Quiz9

An educational illustration depicting neural networks and computer vision, featuring icons or graphics of face detection, convolutional layers, and object detection in an engaging style.

Deep Learning and Computer Vision Quiz

Test your knowledge on computer vision and neural networks with our engaging quiz! This quiz covers a variety of topics including algorithms, neural network architectures, and classifiers.

12 thought-provoking questions
Multiple choice and checkbox formats
Perfect for students and professionals alike

12 Questions3 MinutesCreated by LearningWave329

Which are true regarding the Viola-Jones face detector?

It only detects real human faces

It extracts Haar-like features from the images

It's very robust for side faces

Can be adapted to detect other things

I want to create my onw Cascade Classifier of British car license plates. What will I need to do?

Train the classifier with thousands of images of British license plates

Train the classifier with thousands of images of British license plates, and thousands of something else

Train the classifier with hundreds of images of British license plates

Train the classifier with thousands of images of British license plates, and thousands of Portuguese license plates

Regarding the cascade approach used in the Viola-Jones face detector, which statements are true?

The cascading process doesn't use any Haar-like features during classification

Only images "patches" that will highly likely have a face on it will be allowed to go through the cascade

There are Haar-like features than can tell me with a great certainty that an image "patch" doesn't contain a face; those are found during training and, then, used first in the cascade during classification

Usually, the majority of "patches" in one image won't contain a face, therefore, rejecting those immediately, will prevent calculating more detailed Haar-like features on them, resulting in great performance

What does a non-covolutional Artificial Neural Network needs to adjust during the learning (i.e. training) phase?

The training dataset

The contributions (i.e. weights) of each neuron in the previous layer to each neuron in the new layer

The labels

The bias of each neuron

What does a Convolutional Neural Network needs to adjust during the learning (i.e. training) phase?

The contributions (i.e. weights) of each neuron in the previous layer to each neuron in the new layer, when talking about dense layers

The contributions (i.e. weights) of each neuron in the previous layer to each neuron in the new layer, when talking about convolutional layers

The labels

The values of the filters, when talking about the convolutional layers

N the context of Neural Networks what is the idea behind the technique known as Dropout?

It consists in randomly "shutting down" some neurons during testing, so that the model doesn't overfit to the training data (i.e. Very clever at recognizing the training data, but very stupid at recognizing unseen data), with the benefit that it also results in better performance during training

Even without looking at any of the labels of the dataset, why can I immediately reach the conclusion that the model is not expecting one-hot encoded labels? model = tf.keras.models.Sequential([ tf.keras.layers.Flatten(input_shape=(28, 28)), tf.keras.layers.Dense(128, activation='relu'), tf.keras.layers.Dropout(0.2), tf.keras.layers.Dense(10, activation='softmax') ]) model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

The loss function is the Sparse Categorical Cross Entropy

The loss function is the Categorical Cross Entropy

The activation function of the output layer is the Softmax

It uses the Adam Optimizer

How can I tell that we are dealing with a multi-classification problem, just by looking a the follwing model definition? model = tf.keras.models.Sequential([ tf.keras.layers.Flatten(input_shape=(28, 28)), tf.keras.layers.Dense(128, activation='relu'), tf.keras.layers.Dropout(0.2), tf.keras.layers.Dense(10, activation='softmax') ]) model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

It uses the Adam Optimizer

The loss function is the Sparse Categorical Cross Entropy

The activation function of the output layer is the Softmax

The activation function of the output layer is the Sigmoid

What is purpose of the definition of the first "layer" in the following model? model = tf.keras.models.Sequential([ tf.keras.layers.Flatten(input_shape=(28, 28)), tf.keras.layers.Dense(128, activation='relu'), tf.keras.layers.Dropout(0.2), tf.keras.layers.Dense(10, activation='softmax') ])

Convert a 28×28 color image into a vector of 2352 elements

Convert a 28×28 grayscale image into a 3D vector of 784 elements

Convert a 28×28 grayscale image into a vector of 784 elements

Tests performed on a given classifier achieved an AUC (Area Under the Curve) of 0.96. The result is very promising. True or false?

True

False

Which measure(s) tell(s) me how well does a classifier perform in rejecting a sample (e.g. reporting "not cancer")?

False Positive Rate

Fall-out

FPR

F-score

Which are some strengths of the YOLO object detector?

Besides classifying, it also returns the location of the object (i.e. Performs detection)

Works in real time

Doesn't need much data to train

Doesn't need training

Quiz9

Deep Learning and Computer Vision Quiz

More Quizzes