I'm trying to finetune the last layer of the VGG-16. Here is the part of the code where i make the new model:
def train2false(model):
for layer in model.layers:
layer.trainable = False
return model
def define_training_layers(model):
model.layers = model.layers[0:21]
model = train2false(model)
last_layer = model.get_layer('fc7')
out = Dense(n_classes, activation='softmax', name='fc8')(last_layer)
model = Model(input=model.input, output=out)
return model
def compile_model(epochs, lrate, model):
decay = lrate / epochs
sgd = SGD(lr=lrate, momentum=0, decay=0.0002, nesterov=True)
model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])
print (model.summary())
return model
def train_evaluate(model, X_train, y_train, X_test, y_test, epochs):
model.fit(X_train, y_train, validation_data=(X_test, y_test), nb_epoch=epochs, batch_size=32)
# Final evaluation of the model
scores = model.evaluate(X_test, y_test, verbose=0)
print("Accuracy: %.2f%%" % (scores[1]*100))
return model
X_train, X_test, labels_test, labels_train, n_classes = load_dataset()
image_input = Input(shape=(3, 224, 224))
vgg_model = VGGFace(input_tensor= image_input, include_top=True)
custom_vgg_model = define_training_layers(vgg_model)
custom_vgg_model = compile_model(epochs=50, lrate=0.001, model=custom_vgg_model)
custom_vgg_model = train_evaluate(custom_vgg_model, X_train=X_train, y_train=labels_train, X_test=X_test, y_test=labels_test, epochs=50)
I get the following error:
tensorflow.python.framework.errors_impl.InvalidArgumentError:
Dimension 1 in both shapes must be equal, but are 1000 and 2622 for
'Assign_30' (op: 'Assign') with input shapes: [4096,1000],
[4096,2622].
It works for me if i try to finetune all the fully connected part with include_top=False instead of just the softmax activation.
Is there something that i'm missing?
Solved!!! I've take the pre-trained weights from https://github.com/rcmalli/keras-vggface/releases/download/v1.0/rcmalli_vggface_th_weights_th_ordering.h5 which has 2622 number of outputs and i had 1000 outputs. So just change the number of outputs for the last layer in VGG.py
Related
I'm trying binary classification of text with bi-lstm model but getting this error: ValueError: Dimensions must be equal, but are 2 and 64 for '{{node binary_crossentropy/mul}} = Mul[T=DT_FLOAT](binary_crossentropy/Cast, binary_crossentropy/Log)' with input shapes: [?,2], [?,64].
I am a beginner please provide some valuable solutions.
text=df['text']
label=df['label']
X = pad_sequences(X, maxlen=max_len,padding=pad_type,truncating=trunc_type)
Y = pd.get_dummies(label).values
X_train, X_test, Y_train, Y_test = train_test_split(X,Y, test_size = 0.20)
print(X_train.shape,Y_train.shape)
print(X_test.shape,Y_test.shape)
#model creation
model=tf.keras.Sequential([
# add an embedding layer
tf.keras.layers.Embedding(word_count, 16, input_length=max_len),
tf.keras.layers.Dropout(0.2),
# add another bi-lstm layer
tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(2,return_sequences=True)),
# add a dense layer
tf.keras.layers.Dense(32, activation=tf.keras.activations.relu),
tf.keras.layers.Dense(32, activation=tf.keras.activations.relu),
tf.keras.layers.Dense(32, activation=tf.keras.activations.relu),
tf.keras.layers.Dense(32, activation=tf.keras.activations.softmax),
# add the prediction layer
tf.keras.layers.Dense(1, activation=tf.keras.activations.sigmoid),
])
model.compile(loss=tf.keras.losses.BinaryCrossentropy(), optimizer=tf.keras.optimizers.Adam(), metrics=['accuracy'])
model.summary()
history = model.fit(X_train, Y_train, validation_data=(X_test, Y_test), epochs = 10, batch_size=batch_size, callbacks = [callback_func], verbose=1)
The output dimension of the prediction layer of the binary classification should be 2:
# add the prediction layer
tf.keras.layers.Dense(2, activation=tf.keras.activations.sigmoid)
Flatten:
#model creation
model=tf.keras.Sequential([
# add an embedding layer
tf.keras.layers.Embedding(word_count, 16, input_length=max_len),
tf.keras.layers.Dropout(0.2),
# add another bi-lstm layer
tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(2,return_sequences=True)),
