Merge 5d16cc4 into 406a931

.travis.yml

@@ -7,12 +7,13 @@ git:
 language: python
 
 python:
-  - 3.6
+  # - 3.6
   - 3.7
 
 env:
   - TF_VERSION=2.1.1
   - TF_VERSION=2.2.0
+  - TF_VERSION=2.3.0
 
 install:
   # code below is taken from https://github.com/keras-team/keras/blob/master/.travis.yml

Dockerfile

@@ -1,9 +1,9 @@
 # Use tensorflow/tensorflow as the base image
 # Change the build arg to edit the tensorflow version.
 # Only supporting python3.
-ARG TF_VERSION=2.1.0-gpu
+ARG TF_VERSION=2.1.1-gpu
 
-FROM tensorflow/tensorflow:${TF_VERSION}-py3
+FROM tensorflow/tensorflow:${TF_VERSION}
 
 # System maintenance
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \

deepcell/applications/cell_tracking.py

@@ -67,7 +67,7 @@ def CellTrackingModel(input_shape=(32, 32, 1),
             'CellTrackingModel.h5',
             WEIGHTS_PATH,
             cache_subdir='models',
-            md5_hash='3349b363fdad0266a1845ba785e057a6')
+            file_hash='3349b363fdad0266a1845ba785e057a6')
 
         model.load_weights(weights_path)
 

deepcell/applications/cytoplasm_segmentation.py

@@ -123,7 +123,7 @@ def __init__(self,
                 os.path.basename(WEIGHTS_PATH),
                 WEIGHTS_PATH,
                 cache_subdir='models',
-                md5_hash='104a7d7884c80c37d2bce6d1c3a17c7a'
+                file_hash='104a7d7884c80c37d2bce6d1c3a17c7a'
             )
 
             model.load_weights(weights_path, by_name=True)

deepcell/applications/label_detection.py

@@ -103,7 +103,7 @@ def LabelDetectionModel(input_shape=(None, None, 1),
                 local_name,
                 MOBILENETV2_WEIGHTS_PATH,
                 cache_subdir='models',
-                md5_hash='14d4b2f7c77d334c958d2dde79972e6e')
+                file_hash='14d4b2f7c77d334c958d2dde79972e6e')
         else:
             raise ValueError('Backbone %s does not have a weights file.' %
                              backbone)

deepcell/applications/multiplex_segmentation.py

@@ -33,15 +33,15 @@
 
 from tensorflow.keras.utils import get_file
 
-from deepcell_toolbox.deep_watershed import deep_watershed_subcellular, format_output_multiplex
-from deepcell_toolbox.processing import phase_preprocess
+from deepcell_toolbox.multiplex_utils import \
+    multiplex_preprocess, multiplex_postprocess, format_output_multiplex
 
 from deepcell.applications import Application
 from deepcell.model_zoo import PanopticNet
 
 
 WEIGHTS_PATH = ('https://deepcell-data.s3-us-west-1.amazonaws.com/'
-                'model-weights/Multiplex_Segmentation_20200714_subcellular.h5')
+                'model-weights/Multiplex_Segmentation_20200908_2_head.h5')
 
 
 class MultiplexSegmentation(Application):
@@ -109,8 +109,8 @@ def __init__(self,
                  use_pretrained_weights=True,
                  model_image_shape=(256, 256, 2)):
 
-        whole_cell_classes = [1, 1, 2, 3]
-        nuclear_classes = [1, 1, 2, 3]
+        whole_cell_classes = [1, 3]
+        nuclear_classes = [1, 3]
         num_semantic_classes = whole_cell_classes + nuclear_classes
         num_semantic_heads = len(num_semantic_classes)
 
