Experimental run with CubeRun and MLSP 2013

  - separate train/validation : 230/92

Using TensorFlow backend.
Wed, 16 Nov 2016 17:04:16 +0000
Wed, 16 Nov 2016 17:04:45 +0000
X_train shape:  (230, 257, 624, 1)
Y_train shape:  (230, 19)
Train on 230 samples, validate on 92 samples
Epoch 1/20
230/230 [==============================] - 234s - loss: 0.2889 - acc: 0.9096 -
val_loss: 0.6004 - val_acc: 0.9382
Epoch 2/20
230/230 [==============================] - 280s - loss: 0.2331 - acc: 0.9357 -
val_loss: 0.4799 - val_acc: 0.9388
Epoch 3/20
230/230 [==============================] - 267s - loss: 0.2178 - acc: 0.9405 -
val_loss: 0.3822 - val_acc: 0.9336
Epoch 4/20
230/230 [==============================] - 263s - loss: 0.1884 - acc: 0.9467 -
val_loss: 0.3444 - val_acc: 0.9251
Epoch 5/20
230/230 [==============================] - 267s - loss: 0.1899 - acc: 0.9458 -
val_loss: 0.2372 - val_acc: 0.9354
Epoch 6/20
230/230 [==============================] - 259s - loss: 0.1735 - acc: 0.9490 -
val_loss: 0.2278 - val_acc: 0.9382
Epoch 7/20
230/230 [==============================] - 274s - loss: 0.1657 - acc: 0.9533 -
val_loss: 0.2417 - val_acc: 0.9388
Epoch 8/20
230/230 [==============================] - 290s - loss: 0.1501 - acc: 0.9545 -
val_loss: 0.2337 - val_acc: 0.9399
Epoch 9/20
230/230 [==============================] - 282s - loss: 0.1414 - acc: 0.9604 -
val_loss: 0.2455 - val_acc: 0.9319
Epoch 10/20
230/230 [==============================] - 281s - loss: 0.1222 - acc: 0.9613 -
val_loss: 0.2539 - val_acc: 0.9371
Epoch 11/20
230/230 [==============================] - 285s - loss: 0.1108 - acc: 0.9689 -
val_loss: 0.3698 - val_acc: 0.9416
Epoch 12/20
230/230 [==============================] - 289s - loss: 0.0974 - acc: 0.9698 -
val_loss: 0.3457 - val_acc: 0.9365
Epoch 13/20
230/230 [==============================] - 287s - loss: 0.0939 - acc: 0.9712 -
val_loss: 0.2873 - val_acc: 0.9405
Epoch 14/20
230/230 [==============================] - 269s - loss: 0.0789 - acc: 0.9748 -
val_loss: 0.3948 - val_acc: 0.9394
Epoch 15/20
230/230 [==============================] - 266s - loss: 0.0611 - acc: 0.9808 -
val_loss: 0.3758 - val_acc: 0.9399
Epoch 16/20
230/230 [==============================] - 259s - loss: 0.0582 - acc: 0.9817 -
val_loss: 0.3478 - val_acc: 0.9405
Epoch 17/20
230/230 [==============================] - 259s - loss: 0.0450 - acc: 0.9842 -
val_loss: 0.2778 - val_acc: 0.9411
Epoch 18/20
230/230 [==============================] - 275s - loss: 0.0380 - acc: 0.9870 -
val_loss: 0.4107 - val_acc: 0.9434
Epoch 19/20
230/230 [==============================] - 287s - loss: 0.0347 - acc: 0.9888 -
val_loss: 0.2777 - val_acc: 0.9354
Epoch 20/20
230/230 [==============================] - 288s - loss: 0.0308 - acc: 0.9911 -
val_loss: 0.4374 - val_acc: 0.9411
Wed, 16 Nov 2016 18:35:59 +0000
Wed, 16 Nov 2016 18:36:00 +0000
The weights have been saved in: ../weights/2016_11_16_18:35:59_cuberun.h5

bird/train.py

@@ -6,7 +6,7 @@
 import loader
 
 # Settings
-nb_epoch = 60
+nb_epoch = 20
 nb_classes = 19
 #nb_classes = 2
 batch_size = 8
@@ -72,8 +72,8 @@
 #print("Validation class dict: ", validation_class_dict.class_indices)
 ############################################################################
 
-X_valid, Y_valid = loader.load_data(train_path, labels_path, size=100,
-                                    nb_classes=nb_classes, image_shape=(cols, rows))
+#X_valid, Y_valid = loader.load_data(train_path, labels_path, size=100,
+                                    #nb_classes=nb_classes, image_shape=(cols, rows))
 
 
 # Setup compile
@@ -95,9 +95,16 @@
 #for e in range(nb_epoch):
     #print("epoch %d" % e)
 print(strftime("%a, %d %b %Y %H:%M:%S +0000", localtime()))
-X_train, Y_train = loader.load_all_data(train_path, labels_path,
-                                        nb_classes=nb_classes,
-                                        image_shape=(cols, rows));
+X, Y, filenames = loader.load_all_data(train_path, labels_path,
+                            nb_classes=nb_classes,
+                            image_shape=(cols, rows));
+
+X_train = X[:230]
+Y_train = Y[:230]
+X_valid = X[230:]
+Y_valid = Y[230:]
+
+
 print(strftime("%a, %d %b %Y %H:%M:%S +0000", localtime()))
 print("X_train shape: ", X_train.shape)
 print("Y_train shape: ", Y_train.shape)