# add flatten
tf.keras.layers.Flatten(), #<========================
# add a dense layer
tf.keras.layers.Dense(32, activation=tf.keras.activations.relu),
tf.keras.layers.Dense(32, activation=tf.keras.activations.relu),
tf.keras.layers.Dense(32, activation=tf.keras.activations.relu),
tf.keras.layers.Dense(32, activation=tf.keras.activations.softmax),
# add the prediction layer
tf.keras.layers.Dense(2, activation=tf.keras.activations.sigmoid),
])
I have to following model:
#load VGG16 as convolutional base
conv_base = VGG16(weights='imagenet',
include_top=False,
input_shape=(150, 150, 3))
checkpoint_filepath = "/content/gdrive/MyDrive/Colab Notebooks/Deep_Learning/models/Benign_Melign_VGG_noAug_final.h5"
model_checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(
filepath=checkpoint_filepath,
save_weights_only=True,
monitor='val_loss',
mode='auto',
save_best_only=True)
#add custom fully-connected network on top of the already-trained base network
model = models.Sequential()
model.add(conv_base)
model.add(layers.Flatten())
model.add(layers.Dense(256, activation="relu"))
model.add(layers.Dense(1, activation="sigmoid"))
#freeze convolutional base
conv_base.trainable = False
model.compile(loss="binary_crossentropy",
optimizer=optimizers.Adam(lr=1e-3), # lr = 0.0001
metrics=METRICS)
#train fully-connected added part
history = model.fit(train_generat.flow(train_dataset_split,
train_labels_split,
batch_size=BATCH_SIZE,
shuffle=False),
steps_per_epoch=len(train_dataset_split) // BATCH_SIZE,
epochs=100,
validation_data=valid_generat.flow(valid_dataset_split,
valid_labels_split,
batch_size=BATCH_SIZE,
shuffle=False),
validation_steps=len(valid_labels_split) // BATCH_SIZE,
callbacks=[model_checkpoint_callback])
model.summary()
model.save('Benign_Melign_VGG_noAug_final.h5')
the error is:
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/base_layer.py in get_config(self)
719 raise NotImplementedError('Layer %s has arguments in `__init__` and '
720 'therefore must override `get_config`.' %
--> 721 self.__class__.__name__)
722 return config
723
NotImplementedError: Layer ModuleWrapper has arguments in `__init__` and therefore must override `get_config`.
there is a problem in the last line when I want to save the model. then I also need to load it later. I need to save and load the whole model, not just the weights. can you help e with this?
I have a simple MLP built in Keras. The shapes of my inputs are:
X_train.shape - (6, 5)
Y_train.shape - 6
Create the model
model = Sequential()
model.add(Dense(32, input_shape=(X_train.shape[0],), activation='relu'))
model.add(Dense(Y_train.shape[0], activation='softmax'))
# Compile and fit
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(X_train, Y_train, epochs=10, batch_size=1, verbose=1, validation_split=0.2)
# Get output vector from softmax
output = model.layers[-1].output
This gives me the error:
ValueError: Error when checking input: expected dense_1_input to have shape (6,) but got array with shape (5,).
I have two questions:
Why do I get the above error and how can I solve it?
Is output = model.layers[-1].output the way to return the softmax vector for a given input vector? I haven't ever done this in Keras.
in the input layer use input_shape=(X_train.shape[1],) while your last layer has to be a dimension equal to the number of classes to predict
the way to return the softmax vector is model.predict(X)
here a complete example
n_sample = 5
n_class = 2
X = np.random.uniform(0,1, (n_sample,6))
y = np.random.randint(0,n_class, n_sample)
model = Sequential()
model.add(Dense(32, input_shape=(X.shape[1],), activation='relu'))
model.add(Dense(n_class, activation='softmax'))
# Compile and fit
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(X, y, epochs=10, batch_size=1, verbose=1)
# Get output vector from softmax
model.predict(X)
custom-loss-function of keras got wrong output:
When I use a Bayesian layer (tensorflow_probability.layers.DenseFlipout), and use my custom loss function, I got a wrong output loss. But if I replace Bayesian layer by a traditional tf.keras.layers.Dense layer, the output is correct. Can anybody help me ?