@@ -128,7 +128,7 @@ def __init__(self,
                 os.path.basename(WEIGHTS_PATH),
                 WEIGHTS_PATH,
                 cache_subdir='models',
-                md5_hash='a2f26a7b5cc9a86d68e65a7bd27c32d0'
+                file_hash='4e440b0e329dd5c24c1162efa0a33bc9'
             )
 
             model.load_weights(weights_path)
@@ -138,8 +138,8 @@ def __init__(self,
         super(MultiplexSegmentation, self).__init__(model,
                                                     model_image_shape=model_image_shape,
                                                     model_mpp=0.5,
-                                                    preprocessing_fn=phase_preprocess,
-                                                    postprocessing_fn=deep_watershed_subcellular,
+                                                    preprocessing_fn=multiplex_preprocess,
+                                                    postprocessing_fn=multiplex_postprocess,
                                                     format_model_output_fn=format_output_multiplex,
                                                     dataset_metadata=self.dataset_metadata,
                                                     model_metadata=self.model_metadata)
@@ -150,8 +150,8 @@ def predict(self,
                 image_mpp=None,
                 preprocess_kwargs={},
                 compartment='whole-cell',
-                postprocess_kwargs_whole_cell={},
-                postprocess_kwargs_nuclear={}):
+                postprocess_kwargs_whole_cell=None,
+                postprocess_kwargs_nuclear=None):
         """Generates a labeled image of the input running prediction with
         appropriate pre and post processing functions.
 
@@ -182,6 +182,20 @@ def predict(self,
             np.array: Model output
         """
 
+        if postprocess_kwargs_whole_cell is None:
+            postprocess_kwargs_whole_cell = {'maxima_threshold': 0.1, 'maxima_model_smooth': 0,
+                                             'interior_threshold': 0.3, 'interior_model_smooth': 2,
+                                             'small_objects_threshold': 15,
+                                             'fill_holes_threshold': 15,
+                                             'radius': 2}
+
+        if postprocess_kwargs_nuclear is None:
+            postprocess_kwargs_nuclear = {'maxima_threshold': 0.1, 'maxima_model_smooth': 0,
+                                          'interior_threshold': 0.3, 'interior_model_smooth': 2,
+                                          'small_objects_threshold': 15,
+                                          'fill_holes_threshold': 15,
+                                          'radius': 2}
+
         # create dict to hold all of the post-processing kwargs
         postprocess_kwargs = {'whole_cell_kwargs': postprocess_kwargs_whole_cell,
                               'nuclear_kwargs': postprocess_kwargs_nuclear,

deepcell/applications/multiplex_segmentation_test.py

@@ -44,7 +44,7 @@ def test_multiplex_app(self):
             # test output shape
             shape = app.model.output_shape
             self.assertIsInstance(shape, list)
-            self.assertEqual(len(shape), 8)
+            self.assertEqual(len(shape), 4)
 
             # test predict with default
             x = np.random.rand(1, 500, 500, 2)

deepcell/applications/nuclear_segmentation.py

@@ -123,7 +123,7 @@ def __init__(self,
                 os.path.basename(WEIGHTS_PATH),
                 WEIGHTS_PATH,
                 cache_subdir='models',
-                md5_hash='42ca0ebe4b7b0f782eaa4733cdddad88'
+                file_hash='42ca0ebe4b7b0f782eaa4733cdddad88'
             )
 
             model.load_weights(weights_path, by_name=True)

deepcell/applications/scale_detection.py

@@ -102,7 +102,7 @@ def ScaleDetectionModel(input_shape=(None, None, 1),
                 local_name,
                 MOBILENETV2_WEIGHTS_PATH,
                 cache_subdir='models',
-                md5_hash='aa78e6b9a4551289dd967f1f5ca83fed')
+                file_hash='aa78e6b9a4551289dd967f1f5ca83fed')
         else:
             raise ValueError('Backbone %s does not have a weights file.' %
                              backbone)

deepcell/datasets/__init__.py

@@ -161,6 +161,14 @@ def load_tracked_data(self, path=None, test_size=0.2, seed=0):
     metadata={}
 )
 
