I want to use the pretrained VGG19 (with imagenet weights) to build a two class classifier using a dataset of about 2.5k images that i've curated and split into 2 classes. It seems that not only is training taking a very long time, but accuracy seems to not increase in the slightest.
Here's my implementation:
def transferVGG19(train_dataset, val_dataset):
# conv_model = VGG19(weights='imagenet', include_top=False, input_shape=(224, 224, 3))
conv_model = VGG19(
include_top=True,
weights="imagenet",
input_tensor=None,
input_shape=(224, 224, 3),
pooling=None,
classes=1000,
classifier_activation="softmax",
)
for layer in conv_model.layers:
layer.trainable = False
input = layers.Input(shape=(224, 224, 3))
scale_layer = layers.Rescaling(scale=1 / 127.5, offset=-1)
x = scale_layer(input)
x = conv_model(x, training=False)
x = layers.Dense(256, activation='relu')(x)
x = layers.Dropout(0.5)(x)
x = layers.Dense(64, activation='relu')(x)
predictions = layers.Dense(1, activation='softmax')(x)
full_model = models.Model(inputs=input, outputs=predictions)
full_model.summary()
full_model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['acc'])
history = full_model.fit(
train_dataset,
epochs=10,
validation_data=val_dataset,
workers=10,
)
Model performance seems to be awful...
I imagine this behaviour comes from my rudimentary understanding of how layers work and how to best the new model's architecture. As VGG19 is trained on 1000 classes, i saw it best fit to add to the output a couple of dense layers to reduce the size of the feature maps, as well as a dropout layer in between to randomly discard neurons and help ease the risk of overfitting. At first i suspected i might have dropped too many neurons, but i was expecting my network to learn slower rather than not at all.
Is there something obviously wrong in my implementation that would cause such poor performance? Any explanation is welcomed. Just to mention, i would rule out the dataset as an issue because i've implemented transfer learning on Xception and have managed to get 98% validation accuracy that was monotonously increasing over 20 epochs. That implementation used different layers (i can provide it if necessary) because i was experimenting with different network layouts.
TLDR; Change include_top= True to False
Explaination-
Model graphs are represented in inverted manner i.e last layers are shown at the top and initial layers are shown at bottom.
When include_top=False, the top dense layers which are used for classification and not representation of data are removed from the pretrained VGG model. Only till the last conv2D layers are preserved.
During transfer-learning, you need to keep the learned representation layers intact and only learn the classification part for your data. Hence you are adding your stack of classification layers i.e.
x = layers.Dense(256, activation='relu')(x)
x = layers.Dropout(0.5)(x)
x = layers.Dense(64, activation='relu')(x)
predictions = layers.Dense(1, activation='softmax')(x)
If you keep the top classification layers of VGG, it will give 1000 probabilities for 1000 classes due to softmax activation at its top layer in model graph.This activation is not relu. We dont need softmax in intermediate layer as softmax "squishes" the unscaled inputs so that sum(input) = 1. Effectively it produces a smooth software defined approximation of argmax. Hence your accuracy is suffering.
Related
I'm trying to train VGG16 models using both transfer learning and training from scratch. I have a dataset with 7k images per category, and 4 different categories. I managed to come up with the transfer learning code no problem, however, the same program but for training from scratch does not seem to be working.
creating the model for transfer learning:
base_model = apps.VGG16(
include_top=False, # This is if we want the final FC layers
weights="imagenet",
input_shape=input_shape,
classifier_activation="softmax",
pooling = pooling,
)
# Freeze the base model
for layer in base_model.layers:
layer.trainable = False
# convert output of base model to a 1D vector
x = Flatten()(base_model.output)
# We create fc_count fully connected layers, relu for all but the last
x = Dense(units=4096, activation='relu')(x) # relu avoids vanishing gradient problem
x = Dense(units=4096, activation='relu')(x) # relu avoids vanishing gradient problem
# The final layer is a softmax layer
prediction = Dense(4, activation='softmax')(x)
model = Model(inputs=base_model.input, outputs=prediction)
model.compile(loss='categorical_crossentropy',
optimizer=optimizers.Adam(learning_rate=0.001),
metrics=['accuracy'])
Meanwhile, for training from scratch:
model = apps.VGG16(
include_top=True, # This is if we want the final FC layers
weights=None,
input_shape=input_shape,
classifier_activation="softmax",
pooling = pooling,
classes = 4 # set the number of outputs to required count
)
model.compile(loss='categorical_crossentropy',
optimizer=optimizers.Adam(learning_rate=0.1), # I've experimented w values as low as 0.001
metrics=['accuracy'])
model.summary()
and the training is done via
history = model.fit(train_images,
validation_data=val_images,
epochs=epochs,
verbose=1, callbacks=callbacks)
Transfer learning takes around 10 epochs to converge, whereas I've gone up to 20 epochs when training from scratch, converging to an accuracy and val_accuracy of exactly 0.2637. I have a ReduceLROnPlateau that does make a difference when transfer learning.
