Pre-trained models (VGG, ResNet, MobileNet) in TensorFlow - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to load the VGG16 pre-trained model without the top classification layer.

TensorFlow

from tensorflow.keras.applications import VGG16
model = VGG16(include_top=[1], weights='imagenet')

Drag options to blanks, or click blank then click option'

ATrue

B'imagenet'

CFalse

DNone

Attempts:

3 left

💡 Hint

Common Mistakes

Using include_top=True when you want to customize the output.

Passing weights=None instead of 'imagenet'.

Setting include_top=False loads the model without the final classification layers, useful for feature extraction.

2fill in blank

medium

Complete the code to preprocess input images for ResNet50 model.

TensorFlow

from tensorflow.keras.applications.resnet50 import preprocess_input
preprocessed_img = preprocess_input([1])

Drag options to blanks, or click blank then click option'

Araw_image

Bmodel

Cinput_shape

Dpredictions

Attempts:

3 left

💡 Hint

Common Mistakes

Passing the model object instead of image data.

Passing predictions instead of raw images.

The preprocess_input function expects raw image data to prepare it for the model.

3fill in blank

hard

Fix the error in the code to correctly load MobileNet with input shape 128x128x3.

TensorFlow

from tensorflow.keras.applications import MobileNet
model = MobileNet(input_shape=[1], weights='imagenet', include_top=False)

Drag options to blanks, or click blank then click option'

A(128, 128, 3)

B[128, 128, 3]

C128, 128, 3

D'128x128x3'

Attempts:

3 left

💡 Hint

Common Mistakes

Using list brackets instead of tuple parentheses.

Passing a string instead of a tuple.

The input_shape argument requires a tuple specifying the dimensions.

4fill in blank

hard

Fill both blanks to create a feature extractor from ResNet50 and freeze its layers.

TensorFlow

from tensorflow.keras.applications import ResNet50
model = ResNet50(include_top=[1], weights='imagenet')
model.[2] = False

Drag options to blanks, or click blank then click option'

AFalse

BTrue

Ctrainable

Dcompile

Attempts:

3 left

💡 Hint

Common Mistakes

Using include_top=True when freezing layers.

Trying to freeze layers by calling compile instead of setting trainable.

Setting include_top=False removes the classification layers. Setting model.trainable = False freezes the model weights.

5fill in blank

hard

Fill all three blanks to build a simple classifier on top of MobileNet features.

TensorFlow

from tensorflow.keras.applications import MobileNet
from tensorflow.keras import layers, models
base_model = MobileNet(include_top=[1], weights='imagenet', input_shape=(224,224,3))
base_model.trainable = [2]
model = models.Sequential([
    base_model,
    layers.GlobalAveragePooling2D(),
    layers.Dense([3], activation='softmax')
])

Drag options to blanks, or click blank then click option'

AFalse

BTrue

C10

DNone

Attempts:

3 left

💡 Hint

Common Mistakes

Setting include_top=True when adding custom layers.

Not freezing the base model causing slow training.

Using None instead of number of classes in Dense layer.

We exclude the top layer (include_top=False) to add our own classifier. We freeze the base model (trainable=False) to keep pre-trained weights fixed. The final Dense layer has 10 units for 10 classes.