diff --git a/modules/deeplearning/icing_fcn.py b/modules/deeplearning/icing_fcn.py
index 96fd4fd93aeaa20083f7e9bafe1a331a7d957380..c99fe687fded9aaba12373eab12e6700e8431343 100644
--- a/modules/deeplearning/icing_fcn.py
+++ b/modules/deeplearning/icing_fcn.py
@@ -575,85 +575,60 @@ class IcingIntensityFCN:
else:
input_2d = self.inputs[0]
- conv = tf.keras.layers.Conv2D(num_filters, 5, strides=[1, 1], padding=padding, activation=activation)(input_2d)
+ conv = tf.keras.layers.Conv2D(num_filters, kernel_size=5, strides=1, padding=padding, activation=activation)(input_2d)
conv = tf.keras.layers.MaxPool2D(padding=padding)(conv)
conv = tf.keras.layers.BatchNormalization()(conv)
print(conv.shape)
num_filters *= 2
- conv = tf.keras.layers.Conv2D(num_filters, 3, strides=[1, 1], padding=padding, activation=activation)(conv)
+ conv = tf.keras.layers.Conv2D(num_filters, kernel_size=3, strides=1, padding=padding, activation=activation)(conv)
conv = tf.keras.layers.MaxPool2D(padding=padding)(conv)
conv = tf.keras.layers.BatchNormalization()(conv)
print(conv.shape)
num_filters *= 2
- conv = tf.keras.layers.Conv2D(num_filters, 3, strides=[1, 1], padding=padding, activation=activation)(conv)
+ conv = tf.keras.layers.Conv2D(num_filters, kernel_size=3, strides=1, padding=padding, activation=activation)(conv)
conv = tf.keras.layers.MaxPool2D(padding=padding)(conv)
conv = tf.keras.layers.BatchNormalization()(conv)
print(conv.shape)
num_filters *= 2
- conv = tf.keras.layers.Conv2D(num_filters, 3, strides=[1, 1], padding=padding, activation=activation)(conv)
+ conv = tf.keras.layers.Conv2D(num_filters, kernel_size=3, strides=1, padding=padding, activation=activation)(conv)
conv = tf.keras.layers.MaxPool2D(padding=padding)(conv)
conv = tf.keras.layers.BatchNormalization()(conv)
print(conv.shape)
- # num_filters *= 2
- # conv = tf.keras.layers.Conv2D(num_filters, 3, strides=[1, 1], padding=padding, activation=activation)(conv)
- # conv = tf.keras.layers.MaxPool2D(padding=padding)(conv)
- # conv = tf.keras.layers.BatchNormalization()(conv)
- # print(conv.shape)
+ return conv
- flat = tf.keras.layers.Flatten()(conv)
-
- return flat
-
- def build_dnn(self, input_layer=None):
+ def build_fcl(self, input_layer):
print('build fully connected layer')
+ num_filters = input_layer.shape[3]
+
drop_rate = 0.5
# activation = tf.nn.relu
# activation = tf.nn.elu
activation = tf.nn.leaky_relu
momentum = 0.99
-
- if input_layer is not None:
- flat = input_layer
- n_hidden = input_layer.shape[1]
- else:
- flat = self.X_img
- n_hidden = self.X_img.shape[1]
-
- fac = 2
-
- fc = build_residual_block(flat, drop_rate, fac*n_hidden, activation, 'Residual_Block_1', doDropout=True, doBatchNorm=True)
-
- fc = build_residual_block(fc, drop_rate, fac*n_hidden, activation, 'Residual_Block_2', doDropout=True, doBatchNorm=True)
-
- fc = build_residual_block(fc, drop_rate, fac*n_hidden, activation, 'Residual_Block_3', doDropout=True, doBatchNorm=True)
-
- fc = build_residual_block(fc, drop_rate, fac*n_hidden, activation, 'Residual_Block_4', doDropout=True, doBatchNorm=True)
+ # padding = "VALID"
+ padding = "SAME"
- fc = build_residual_block(fc, drop_rate, fac*n_hidden, activation, 'Residual_Block_5', doDropout=True, doBatchNorm=True)
+ conv = tf.keras.layers.Conv2D(num_filters, kernel_size=1, strides=1, padding=padding, activation=activation)(input_layer)
- # fc = build_residual_block(fc, drop_rate, fac*n_hidden, activation, 'Residual_Block_6', doDropout=True, doBatchNorm=True)
- #
- # fc = build_residual_block(fc, drop_rate, fac*n_hidden, activation, 'Residual_Block_7', doDropout=True, doBatchNorm=True)
- #
- # fc = build_residual_block(fc, drop_rate, fac*n_hidden, activation, 'Residual_Block_8', doDropout=True, doBatchNorm=True)
+ conv = tf.keras.layers.Conv2D(num_filters, kernel_size=1, strides=1, padding=padding, activation=activation)(conv)
- fc = tf.keras.layers.Dense(n_hidden, activation=activation)(fc)
- fc = tf.keras.layers.BatchNormalization()(fc)
+ conv = tf.keras.layers.Conv2D(num_filters, kernel_size=1, strides=1, padding=padding, activation=activation)(conv)
+ print(conv.shape)
if NumClasses == 2:
activation = tf.nn.sigmoid # For binary
else:
activation = tf.nn.softmax # For multi-class
-
# Called logits, but these are actually probabilities, see activation
- logits = tf.keras.layers.Dense(NumLogits, activation=activation)(fc)
+ logits = tf.keras.layers.Conv2D(1, kernel_size=1, strides=1, padding=padding, activation=activation)(conv)
+
print(logits.shape)
-
+
self.logits = logits
def build_training(self):
@@ -705,6 +680,7 @@ class IcingIntensityFCN:
labels = mini_batch[2]
with tf.GradientTape() as tape:
pred = self.model(inputs, training=True)
+ pred = tf.reshape(pred, (pred.shape[0], NumLogits))
loss = self.loss(labels, pred)
total_loss = loss
if len(self.model.losses) > 0:
@@ -725,6 +701,7 @@ class IcingIntensityFCN:
inputs = [mini_batch[0], mini_batch[1]]
labels = mini_batch[2]
pred = self.model(inputs, training=False)
+ pred = tf.reshape(pred, (pred.shape[0], NumLogits))
t_loss = self.loss(labels, pred)
self.test_loss(t_loss)
@@ -924,14 +901,14 @@ class IcingIntensityFCN:
f.close()
def build_model(self):
- flat = self.build_cnn()
+ cnn = self.build_cnn()
# flat_1d = self.build_1d_cnn()
# flat = tf.keras.layers.concatenate([flat, flat_1d, flat_anc])
# flat = tf.keras.layers.concatenate([flat, flat_1d])
# self.build_dnn(flat)
- if self.USE_FLIGHT_ALTITUDE:
- flat = tf.keras.layers.concatenate([flat, self.inputs[1]])
- self.build_dnn(flat)
+ ## if self.USE_FLIGHT_ALTITUDE:
+ ## flat = tf.keras.layers.concatenate([flat, self.inputs[1]])
+ self.build_fcl(cnn)
self.model = tf.keras.Model(self.inputs, self.logits)
def restore(self, ckpt_dir):