Back to course
Next page
Interactive Quiz
Test your knowledge!
1
What is the main task of face verification?
A
Automatically identify the person in a photo from a complete database
B
Verify if two photos belong to the same person
C
Generate a realistic image of a face from an example photo
D
Recognize all people present in a group image
2
What is the main objective of the contrastive loss function in training a face verification model?
A
Minimize the distance between different images and maximize the distance between similar images
B
Maximize the distance between all images, whether they are similar or not
C
Minimize the distance between images of the same category and ensure a minimum margin between images of different categories
D
Minimize the distance between different images only
3
Why is a margin (parameter \(m\)) used in the contrastive loss function?
A
To speed up training by reducing model complexity
B
To prevent the model from learning a trivial constant function and improve generalization
C
To increase the size of the images processed by the model
D
To ensure that the distance between similar images is always zero
4
In the Siamese Network model architecture presented, what is the role of the convolutional network?
A
Generate synthetic images from embeddings
B
Extract relevant features from images while being robust to geometric distortions
C
Apply normalization to input images
D
Directly calculate the pixel distance between two images
5
What is the main difference between face verification and face recognition?
A
Face verification identifies a person in a database, face recognition only compares two images
B
Face verification compares a photo to a given identity, face recognition automatically identifies who the person is from a larger database
C
Face recognition is only used for face detection, face verification for facial recognition
D
There is no difference, the terms are interchangeable
6
What is the principle of triplet loss used in face recognition?
A
Minimize the distance between one image and all other images in the database
B
Maximize the distance between two randomly selected images in the dataset
C
Minimize the distance between the anchor and the positive while maximizing the distance between the anchor and the negative, with a margin
D
Use three identical images to increase model robustness
7
In unsupervised contrastive training (SimCLR method), how are positive pairs formed?
A
Two random different images from the dataset
B
One image and another very similar image belonging to the same class (known label)
C
Two images derived from the same image but transformed by different augmentations
D
Two images generated by an adversarial generator
8
What is the main advantage of unsupervised contrastive training for images?
A
It requires a very large number of labeled images
B
It allows learning relevant representations without using labels
C
It is only applicable to face images
D
It guarantees perfect performance on all classification tasks
9
Among the following unsupervised methods, which one relies on adversarial training between a generator and a discriminator?
A
Autoencoders
B
Generative Adversarial Networks (GAN)
C
SimCLR method
D
Transformation prediction (RotNet)
10
What is the main reason why fine-tuning is useful after unsupervised contrastive training?
A
To train a model to recognize only artificially generated images
B
To adapt a pre-trained model on a general task to a specific task with labels
C
To reduce the size of the initial model
D
To transform a supervised model into an unsupervised model
Score: 0/10
Score: 0/10