Replace train create_generators with Plugin impl

Replaced entire create_generators routine with PluginManager based
system. Includes error handling for plugins which may not conform to API
entirely.
Dictionary based capture to support extensibility.
Additionally, changed plugin path to more correctly represent project
root.

keras_retinanet/bin/train.py

@@ -37,8 +37,6 @@
 from .. import layers
 from ..callbacks import RedirectModel
 from ..callbacks.eval import Evaluate
-from ..preprocessing.pascal_voc import PascalVocGenerator
-from ..preprocessing.csv_generator import CSVGenerator
 from ..models.resnet import resnet50_retinanet, custom_objects
 from ..utils.transform import random_transform_generator
 from ..utils.keras_version import check_keras_version
@@ -121,55 +119,25 @@ def create_generators(args):
     # create random transform generator for augmenting training data
     transform_generator = random_transform_generator(flip_x_chance=0.5)
 
-    if args.dataset_type == 'coco':
-        # import here to prevent unnecessary dependency on cocoapi
-        from ..preprocessing.coco import CocoGenerator
+    generators = None
 
-        train_generator = CocoGenerator(
-            args.coco_path,
-            'train2017',
-            transform_generator=transform_generator,
-            batch_size=args.batch_size
-        )
-
-        validation_generator = CocoGenerator(
-            args.coco_path,
-            'val2017',
-            batch_size=args.batch_size
-        )
-    elif args.dataset_type == 'pascal':
-        train_generator = PascalVocGenerator(
-            args.pascal_path,
-            'trainval',
-            transform_generator=transform_generator,
-            batch_size=args.batch_size
-        )
-
-        validation_generator = PascalVocGenerator(
-            args.pascal_path,
-            'test',
-            batch_size=args.batch_size
-        )
-    elif args.dataset_type == 'csv':
-        train_generator = CSVGenerator(
-            args.annotations,
-            args.classes,
-            transform_generator=transform_generator,
-            batch_size=args.batch_size
-        )
-
-        if args.val_annotations:
-            validation_generator = CSVGenerator(
-                args.val_annotations,
-                args.classes,
-                batch_size=args.batch_size
-            )
+    for plugin in PluginManagerSingleton.get().getAllPlugins():
+        try:
+            getattr(plugin.plugin_object, 'dataset_type')
+        except AttributeError:
+            print("Plugin ({}) does not contain a definition for dataset_type and cannot be utilised.".format(plugin.name))
         else:
-            validation_generator = None
-    else:
+            if plugin.plugin_object.dataset_type.lower() == args.dataset_type.lower():
+                if callable(getattr(plugin.plugin_object, 'create_generators', None)):
+                    generators = plugin.plugin_object.create_generators(args, transform_generator=transform_generator)
+                    break
+                else:
+                    print("Plugin ({}) does not contain a definition for get_generator and cannot be utilised".format(plugin.name))
+
+    if generators is None:
         raise ValueError('Invalid data type received: {}'.format(args.dataset_type))
 
-    return train_generator, validation_generator
+    return generators["train_generator"], generators["validation_generator"]
 
 
 def check_args(parsed_args):
@@ -247,11 +215,10 @@ def load_plugins(plugin_path):
         print("Loaded: {}".format(n.name))
 
 def main(args=None):
-
     #Load plugins first, as their procedures are needed for parsing args.
     print("Loading plugins...")
     # Load Plugins
-    load_plugins(['keras_retinanet/plugins'])
+    load_plugins(['plugins'])
     print("Loaded plugins.")
 
     # parse arguments