Add ssTEM dataset

docs/source/datasets.rst

@@ -188,3 +188,13 @@ SBD
 .. autoclass:: SBDataset
   :members: __getitem__
   :special-members:
+
+ssTEM
+~~~~~
+
+.. note ::
+    This requires `skimage` to be installed
+
+.. autoclass:: ssTEM
+   :members: __getitem__
+   :special-members:

torchvision/datasets/__init__.py

@@ -18,6 +18,7 @@
 from .celeba import CelebA
 from .sbd import SBDataset
 from .vision import VisionDataset
+from .sstem import ssTEM
 
 __all__ = ('LSUN', 'LSUNClass',
            'ImageFolder', 'DatasetFolder', 'FakeData',

torchvision/datasets/sstem.py

@@ -0,0 +1,129 @@
+from __future__ import print_function
+
+import os
+
+from PIL import Image
+import torch.utils.data as data
+
+from .utils import download_url, check_integrity
+from .vision import VisionDataset
+
+
+class ssTEM(VisionDataset):
+    """Dataset for `ISBI Challenge: Segmentation of neuronal structures
+    in EM stacks <http://brainiac2.mit.edu/isbi_challenge/>`_.
+
+    Args:
+        root (string): Root directory where dataset exists or will be saved
+            to if download is set to True.
+        train (bool, optional): If True, creates dataset from training set,
+            otherwise creates from test set.
+        transforms (callable, optional): A function/transform that takes in
+            two PIL images and returns transformed versions.
+        download (bool, optional): If true, downloads the dataset from the
+            internet and puts it in root directory. If dataset is already
+            downloaded, it is not downloaded again.
+    """
+    base_url = 'http://brainiac2.mit.edu/isbi_challenge/sites/default/files/'
+
+    meta = {
+        'train': {
+            'data': {
+                'filename': 'train-volume.tif',
+                'md5': '465461edbe0254630c4ec5577f1e7764'
+            },
+            'labels': {
+                'filename': 'train-labels.tif',
+                'md5': '657fe6b728c6dd0152e295c6d800001d'
+            }
+        },
+        'test': {
+            'data': {
+                'filename': 'test-volume.tif',
+                'md5': '9767660d7abe4e0ecbdd0061a16058ad'
+            }
+        }
+    }
+
+    def __init__(self, root, train=True, transforms=None, download=False):
+        # Lazy import
+        import skimage.io
+
+        super(ssTEM, self).__init__(root, transforms)
+
+        self.train = train
+
+        if download:
+            self.download()
+
+        if not self._check_integrity():
+            raise RuntimeError('Dataset not found or corrupted.' +
+                               ' You can use download=True to download it')
+
+        if train:
+            self.data = skimage.io.imread(os.path.join(
+                self.root, self.meta['train']['data']['filename']))
+            self.labels = skimage.io.imread(os.path.join(
+                self.root, self.meta['train']['labels']['filename']))
+        else:
+            self.data = skimage.io.imread(os.path.join(
+                self.root, self.meta['test']['data']['filename']))
+
+    def __getitem__(self, index):
+        """
+        Args:
+            index (int): Index
+        Returns:
+            tuple: (image, target) where target is index of the target class.
+        """
+        img = self.data[index]
+        img = Image.fromarray(img)
+
+        target = None
+        if self.train:
+            target = self.labels[index]
+            target = Image.fromarray(img)
+
+        if self.transforms is not None:
+            img, target = self.transforms(img, target)
+
+        return img, target
+
+    def __len__(self):
+        return len(self.data)
+
+    def _check_integrity(self):
+        if self.train:
+            dataset = 'train'
+            records = ['data', 'labels']
+        else:
+            dataset = 'test'
+            records = ['data']
+
+        for record in records:
+            filename = self.meta[dataset][record]['filename']
+            fpath = os.path.join(self.root, filename)
+            md5 = self.meta[dataset][record]['md5']
+
+            if not check_integrity(fpath, md5):
+                return False
+
+        return True
+
+    def download(self):
+        if self._check_integrity():
+            print('Files already downloaded and verified')
+            return
+
+        if self.train:
+            dataset = 'train'
+            records = ['data', 'labels']
+        else:
+            dataset = 'test'
+            records = ['data']
+
+        for record in records:
+            filename = self.meta[dataset][record]['filename']
+            md5 = self.meta[dataset][record]['md5']
+
+            download_url(self.base_url + filename, self.root, filename, md5)