My first image classification model has an error

General Discussion
,
#1

I was following steps from a Nich Renotte youtube where he was classifying happy/sad face pictures. I tried modifying the model to recognize pictures from 10 classes of litter rather than 2. What would cause my training results to be so poor? I only have 175 training images, I know its low but thought it would kinda work?

CNN_loss_summary
CNN_loss_summary1087×740 168 KB

3. CREATE A DATASET

Load the image Data into a Tensorflow Pipeline tf.data.Dataset?? to display help.

data = tf.keras.utils.image_dataset_from_directory(‘Train’)

data_iterator = data.as_numpy_iterator()
batch = data_iterator.next()

4. SCALE THE DATA

Preprocess the data convert to grayscale x are the images, y are the labels

data = data.map(lambda x,y: (x/255, y))
data.as_numpy_iterator().next()

5. SPLIT THE DATA INTO TRAINING AND TESTING

len(data)

train_size = int(len(data).7)
val_size = int(len(data).2)
test_size = int(len(data)*.1)+1
train_size+val_size+test_size

train = data.take(train_size)
val = data.skip(train_size).take(val_size)
test = data.skip(train_size+val_size).take(test_size)

6. BUILD THE DEEPLEARNING NETWORK

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Dense, Flatten

model = Sequential()
model.add(Conv2D(16, (3,3), 1, activation = ‘relu’, input_shape=(256,256,3)))
model.add(MaxPooling2D())
model.add(Conv2D(32, (3,3), 1, activation = ‘relu’))
model.add(MaxPooling2D())
model.add(Conv2D(16, (3,3), 1, activation = ‘relu’))
model.add(MaxPooling2D())
model.add(Flatten())
model.add(Dense(256, activation=‘relu’))
model.add(Dense(1, activation = ‘sigmoid’))

model.compile(‘adam’, loss=tf.losses.BinaryCrossentropy(), metrics=[‘accuracy’])
model.summary()

7. TRAIN THE MODEL

logdir = ‘logs’
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=logdir)
hist = model.fit(train, epochs=10, validation_data=val, callbacks=[tensorboard_callback])

Thanks for any advice.

#3

Hi @Frank_Ogiamien, welcome to the Forum.

Can you try to change the final output layer of your network

From:
model.add(Dense(1, activation = ‘sigmoid’))
To:
model.add(Dense(2, activation = ‘softmax’))

Since you’ll need 2 classes (happy/sad), your last output layer should have a size of 2 as well.

#4

Verified - Khushboo Gupta

#5

Thanks Dennis, I gave that a try but it didn’t like it. I got:
ValueError: logits and labels must have the same shape, received ((None, 2) vs (None, 1)).
I was hoping it could do 10 labels for this project based on the images I’m using.
I’ll try and find some tutorials that train on many classes. Thanks

#6

If you have several classes, you should either apply one-hot encoding to your labels and use tf.keras.losses.CategoricalCrossentropy or leave the labels as they are and use tf.keras.losses.SparseCategoricalCrossentropy.

1 Like
#8

Thanks Ekaterina, I will research these methods and give it a try.