- minor changes

mini_imagenet_dataloader.py

@@ -31,6 +31,8 @@
 parser.add_argument('--shuffle_images', dest='shuffle_images', action='store_true')
 parser.add_argument('--enable_idx_increase', dest='enable_idx_increase', action='store_true')
 parser.add_argument('--use_independent_base', dest='use_independent_base', action='store_true')
+parser.add_argument('--train_indep_and_dependent', dest='train_indep_and_dependent', action='store_true')
+parser.add_argument('--tensorboard_log_dir', dest='tensorboard_log_dir',  type=str, default='./logs')
 args = parser.parse_args()
 
 class MiniImageNetDataLoader(object):
@@ -264,7 +266,9 @@ def get_batch(self, phase='train', idx=0):
 base_train_img, base_train_label, base_test_img, base_test_label = \
         dataloader.get_batch(phase='train', idx=0) 
 
-train_epoch_size = int(base_train_img.shape[0] / 2) # as double is generated for the base and train 
+train_epoch_size = base_train_img.shape[0]
+if not args.train_indep_and_dependent:
+    train_epoch_size = int(train_epoch_size / 2) # as double is generated for the base and train
 test_epoch_size = base_test_img.shape[0]
 
 print("epoch training size:", train_epoch_size, base_train_label.shape[0], "epoch testing size", test_epoch_size)
@@ -291,23 +295,25 @@ def generate(self, phase='train'):
     def generate_add_samples(self, phase = 'train'):
         self.idx = 0
         while True:
-            episode_train_img, episode_train_label, episode_test_img, episode_test_label = \
+            batch_train_img, batch_train_label, episode_test_img, episode_test_label = \
                 dataloader.get_batch(phase=args.dataset, idx=self.idx)
 
             # this depends on what we are trying to train.
             # care must be taken, that with a different dataset the labels have a different meaning. Thus if we use a new dataset, we must 
             # use network_base which fits to the database. Therefore there must be taken images with label from the same dataset.
-            network_base_img = episode_train_img[:train_epoch_size]
-            network_base_img = episode_train_label[:train_epoch_size]
+            network_base_img = batch_train_img[:train_epoch_size]
+            network_base_img = batch_train_label[:train_epoch_size]
 
             #only half is used now, as the rest is reserved for independend base    
-            episode_train_img = episode_train_img[train_epoch_size:]
-            episode_train_label = episode_train_label[train_epoch_size:]
+            episode_train_img = batch_train_img[train_epoch_size:]
+            episode_train_label = batch_train_label[train_epoch_size:]
 
             if not args.use_independent_base:
                 network_base_img = episode_train_img
                 network_base_label = episode_train_label
-
+            if args.train_indep_and_dependent:
+                episode_train_img = batch_train_img
+                episode_train_label = batch_train_label
             if phase == 'train':
                 if args.enable_idx_increase:
                     self.idx += 1 # only train phase allowed to change
@@ -490,7 +496,7 @@ def all_layers(model):
 
 
 checkpointer = ModelCheckpoint(filepath='checkpoints/model-{epoch:02d}.hdf5', verbose=1)
-tensorboard = TensorBoard()
+tensorboard = TensorBoard(log_dir = args.tensorboard_log_dir)
 lambda_model.fit_generator(keras_gen_train.generate_add_samples(), train_epoch_size, args.epochs, 
                            validation_data=keras_gen_train.generate_add_samples('test'), validation_steps=test_epoch_size, callbacks = [tensorboard], workers = 0) 
 #workers = 0 is a work around to correct the number of calls to the validation_data generator