+#:
+multiplex_tissue = Dataset(
+    path='20200810_tissue_dataset.npz',
+    url='https://deepcell-data.s3.amazonaws.com/multiplex/20200810_tissue_dataset.npz',
+    file_hash='1e573b72123fd86e45433402094bf0d0',
+    metadata={}
+)
+
 # pylint: disable=wrong-import-position
 from deepcell.datasets import cytoplasm
 from deepcell.datasets import phase

deepcell/image_generators/__init__.py

@@ -124,9 +124,17 @@ def _transform_masks(y, transform, data_format=None, **kwargs):
                           '`outer-distance` transform instead.'.format(transform),
                           DeprecationWarning)
 
-        bins = kwargs.pop('distance_bins', None)
-        erosion = kwargs.pop('erosion_width', 0)
         by_frame = kwargs.pop('by_frame', True)
+        bins = kwargs.pop('distance_bins', None)
+
+        distance_kwargs = {
+            'bins': bins,
+            'erosion_width': kwargs.pop('erosion_width', 0),
+        }
+
+        # If using 3d transform, pass in scale arg
+        if y.ndim == 5 and not by_frame:
+            distance_kwargs['sampling'] = kwargs.pop('sampling', [0.5, 0.217, 0.217])
 
         if data_format == 'channels_first':
             y_transform = np.zeros(tuple([y.shape[0]] + list(y.shape[2:])))
@@ -146,9 +154,7 @@ def _transform_masks(y, transform, data_format=None, **kwargs):
                 mask = y[batch, 0, ...]
             else:
                 mask = y[batch, ..., 0]
-
-            y_transform[batch] = _distance_transform(
-                mask, bins=bins, erosion_width=erosion)
+            y_transform[batch] = _distance_transform(mask, **distance_kwargs)
 
         y_transform = np.expand_dims(y_transform, axis=-1)
 
@@ -166,11 +172,19 @@ def _transform_masks(y, transform, data_format=None, **kwargs):
                           '`inner-distance` transform instead.'.format(transform),
                           DeprecationWarning)
 
-        bins = kwargs.pop('distance_bins', None)
-        erosion = kwargs.pop('erosion_width', 0)
         by_frame = kwargs.pop('by_frame', True)
-        alpha = kwargs.pop('alpha', 0.1)
-        beta = kwargs.pop('beta', 1)
+        bins = kwargs.pop('distance_bins', None)
+
+        distance_kwargs = {
+            'bins': bins,
+            'erosion_width': kwargs.pop('erosion_width', 0),
+            'alpha': kwargs.pop('alpha', 0.1),
+            'beta': kwargs.pop('beta', 1)
+        }
+
+        # If using 3d transform, pass in scale arg
+        if y.ndim == 5 and not by_frame:
+            distance_kwargs['sampling'] = kwargs.pop('sampling', [0.5, 0.217, 0.217])
 
         if data_format == 'channels_first':
             y_transform = np.zeros(tuple([y.shape[0]] + list(y.shape[2:])))
@@ -190,14 +204,11 @@ def _transform_masks(y, transform, data_format=None, **kwargs):
                 mask = y[batch, 0, ...]
             else:
                 mask = y[batch, ..., 0]
-
-            y_transform[batch] = _distance_transform(mask, bins=bins,
-                                                     erosion_width=erosion,
-                                                     alpha=alpha, beta=beta)
+            y_transform[batch] = _distance_transform(mask, **distance_kwargs)
 
         y_transform = np.expand_dims(y_transform, axis=-1)
 
-        if bins is None:
+        if distance_kwargs['bins'] is None:
             pass
         else:
             # convert to one hot notation
@@ -244,6 +255,9 @@ def _transform_masks(y, transform, data_format=None, **kwargs):
 from deepcell.image_generators.semantic import SemanticMovieGenerator
 from deepcell.image_generators.semantic import SemanticMovieIterator
 
+from deepcell.image_generators.semantic import Semantic3DGenerator
+from deepcell.image_generators.semantic import Semantic3DIterator
+
 from deepcell.image_generators.sample import SampleDataGenerator
 from deepcell.image_generators.sample import ImageSampleArrayIterator
 from deepcell.image_generators.sample import SampleMovieDataGenerator