add documentation for ImageNet dataset (#859)

docs/source/datasets.rst

@@ -7,7 +7,7 @@ Hence, they can all be passed to a :class:`torch.utils.data.DataLoader`
 which can load multiple samples parallelly using ``torch.multiprocessing`` workers.
 For example: ::
 
-    imagenet_data = torchvision.datasets.ImageFolder('path/to/imagenet_root/')
+    imagenet_data = torchvision.datasets.ImageNet('path/to/imagenet_root/')
     data_loader = torch.utils.data.DataLoader(imagenet_data,
                                               batch_size=4,
                                               shuffle=True,
@@ -100,12 +100,11 @@ DatasetFolder
 Imagenet-12
 ~~~~~~~~~~~
 
-This should simply be implemented with an ``ImageFolder`` dataset.
-The data is preprocessed `as described
-here <https://github.com/facebook/fb.resnet.torch/blob/master/INSTALL.md#download-the-imagenet-dataset>`__
+.. autoclass:: ImageNet
+
+.. note ::
+    This requires `scipy` to be installed
 
-`Here is an
-example <https://github.com/pytorch/examples/blob/e0d33a69bec3eb4096c265451dbb85975eb961ea/imagenet/main.py#L113-L126>`__.
 
 CIFAR
 ~~~~~