added PyTorch and Keras DataGenerators (#52)

* added PyTorch and Keras DataGenerators

* adding pyyaml to reqts

* fixing keras imports

* fixing model_io import

environment.yml

@@ -21,6 +21,7 @@ dependencies:
   - scikit-image=0.15.0
   - tensorflow=1.13.1
   - pytorch=1.0.1
+  - pyyaml=5.1
   - pip:
     - affine==2.2.2
     - albumentations==0.2.3

readthedocs-environment.yml

@@ -17,6 +17,7 @@ dependencies:
   - scikit-image=0.15.dev0
   - tensorflow=1.13.1
   - pytorch=1.0.1
+  - pyyaml=5.1
   - pip:
     - affine=2.2.1
     - albumentations=0.2.2

requirements.txt

@@ -16,3 +16,4 @@ torch==1.0.1.post2
 affine==2.2.2
 albumentations==0.2.3
 rio-tiler==1.2.4
+pyyaml==5.1

setup.py

@@ -17,7 +17,8 @@
              'torch>=1.0.1',
              'affine>=2.2.1',
              'albumentations>=0.2.2',
-             'rio-tiler>=1.2.4'
+             'rio-tiler>=1.2.4',
+             'pyyaml>=5.1'
              ]
 
 version = '0.0.1'

solaris/data/config_skeleton.yml

@@ -1,5 +1,5 @@
 ################################################################################
-################# CW-NETS MODEL CONFIGURATION SKELETON #########################
+################# SOLARIS MODEL CONFIGURATION SKELETON #########################
 ################################################################################
 
 # This skeleton lays out the required instructions for running a model using
@@ -20,8 +20,11 @@ nn_framework:  # if not using a model included with the package, use this
 data_specs:
   width: # width of the input images taken in by the neural net.
   height:  # height of the input images taken in by the neural net.
-  bit_depth:  # 8 or 16.
+  image_type: normalized # format of images. options are 'normalized',
+                         # 'zscore', '8bit', '16bit'
   channels:  # number of channels in the input imagery.
+  label_type: mask  # one of ['mask', 'bbox']
+  mask_channels: 1  # number of channels in the training mask
   val_holdout_frac:  # if empty, assumes that separate data ref files define the
                      # training and validation dataset. If a float between 0 and
                      # 1, indicates the fraction of training data that's held
@@ -48,6 +51,7 @@ training_augmentation:  # augmentations for use with training data
     # include augmentations here. See the documentation for options and
     # required arguments.
   p: 1.0  # probability of applying the entire training augmentation pipeline.
+  shuffle: true  # should the image order be shuffled in each epoch.
 validation_augmentation:  # augmentations for use with validation data
   augmentations:
     # include augmentations here

solaris/nets/callbacks.py

@@ -1,4 +1,4 @@
-from keras.callbacks import Callback
+from tensorflow.keras.callbacks import Callback
 import numpy as np