I'm training on a NVIDIA GeForce RTX 3060 Laptop GPU.
EDIT: I should mention that I am getting loss of nan when training from scratch
Problem got resolved by switching to the SGD optimizer
I am new to machine learning and I am currently trying to create a siamese network that can predict the similarity of brand logos.
I have a dataset with ~210.000 brand logos.
The CNN for the siamese network looks like the following:
def build_cnn(inputShape, embeddingDim=48):
# specify the inputs for the feature extractor network
inputs = Input(shape=inputShape)
# define the first set of CONV => RELU => POOL => DROPOUT layers
x = Conv2D(64, (2, 2), padding="same", activation="relu")(inputs)
x = MaxPooling2D(pool_size=(5, 5))(x)
x = Dropout(0.3)(x)
# second set of CONV => RELU => POOL => DROPOUT layers
x = Conv2D(64, (2, 2), padding="same", activation="relu")(x)
x = MaxPooling2D(pool_size=2)(x)
x = Dropout(0.3)(x)
pooledOutput = GlobalAveragePooling2D()(x)
outputs = Dense(embeddingDim)(pooledOutput)
# build the model
model = Model(inputs, outputs)
model.summary()
plot_model(model, to_file=os.path.sep.join([config.BASE_OUTPUT,'model_cnn.png']))
# return the model to the calling function
return model
The siamese network looks like this (the model here is the cnn described above):
imgA = Input(shape=config.IMG_SHAPE)
imgB = Input(shape=config.IMG_SHAPE)
featureExtractor = siamese_network.build_cnn(config.IMG_SHAPE)
featsA = featureExtractor(imgA)
featsB = featureExtractor(imgB)
distance = Lambda(euclidean_distance)([featsA, featsB])
outputs = Dense(1, activation="sigmoid")(distance)
model = Model(inputs=[imgA, imgB], outputs=outputs)
My first test was with 800 positive and 800 negative pairs and the accuracy and loss looks like this:
My thoughts to this were that there is some overfitting happening and my approach was to create more training data (2000 positive and negative pairs) and train the model again, but unfortunately the model was not improving at all after, even after 20+ epochs.
For both cases I used the following to train my network:
model.compile(loss="binary_crossentropy", optimizer="adam", metrics=["accuracy"])
print("[INFO] training model...")
history = model.fit(
[pairTrain[:, 0], pairTrain[:, 1]], labelTrain[:],
validation_data=([pairTest[:, 0], pairTest[:, 1]], labelTest[:]),
batch_size=10,
shuffle=True,
epochs=50)
I cannot figure out what is happening here, so I am really gratefull for every help.
My question here is why is the siamese network learning (or at least it looks like it is learning) with less training data, but as soon as I add more the accuracy is constant and not improving at all?
EDIT
According to Albertos comment I tried it with selu (still not working):
EDIT2
With LeakyReLU it looks like this:
My latest training result with 10k Pairs looks like this:
I've seen this happening other times, idk if it's a case but this Github issue has literally the same loss
As in that case, I think that is more a problem of initialization, so at this point I think you should use He-initialziation, or non-saturating activation function (for example, try tf.keras.layers.LeakyReLU or tf.keras.activations.selu)
I wanted to classify images which consist five classes. I wanted to use CNN. But when I try with several models, the training accuracy will not increase than 20%. Please some one help me to overcome this. Mostly model will trained within 3 epoches and when epoches increase there is no improvement in accuracy. Can anyone suggest me a solution or model or can specify what could be the problem?