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data as mnist_data
train, valid, test = mnist_data.read_data_sets('~/code/Python')
num_classes = 10
from tensorflow import keras
import tensorflow_probability as tfp
model = keras.Sequential()
#model.add(keras.layers.Dense(10, activation = 'softmax', input_shape=(784,)))
model.add(tfp.layers.DenseFlipout(10, activation = 'softmax', input_shape=(784,)))
sgd = keras.optimizers.SGD(lr=.1, momentum=0.9, nesterov=True)
def my_loss(y_true,y_pred):
return tf.reduce_mean((y_true-y_pred)**2)
model.compile(loss=my_loss, optimizer=sgd, metrics=['accuracy'])
x_train, y_train = train.images, train.labels
x_test, y_test = test.images, test.labels
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
model.fit(x_train, y_train,
batch_size=128,
epochs=10,
validation_data=(x_test, y_test),
shuffle=True)
I'm fine-tuning the Inception V3 model with Keras, in order to convert it with coremltools into a .mlmodel file.
However, when converting the model coremltools throws an error saying the following when the converter reaches the last layer of the model:
coremltools/models/neural_network.py", line 2501, in set_pre_processing_parameters
channels, height, width = array_shape
ValueError: need more than 1 value to unpack
I used the code from the Keras documentation on applications found here: https://keras.io/applications/#fine-tune-inceptionv3-on-a-new-set-of-classes
And added a piece of code loading my dataset from the VGG example found here: https://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html
My final script looks like this, using TesorFlow as backend:
LOAD THE DATA
from keras.preprocessing.image import ImageDataGenerator
img_width, img_height = 299, 299
train_data_dir = 'data/train'
validation_data_dir = 'data/validation'
nb_train_samples = 358
nb_validation_samples = 21
epochs = 1
batch_size = 15
train_datagen = ImageDataGenerator(
rescale=1. / 255,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True)
test_datagen = ImageDataGenerator(rescale=1. / 255)
train_generator = train_datagen.flow_from_directory(
train_data_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode='categorical')
validation_generator = test_datagen.flow_from_directory(
validation_data_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode='categorical')
TRAIN THE MODEL
base_model = InceptionV3(weights='imagenet', include_top=False)
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(1024, activation='relu')(x)
predictions = Dense(7, activation='softmax')(x)
model = Model(inputs=base_model.input, outputs=predictions)
for layer in base_model.layers:
layer.trainable = False
model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
model.fit_generator(
train_generator,
steps_per_epoch=nb_train_samples // batch_size,
epochs=epochs,
validation_data=validation_generator,
validation_steps=nb_validation_samples // batch_size)
for i, layer in enumerate(base_model.layers):
print(i, layer.name)
for layer in model.layers[:249]:
layer.trainable = False
for layer in model.layers[249:]:
layer.trainable = True
from keras.optimizers import SGD
model.compile(optimizer=SGD(lr=0.0001, momentum=0.9), loss='categorical_crossentropy', metrics=['accuracy'])
model.fit_generator(
train_generator,
steps_per_epoch=nb_train_samples // batch_size,
epochs=epochs,
validation_data=validation_generator,
validation_steps=nb_validation_samples // batch_size)
model.save('finetuned_inception.h5')
I'm writing here in response to #SwimBikeRun's request (as I need a bit more space)
I was converting YOLO to Keras and then Keras to CoreML. For conversion I was using this script https://github.com/qqwweee/keras-yolo3/blob/master/convert.py
In the conversion-process the model was eventually created like that:
input_layer = Input(shape=(None, None, 3))
...
model = Model(inputs=input_layer, outputs=[all_layers[i] for i in out_index])
And those "None"-inputs was what made CoreML conversion fail. For CoreML the input-size to your model must be known. So I changed it to this:
input_layer = Input(shape=(416, 416, 3)
Your input-size will probably vary.
For your original question:
Maybe check your base_model.input size for the same problem.