Why generative models create visual content in Computer Vision - Test Your Understanding

Practice - 5 Tasks

1fill in blank

easy

Complete the code to import the library used for creating generative models.

Computer Vision

import [1]

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

Atensorflow

Bnumpy

Cpandas

Dmatplotlib

Attempts:

3 left

💡 Hint

Common Mistakes

Choosing numpy which is for numerical operations but not specifically for generative models.

Choosing matplotlib which is for plotting, not model building.

TensorFlow is a popular library used to build generative models for visual content.

2fill in blank

medium

Complete the code to define the input shape for a generative model that creates 28x28 grayscale images.

Computer Vision

input_shape = ([1], [2], 1)

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

A28

B3

C64

D1

Attempts:

3 left

💡 Hint

Common Mistakes

Using 3 channels which is for color images.

Using 64 which is a common size but not for this example.

The input shape for 28x28 grayscale images is (28, 28, 1).

3fill in blank

hard

Fix the error in the code to generate random noise input for the generator.

Computer Vision

noise = tf.random.normal(shape=([1], 100))

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

A100

B64

C32

D1

Attempts:

3 left

💡 Hint

Common Mistakes

Using 1 which is too small for batch processing.

Using 100 which is the noise vector size, not batch size.

The batch size for noise input is often 64 for training generative models.

4fill in blank

hard

Fill both blanks to complete the generator model's output layer for 28x28 grayscale images.

Computer Vision

model.add(tf.keras.layers.Dense(28 * 28 * [1], activation='[2]'))

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

A1

Brelu

Csigmoid

D3

Attempts:

3 left

💡 Hint

Common Mistakes

Using 3 channels which is for color images.

Using relu activation which is not suitable for output pixel values.

The output layer should produce 28*28*1 values for grayscale images with sigmoid activation to get pixel values between 0 and 1.

5fill in blank

hard

Fill all three blanks to complete the training step that compiles the generative model with optimizer and loss.

Computer Vision

model.compile(optimizer='[1]', loss='[2]', metrics=['[3]'])

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

Aadam

Bmse

Caccuracy

Dsgd

Attempts:

3 left

💡 Hint

Common Mistakes

Using SGD optimizer which is less common for generative models.

Using categorical crossentropy loss which is for classification.

Adam optimizer is commonly used for generative models, mean squared error (mse) is a typical loss for image generation, and accuracy can be used as a metric.