCdevkit/VOC2011 is loaded from data/. *Other zip files are properly extracted to and loaded from data/VOCdevkit/:

from torchvision import datasets

train_data = datasets.VOCSegmentation(
    root="data",
    year="2011",
    image_set="trainval",
    download=True,
)

from torchvision import datasets

train_data = datasets.VOCDetection(
    root="data",
    year="2011",
    image_set="trainval",
    download=True,
)

data
 |-VOCdevkit
 |  |-VOC2007
 |  |-VOC2008
 |  |-VOC2009
 |  |-VOC2010
 |  └-VOC2012
 └-TrainVal # Here
    └-VOCdevkit
       └-VOC2011

In addition, I moved VOC2011/ from data/TrainVal/VOCdevkit/ to data/VOCdevkit/, then it was not loaded:

data
 |-VOCdevkit
 |  |-VOC2007
 |  |-VOC2008
 |  |-VOC2009
 |  |-VOC2010
 |  |-VOC2011<---┐
 |  └-VOC2012    |
 └-TrainVal      |
    └-VOCdevkit  |
       └-       -┘

Versions

import torchvision

torchvision.__version__ # '0.20.1'

Start by running the VOCSegmentation and VOCDetection examples with year="2011", image_set="trainval", and download=True, then inspect the dataset download and extraction entry points. Done means VOC2011 is extracted under data/VOCdevkit/ and is subsequently loaded from that location, consistently with the other VOC years.