Merge branch 'master' into super-dev

README.rst

@@ -8,18 +8,13 @@ Inferno
 .. image:: https://img.shields.io/pypi/v/inferno.svg
         :target: https://pypi.python.org/pypi/inferno-pytorch
 
-.. image:: https://img.shields.io/travis/infern-pytorch/inferno.svg
+.. image:: https://travis-ci.org/inferno-pytorch/inferno.svg?branch=master
         :target: https://travis-ci.org/inferno-pytorch/inferno
 
 .. image:: https://readthedocs.org/projects/inferno-pytorch/badge/?version=latest
         :target: http://inferno-pytorch.readthedocs.io/en/latest/?badge=latest
         :alt: Documentation Status
 
-.. image:: https://pyup.io/repos/github/inferno-pytorch/inferno/shield.svg
-     :target: https://pyup.io/repos/github/inferno-pytorch/inferno/
-     :alt: Updates
-
-
 
 .. image:: http://svgshare.com/i/2j7.svg
 
@@ -34,7 +29,7 @@ It's a work-in-progress, but the first stable release (0.2) is underway!
 
 
 * Free software: Apache Software License 2.0
-* Documentation: https://pytorch-inferno.readthedocs.io (Work in progress).
+* Documentation: http://inferno-pytorch.readthedocs.io (Work in Progress).
 
 
 Features
@@ -131,11 +126,11 @@ and navigate to `localhost:6007` with your browser.
 Installation
 ------------------------
 
-Conda packages for linux and mac (only python 3) are available via
+Conda packages for linux (only python 3) are available via
 
 .. code:: bash
 
-  $ conda install -c inferno-pytorch inferno
+  $ conda install -c pytorch -c conda-forge inferno
 
 
 

inferno/io/transform/image.py

@@ -83,105 +83,6 @@ def image_function(self, image):
                 ShapeError)
         return rescaled_image
 
-class RandomScaleSegmentation(Transform):
-
-    """ Random Scale input and label image
-
-
-
-        Parameters
-
-        ----------
-
-        scale_range : tuple of floats defining (min, max) scales
-
-                      maximum angle of rotation
-
-        resize  : if True, image is cropped or padded to the original size
-
-    """
-
-    def __init__(self, scale_range, resize=True, pad_const=0, **super_kwargs):
-
-        super(RandomScaleSegmentation, self).__init__(**super_kwargs)
-
-        self.scale_range = scale_range
-
-        self.resize = resize
-
-        self.pad_const = pad_const
-
-
-
-    def build_random_variables(self):
-
-        np.random.seed()
-
-        self.set_random_variable('seg_scale',
-
-                 np.random.uniform(low=self.scale_range[0],
-
-                                   high=self.scale_range[1]))
-
-
-
-    def batch_function(self, image):
-
-        scale = self.get_random_variable('seg_scale')
-
-        image_shape = np.array(image[0].shape[1:])
-
-
-
-        with catch_warnings():
-
-            simplefilter('ignore')
-
-            img = np.stack([zoom(x, scale, order=3) for x in image[0]])
-
-            seg = np.stack([zoom(x, scale, order=0) for x in image[1]])
-
-        new_shape = np.array(img.shape[1:])
-
-
-
-        if self.resize:
-
-            if scale > 1.:
-
-                # pad image to original size
-
-                crop_l = (new_shape - image_shape) // 2
-
-                crop_r = new_shape - image_shape - crop_l
-
-                cropping = [slice(None)] + [slice(c[0] if c[0] > 0 else None,
-
-                                                 -c[1] if c[1] > 0 else None) for c in zip(crop_l, crop_r)]
-
-                img = img[cropping]
-
-                seg = seg[cropping]
-
-            else:
-
-                # crop image to original size
-
-                pad_l = (image_shape - new_shape) // 2
-
-                pad_r = image_shape - new_shape - pad_l
-
-                padding = [(0,0)] + list(zip(pad_l, pad_r))
-
-                img = np.pad(img, padding, 'mirror', constant_values=self.pad_const)
-
-
-
-                seg = np.pad(seg, padding, 'mirror', constant_values=self.pad_const)
-
-
-
-        return img, seg
 
 class RandomCrop(Transform):
     """Crop input to a given size.
@@ -649,6 +550,7 @@ class RandomScaleSegmentation(Transform):
         scale_range : tuple of floats defining (min, max) scales
                       maximum angle of rotation
         resize  : if True, image is cropped or padded to the original size
+        pad_const: value used for constant padding
     """
     def __init__(self, scale_range, resize=True, pad_const=0, **super_kwargs):
         super(RandomScaleSegmentation, self).__init__(**super_kwargs)
@@ -687,7 +589,7 @@ def batch_function(self, image):
                 pad_r = image_shape - new_shape - pad_l
                 padding = [(0,0)] + list(zip(pad_l, pad_r))
                 img = np.pad(img, padding, 'constant', constant_values=self.pad_const)
-                
+
                 seg = np.pad(seg, padding, 'constant', constant_values=self.pad_const)
 
         return img, seg

inferno/trainers/basic.py

@@ -533,7 +533,7 @@ def build_metric(self, method, **kwargs):
         elif isinstance(method, str):
             assert hasattr(metrics, method), \
                 "Could not find the metric '{}'.".format(method)
-            self._metric = getattr(metrics, method)()
+            self._metric = getattr(metrics, method)(**kwargs)
         else:
             raise NotImplementedError
         return self
@@ -1657,7 +1657,7 @@ def save_model(self, to_directory=None):
                    pickle_module=self.pickle_module)
         return self
 
-    def load(self, from_directory=None, best=False, filename=None):
+    def load(self, from_directory=None, best=False, filename=None, map_location=None):
         """
         Load the trainer from checkpoint.
 
@@ -1671,6 +1671,8 @@ def load(self, from_directory=None, best=False, filename=None):
             'best_checkpoint.pytorch'.
         filename : str
             Overrides the default filename.
+        device : function, torch.device, string or a dict
+            Specify how to remap storage locations.
 
         Returns
         -------
@@ -1684,7 +1686,8 @@ def load(self, from_directory=None, best=False, filename=None):
             filename = self._best_checkpoint_filename if best else self._checkpoint_filename
         # Load the dictionary
         config_dict = torch.load(os.path.join(from_directory, filename),
-                                 pickle_module=self.pickle_module)
+                                 pickle_module=self.pickle_module, map_location=map_location)
+
         # This is required to prevent an infinite save loop?
         self._is_iteration_with_best_validation_score = False
         # Set config