Below is one of the model i have used
#defining training and test sets
x_train,x_val,y_train,y_val=train_test_split(x,y,test_size=0.2, random_state=42)
print('Training data and target sizes: \n{}, {}'.format(x_train.shape,y_train.shape))
print('Test data and target sizes: \n{}, {}'.format(x_val.shape,y_val.shape))
Training data and target sizes:
(2398, 224, 224, 3), (2398,)
Test data and target sizes:
(600, 224, 224, 3), (600,)
img_rows, img_cols, img_channel = 224, 224, 3
base_model = applications.inception_v3.InceptionV3(include_top=False, weights='imagenet',pooling='avg', input_shape=(img_rows, img_cols, img_channel))
print(base_model.summary())
#Adding custom Layers
add_model = Sequential()
add_model.add(Dense(1024, activation='relu',input_shape=base_model.output_shape[1:]))
add_model.add(Dropout(0.60))
add_model.add(Dense(1, activation='sigmoid'))
print(add_model.summary())
# creating the final model
model = Model(inputs=base_model.input, outputs=add_model(base_model.output))
# compile the model
opt = optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
reduce_lr = ReduceLROnPlateau(monitor='val_acc',
patience=5,
verbose=1,
factor=0.1,
cooldown=10,
min_lr=0.00001)
model.compile(
loss='categorical_crossentropy',
metrics=['acc'],
optimizer='adam'
)
print(model.summary())
n_fold = 5
kf = model_selection.KFold(n_splits = n_fold, shuffle = True)
eval_fun = metrics.roc_auc_score
model.fit(x_train,y_train,epochs=50,batch_size=50,validation_data=(x_val,y_val))
is it okay could you share the part of the code where you're fitting the model. It's not available in the post.
And since the output is not reproducible due to lack of data, I suggest you go through this link https://www.kaggle.com/kenconstable/alzheimer-s-multi-class-classification
It's really well explained and it has given the best practices of multi-class-classification based on transfer learning as well as from scratch. In case you don't find this helpful, It would be helpful to share the training script including the model.fit() code.
Okay, so here's the issue,
In your code, you may be creating a base model with inception V3, however, you are not really adding that base model to your add_model variable.
Your add_model variable is essentially a dense network and not a CNN. Also, another thing, although it's not a big deal is that you're creating your own optimiser opt and not using it in model.compile
Can you please try this code out and let me know if it works:
# function to build the model
def build_transfer_model(conv_base,dropout,dense_node,learn_rate,metric):
"""
Build and compile a transfer learning model
Input: a base model, dropout rate, the number of filters in the dense node,
the learning rate and performance metrics
Output: A compiled CNN model
"""
# clear previous run
backend.clear_session()
# build the model
model = Sequential()
model.add(conv_base)
model.add(Dropout(dropout))
model.add(BatchNormalization())
model.add(Flatten())
model.add(Dense(dense_node,activation='relu'))
model.add(Dense(1,activation='sigmoid'))
# complile the model
model.compile(
optimizer = tensorflow.keras.optimizers.Adam(lr=learn_rate),
loss = 'categorical_crossentropy',
metrics = metric )
model.summary()
return model
img_rows, img_cols, img_channel = 224, 224, 3
base_model = applications.inception_v3.InceptionV3(include_top=False, weights='imagenet',pooling='avg', input_shape=(img_rows, img_cols, img_channel))
model = build_transfer_model(conv_base=base_model,dropout=0.6,dense_node =1024,learn_rate=0.001,metric=['acc'])
print(model.summary())
model.fit(x_train,y_train,epochs=50,batch_size=50,validation_data=(x_val,y_val))
If you pay attention in the function, the first thing we are adding to the instance of Sequential() is the base layer (InceptionV3 in your case). But you were adding a dense layer directly. Although it may get the weights from the output layer of the base inception V3, it will be a dense network, not a CNN. So please check this out.
I may have changed the variable names, although I have tried not to do the same. And, please change the order of the layers in the build_transfer_model function according to your requirement.
