Classification report for CNN

Question

train_datagen = ImageDataGenerator(rescale=1./255)  # Rescale pixel values
test_datagen = ImageDataGenerator(rescale=1./255)   # Rescale pixel values

train_generator = train_datagen.flow_from_directory(
    train_data_path,
    target_size=(150, 150),
    batch_size=32,
    class_mode='categorical'
  )

test_generator = test_datagen.flow_from_directory(
    test_data_path,
    target_size=(150, 150),
    batch_size=32,
    class_mode='categorical'
  )

# Set the seed for NumPy
np.random.seed(42)

# Set the seed for TensorFlow
tf.random.set_seed(42)

# Define the model architecture
L1 = 32
L2 = 64
L3 = 128
L4 = 128
L5 = 512

model1 = Sequential()
model1.add(Conv2D(L1, (3,3), activation='relu', input_shape = (150, 150, 3),
                 kernel_initializer = 'he_normal',kernel_regularizer = regularizers.l2(0.01)))
model1.add(MaxPooling2D(2, 2))

model1.add(Conv2D(L2, (3,3), activation='relu', padding='same'))
model1.add(MaxPooling2D(2,2))

model1.add(Conv2D(L3, (3,3), activation='relu', padding='same'))
model1.add(MaxPooling2D(2,2))

model1.add(Conv2D(L4, (3,3), activation='relu', padding='same'))
model1.add(MaxPooling2D(2,2))

model1.add(Flatten())
model1.add(Dense(L5, activation='relu'))
model1.add(Dropout(0.3))
model1.add(Dense(2, activation='sigmoid'))

# Compile the model with a regression-specific loss function
model1.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

# training the model for 10 epochs
history1 = model1.fit(
    train_generator,
    epochs=10,
    steps_per_epoch=len(train_generator),
    validation_data=test_generator,
    validation_steps=len(test_generator)
)

Hello,

I'm doing image classification for gender. In total, I have 3500 male images and 3500 female images for training dataset and 700 male images and 700 females images for testing dataset. When I run the algorithm above, this is the model performance that I got:

It seems like the model is performing quite good, but when I run the classification report below:

# Make predictions on the test dataset
y_pred = model1.predict(test_generator)
y_pred_classes = np.argmax(y_pred, axis=1)  # Convert probabilities to class labels

# Get true labels
y_true = test_generator.classes

# Define class labels
class_labels = {0: 'female', 1: 'male'} 

# Compute confusion matrix
conf_matrix = confusion_matrix(y_true, y_pred_classes)

# Compute classification report
class_report = classification_report(y_true, y_pred_classes)

print("\nClassification Report:")
print(class_report)

This is the result that I got from the report:

The accuracy from the classification report is only 0.5. May I know why the accuracy from the classification report seems like randomly predicting on the testing dataset instead of using the trained model? Where I did wrongly? Appreciate any help or advise, thank you very much.