snapshot..

modules/deeplearning/icing_fcn.py

@@ -20,8 +20,8 @@ if NumClasses == 2:
 else:
     NumLogits = NumClasses
 
-BATCH_SIZE = 128
-NUM_EPOCHS = 50
+BATCH_SIZE = 64
+NUM_EPOCHS = 80
 
 TRACK_MOVING_AVERAGE = False
 EARLY_STOP = True
@@ -317,9 +317,17 @@ class IcingIntensityFCN:
             data_alt_lr = np.copy(data_alt)
             label_lr = np.copy(label)
 
-            data = np.concatenate([data, data_ud, data_lr])
-            data_alt = np.concatenate([data_alt, data_alt_ud, data_alt_lr])
-            label = np.concatenate([label, label_ud, label_lr])
+            data_r1 = np.rot90(data, k=1)
+            data_alt_r1 = np.copy(data_alt)
+            label_r1 = np.copy(label)
+
+            data_r2 = np.rot90(data, k=1)
+            data_alt_r2 = np.copy(data_alt)
+            label_r2 = np.copy(label)
+
+            data = np.concatenate([data, data_ud, data_lr, data_r1, data_r2])
+            data_alt = np.concatenate([data_alt, data_alt_ud, data_alt_lr, data_alt_r1, data_alt_r2])
+            label = np.concatenate([label, label_ud, label_lr, label_r1, label_r2])
 
         return data, data_alt, label
 
@@ -721,7 +729,7 @@ class IcingIntensityFCN:
             self.loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False)  # For multi-class
 
         # decayed_learning_rate = learning_rate * decay_rate ^ (global_step / decay_steps)
-        initial_learning_rate = 0.002
+        initial_learning_rate = 0.0006
         decay_rate = 0.95
         steps_per_epoch = int(self.num_data_samples/BATCH_SIZE)  # one epoch
         decay_steps = int(steps_per_epoch)