模型修正

mix/mix_train.py

@@ -19,7 +19,7 @@ from keras.models import Model
 def read_data(path):
     lines = []
     with open(path) as f:
-        for x in range(80000): #380000
+        for x in range(300000): #380000
             lines.append(eval(f.readline().strip()))
 
     random.shuffle(lines)
@@ -74,18 +74,17 @@ def create_cnn(width, height, depth, filters=(4, 6), kernel_size=(5, 6), regress
     # define the model input
     inputs = Input(shape=inputShape)
 
-    # loop over the number of filters
-    for (i, f) in enumerate(filters):
-        # if this is the first CONV layer then set the input
-        # appropriately
-        if i == 0:
-            x = inputs
+    x = inputs
 
-        # CONV => RELU => BN => POOL
-        x = Conv2D(f, kernel_size, padding="same")(x)
-        x = Activation("relu")(x)
-        x = BatchNormalization(axis=chanDim)(x)
-        # x = MaxPooling2D(pool_size=(2, 2))(x)
+    # CONV => RELU => BN => POOL
+    x = Conv2D(32, kernel_size, strides=2, padding="same")(x)
+    x = Activation("relu")(x)
+    x = BatchNormalization(axis=chanDim)(x)
+    # x = MaxPooling2D(pool_size=(2, 2))(x)
+    # if width > 2:
+    #     x = Conv2D(32, (10, 6), padding="same")(x)
+    #     x = Activation("relu")(x)
+    #     x = BatchNormalization(axis=chanDim)(x)
 
     # flatten the volume, then FC => RELU => BN => DROPOUT
     x = Flatten()(x)
@@ -112,8 +111,8 @@ def create_cnn(width, height, depth, filters=(4, 6), kernel_size=(5, 6), regress
 
 # create the MLP and CNN models
 mlp = create_mlp(train_x_c.shape[1], regress=False)
-cnn_0 = create_cnn(18, 18, 1, kernel_size=(6, 6), regress=False, output=256)
-cnn_1 = create_cnn(18, 2, 1, kernel_size=(6,2), regress=False, output=36)
+cnn_0 = create_cnn(18, 18, 1, kernel_size=(10, 6), regress=False, output=256)
+cnn_1 = create_cnn(18, 2, 1, kernel_size=(10,2), regress=False, output=36)
 
 # create the input to our final set of layers as the *output* of both
 # the MLP and CNN
@@ -149,8 +148,8 @@ model.fit(
     [train_x_c, train_x_a, train_x_b], train_y,
     # validation_data=([testAttrX, testImagesX], testY),
     # epochs=int(3*train_x_a.shape[0]/1300),
-    epochs=120,
-    batch_size=4096, shuffle=True)
+    epochs=100,
+    batch_size=2048, shuffle=True)
 
 test_x_a = test_x[:,:18*18]
 test_x_a = test_x_a.reshape(test_x.shape[0], 18, 18, 1)