In case it doesn't work, let me know.
Thanks.
You have to use model.fit() to actually train the model after compiling. Right now, it has randomly initialized weights, and is therefore making random predictions. Since you have five classes, the accuracy is approximately 1/5 = 20%. Training your model may take time depending on model size and amount of data you have.
I am trying to use a CNN architecture to classify text sentences. The architecture of the network is as follows:
text_input = Input(shape=X_train_vec.shape[1:], name = "Text_input")
conv2 = Conv1D(filters=128, kernel_size=5, activation='relu')(text_input)
drop21 = Dropout(0.5)(conv2)
pool1 = MaxPooling1D(pool_size=2)(drop21)
conv22 = Conv1D(filters=64, kernel_size=5, activation='relu')(pool1)
drop22 = Dropout(0.5)(conv22)
pool2 = MaxPooling1D(pool_size=2)(drop22)
dense = Dense(16, activation='relu')(pool2)
flat = Flatten()(dense)
dense = Dense(128, activation='relu')(flat)
out = Dense(32, activation='relu')(dense)
outputs = Dense(y_train.shape[1], activation='softmax')(out)
model = Model(inputs=text_input, outputs=outputs)
# compile
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
I have some callbacks as early_stopping and reduceLR to stop the training and to reduce the learning rate when the validation loss is not improving (reducing).
early_stopping = EarlyStopping(monitor='val_loss',
patience=5)
model_checkpoint = ModelCheckpoint(filepath=checkpoint_filepath,
save_weights_only=False,
monitor='val_loss',
mode="auto",
save_best_only=True)
learning_rate_decay = ReduceLROnPlateau(monitor='val_loss',
factor=0.1,
patience=2,
verbose=1,
mode='auto',
min_delta=0.0001,
cooldown=0,
min_lr=0)
Once the model is trained the history of the training goes as follows:
We can observe here that the validation loss is not improving from epoch 5 on and that the training loss is being overfitted with each step.
I will like to know if I'm doing something wrong in the architecture of the CNN? Aren't enough the dropout layers to avoid the overfitting? Which are other ways to reduce overfitting?
Any suggestion?
Thanks in advance.
Edit:
I have tried also with regularization an the result where even worse:
kernel_regularizer=l2(0.01), bias_regularizer=l2(0.01)
Edit 2:
I have tried to apply BatchNormalization layers after each convolution and the result is the next one:
norm = BatchNormalization()(conv2)
Edit 3:
After applying the LSTM architecture:
text_input = Input(shape=X_train_vec.shape[1:], name = "Text_input")
conv2 = Conv1D(filters=128, kernel_size=5, activation='relu')(text_input)
drop21 = Dropout(0.5)(conv2)
conv22 = Conv1D(filters=64, kernel_size=5, activation='relu')(drop21)
drop22 = Dropout(0.5)(conv22)
lstm1 = Bidirectional(LSTM(128, return_sequences = True))(drop22)
lstm2 = Bidirectional(LSTM(64, return_sequences = True))(lstm1)
flat = Flatten()(lstm2)
dense = Dense(128, activation='relu')(flat)
out = Dense(32, activation='relu')(dense)
outputs = Dense(y_train.shape[1], activation='softmax')(out)
model = Model(inputs=text_input, outputs=outputs)
# compile
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
overfitting can caused by many factors, it happens when your model fits too well to the training set.
To handle it you can do some ways:
Add more data
Use data augmentation
Use architectures that generalize well
Add regularization (mostly dropout, L1/L2 regularization are also possible)
Reduce architecture complexity.
