I am generating a model to read QR codes. I figured that I kept receiving a test accuracy of 0 and a training accuracy of 1.0 due to overfitting. Once I reduced the layers in my CNN, I got a training accuracy of 98% and training loss of 3, while a testing accuracy of 0% and a testing loss of 6.

Here is my code:

# generating QR codes (10 for now)  
base = 'hello'
qr_folder = 'QR Code Images'

os.makedirs(qr_folder, exist_ok = True)

for i in range (1,301): 
    data = f'{base} {i}'
    img = qrcode.make(data)
    img.save(os.path.join(qr_folder, f'QR Code Image {i}.png'))

# preprocess qr code images 
def load_images (image_folder):
    # initilaize images and labels
    images = []
    labels = []
    for file_name in os.listdir(image_folder):
        if file_name.endswith('.png'):
            img_path = os.path.join(image_folder, file_name)
            img = cv2.imread(img_path, cv2.IMREAD_GRAYSCALE)
            img = cv2.resize(img, (128,128))
            images.append(img)
            # file name will be the label
            labels.append(file_name.split('.')[0])
    return np.array(images), np.array(labels)

images, labels = load_images(qr_folder)
images = images/255.0 

images = np.expand_dims(images, axis=-1)

# convert labels to numerical format
label_encoder = LabelEncoder()
y_encoded = label_encoder.fit_transform(labels)

# split into training and testing - 20% testing, 80% training
X_train, X_test, y_train, y_test = train_test_split(images, y_encoded, test_size = 0.2, random_state = 42) # for now random state will be 42, after debugging - change to none 

# data augmentation
datagen = ImageDataGenerator(
    rotation_range = 20,
    width_shift_range = 0.2,
    height_shift_range = 0.2,
    shear_range = 0.2,
    zoom_range = 0.2,
    horizontal_flip = True,
    fill_mode = 'nearest'
)

datagen.fit(X_train)

# create a CNN w/ tensorflow - add more layers model is too simple 
model = tf.keras.Sequential([
    tf.keras.layers.InputLayer(shape = (128, 128, 1)),
    tf.keras.layers.Conv2D(filters = 32, kernel_size = 3, activation = 'relu', kernel_regularizer = tf.keras.regularizers.l2(0.001)),
    tf.keras.layers.MaxPooling2D(),
    tf.keras.layers.Conv2D(filters = 64, kernel_size = 3, activation = 'relu', kernel_regularizer = tf.keras.regularizers.l2(0.001)),
    tf.keras.layers.MaxPooling2D(), 
    tf.keras.layers.Conv2D(filters = 128, kernel_size = 3, activation = 'relu', kernel_regularizer = tf.keras.regularizers.l2(0.0001)),
    #tf.keras.layers.MaxPooling2D(),
    #tf.keras.layers.Conv2D(filters = 64, kernel_size = 6, activation = 'relu', kernel_regularizer=tf.keras.regularizers.l2(0.001)),
    #tf.keras.layers.MaxPooling2D(),
    #tf.keras.layers.Conv2D(filters = 128, kernel_size = 6, activation = 'relu'),
    tf.keras.layers.Flatten(),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Dense(256, activation = 'relu', kernel_regularizer = tf.keras.regularizers.l2(0.001)),
    tf.keras.layers.Dropout(0.5), # add in regularization constant 
    tf.keras.layers.Dense(300, activation = 'softmax')
    #tf.keras.layers.Dense(20, activation = 'sigmoid')
])

model.summary()

# compile model
model.compile(optimizer = tf.keras.optimizers.AdamW(learning_rate = 0.0001),
              loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits = False),
              metrics = ['accuracy'])
              #metrics=[tf.keras.metrics.SparseCategoricalAccuracy()])

# early stopping
#early_stopping = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience = 5, restore_best_weights = True)


# train model changed to 10
history = model.fit(X_train, y_train, epochs = 20, validation_data = (X_test, y_test)) # callbacks = [early_stopping]) 


# something's wrong - plot the training accuracy and loss 
plt.plot(history.history['accuracy'], label = 'accuracy')
plt.plot(history.history['val_accuracy'], label = 'val_accuracy')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.ylim([0, 1])
plt.legend(loc='lower right')

plt.show()

# evaluate model
test_loss, test_acc = model.evaluate(X_test, y_test)
train_loss, train_acc = model.evaluate(X_train, y_train)
print(f'Test accuracy: {test_acc}')
print(f'Testing loss: {test_loss}')
print(f'Training loss: {train_loss}')
print(f'Training accuracy: {train_acc}')