GAN for image generation in Computer Vision - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to define the noise input shape for the GAN generator.

Computer Vision

noise_shape = ([1],)

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

A100

B28

C256

D64

Attempts:

3 left

💡 Hint

Common Mistakes

Using image dimensions like 28 or 64 instead of noise vector size.

Confusing noise shape with output image shape.

The noise input to the GAN generator is usually a vector of length 100, representing random latent space input.

2fill in blank

medium

Complete the code to compile the GAN generator model with the Adam optimizer.

Computer Vision

generator.compile(optimizer=[1], loss='binary_crossentropy')

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

A'adagrad'

B'adam'

C'rmsprop'

D'sgd'

Attempts:

3 left

💡 Hint

Common Mistakes

Using SGD which may be slower to converge.

Using RMSprop or Adagrad which are less common for GANs.

The Adam optimizer is commonly used for GAN training because it adapts learning rates and converges well.

3fill in blank

hard

Fix the error in the discriminator training step by completing the code to train on real images.

Computer Vision

d_loss_real = discriminator.train_on_batch(real_images, [1])

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

Anp.full((batch_size, 1), 0.5)

Bnp.random.rand(batch_size, 1)

Cnp.zeros((batch_size, 1))

Dnp.ones((batch_size, 1))

Attempts:

3 left

💡 Hint

Common Mistakes

Using zeros which label fake images.

Using random or 0.5 labels which confuse training.

Real images are labeled with ones (1) to indicate they are real during discriminator training.

4fill in blank

hard

Fill both blanks to generate fake images and label them for discriminator training.

Computer Vision

noise = np.random.normal(0, 1, (batch_size, [1]))
fake_labels = np.[2]((batch_size, 1))

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

A100

Bones

Czeros

Dfull

Attempts:

3 left

💡 Hint

Common Mistakes

Using ones for fake labels instead of zeros.

Using wrong noise vector size.

Noise vector length is 100. Fake images are labeled with zeros (0) for discriminator training.

5fill in blank

hard

Fill all three blanks to create a dictionary of losses for GAN training metrics.

Computer Vision

losses = {'d_loss_real': d_loss_real, 'd_loss_fake': d_loss_fake, '[1]': g_loss}
print(f"Discriminator real loss: {losses['d_loss_real']}")
print(f"Discriminator fake loss: {losses['d_loss_fake']}")
print(f"[2] loss: {losses['[3]']}")

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

Ag_loss

BGenerator

DDiscriminator

Attempts:

3 left

💡 Hint

Common Mistakes

Using 'Discriminator' label for generator loss.

Mismatching dictionary keys and print labels.

The generator loss key is 'g_loss'. The printed label is 'Generator'. The dictionary key used in print is 'g_loss'.