for more clearly you can read in https://towardsdatascience.com/deep-learning-3-more-on-cnns-handling-overfitting-2bd5d99abe5d
This is screaming Transfer Learning. google-unversal-sentence-encoder is perfect for this use case. Replace your model with
import tensorflow_hub as hub
import tensorflow_text
text_input = Input(shape=X_train_vec.shape[1:], name = "Text_input")
# this next layer might need some tweaking dimension wise, to correctly fit
# X_train in the model
text_input = tf.keras.layers.Lambda(lambda x: tf.squeeze(x))(text_input)
# conv2 = Conv1D(filters=128, kernel_size=5, activation='relu')(text_input)
# drop21 = Dropout(0.5)(conv2)
# pool1 = MaxPooling1D(pool_size=2)(drop21)
# conv22 = Conv1D(filters=64, kernel_size=5, activation='relu')(pool1)
# drop22 = Dropout(0.5)(conv22)
# pool2 = MaxPooling1D(pool_size=2)(drop22)
# 1) you might need `text_input = tf.expand_dims(text_input, axis=0)` here
# 2) If you're classifying English only, you can use the link to the normal `google-universal-sentence-encoder`, not the multilingual one
# 3) both the English and multilingual have a `-large` version. More accurate but slower to train and infer.
embedded = hub.KerasLayer('https://tfhub.dev/google/universal-sentence-encoder-multilingual/3')(text_input)
# this layer seems out of place,
# dense = Dense(16, activation='relu')(embedded)
# you don't need to flatten after a dense layer (in your case) or a backbone (in my case (google-universal-sentence-encoder))
# flat = Flatten()(dense)
dense = Dense(128, activation='relu')(flat)
out = Dense(32, activation='relu')(dense)
outputs = Dense(y_train.shape[1], activation='softmax')(out)
model = Model(inputs=text_input, outputs=outputs)
I think since you are doing a text Classification, adding 1 or 2 LSTM layers might help the network learn better, since it will be able to better associate with the context of the data. I suggest adding the following code before the flatten layer.
lstm1 = Bidirectional(LSTM(128, return_sequence = True))
lstm2 = Bidirectional(LSTM(64))
LSTM layers can help neural network learn association between certain words and might improve the accuracy of your network.
I also Suggest dropping the Max Pooling layers as max pooling especially in text classification can lead the network to drop some of the useful features.
Just keep the convolutional Layers and the dropout. Also remove the Dense layer before flatten and add the aforementioned LSTMs.
It is unclear how you feed the text into your model. I am assuming that you tokenize the text to represent it as a sequence of integers, but do you use any word embedding prior to feeding it into your model? If not, I suggest you to throw atrainable tensorflow Embedding layer at the start of your model. There is a clever technique called Embedding Lookup to speed up its training, but you can save it for later. Try adding this layer to your model. Then your Conv1D layer would have a much easier time working on a sequence of floats. Also, I suggest you throw BatchNormalization after each Conv1D, it should help to speed up convergence and training.
I am new to Deep learning and Keras. What does pretrained weights initialization weights='imagenet' mean when used to define a model in Keras?
ResNet50(weights='imagenet')
Thanks!
This code line creates a network architecture known by the name ResNet50 (you can find more information about it here). The weights='imagenet' makes Keras load the weights of this network, which has been trained on the imagenet data set. Without this information Keras would only be able to prepare the network architecture but would not be able to set any of the weights to "good" values, as it does not know the purpose of the model. This is determined by specifying the data set.
If you are using an other data set, then you are using the model as a pre-trained model. You can find more information about this technique here; but the general idea is: after a model has been trained on any complex (image) data set, it will have learned in its lowest layers (most of the time: convolutions) to detect very basic features, such as edges, corners, etc. This helps the model to learn to analyze your own data set much faster, as it does not have to learn to detect this basic features again.
Following #FlashTek answer we can also train this model on our dataset.
Look at the following code:
model = applications.ResNet50(weights = "imagenet", include_top=False,
input_shape = (img_width, img_height,3))
# Freeze the layers which you don't want to train. Here I am freezing the first 30 layers.
for layer in model.layers[0:30]:
layer.trainable = False
for layer in model.layers[30:]:
layer.trainable = True
#Adding custom Layers
x = Flatten()(model.output)
# x = Dense(1024, activation="relu")(x)
# x = Dropout(0.5)(x)
# x = Dense(1024, activation="relu")(x)
# x = Dropout(0.5)(x)
x = Dense(1024, activation="relu")(x)
predictions = Dense(2, activation="softmax")(x)
In the above code we can are specifying how many layer of resnet we have to train on our dataset by assigning layer.trainable either true to train it on your dataset or false for otherwise.
Apart of that we can also stick layer after the network as shown in Adding custom layers