Machine Learning Computer Vision 2 — Questions and Answers
Question 1: Which pooling operation retains the maximum value within each pooling window?
- Average pooling
- Max pooling (Correct answer)
- Min pooling
- Global pooling
Correct answer: Max pooling
Max pooling selects the highest activation value in each region, preserving the most prominent features.
Question 2: What does the term 'receptive field' refer to in the context of CNNs?
- The number of filters in a layer
- The region of input that influences a particular neuron's output (Correct answer)
- The total number of parameters in the network
- The output resolution of the final layer
Correct answer: The region of input that influences a particular neuron's output
The receptive field is the portion of the input image that a particular neuron in the network can 'see' or respond to.
Question 3: In semantic segmentation, what is the output of the model?
- A bounding box per detected object
- A single class label for the entire image
- A class label for every pixel in the image (Correct answer)
- A set of keypoints for each object
Correct answer: A class label for every pixel in the image
Semantic segmentation assigns a class label to each individual pixel, producing a dense prediction map.
Question 4: Which architecture introduced skip connections (residual connections) to combat the vanishing gradient problem?
- VGGNet
- AlexNet
- ResNet (Correct answer)
- Inception
Correct answer: ResNet
ResNet (Residual Network) introduced skip connections that allow gradients to flow directly through the network, enabling very deep architectures.
Question 5: What is the primary purpose of data augmentation in training computer vision models?
- Reducing training time
- Increasing model depth
- Improving generalization by exposing the model to varied input transformations (Correct answer)
- Decreasing memory usage
Correct answer: Improving generalization by exposing the model to varied input transformations
Data augmentation artificially expands the training set with transformations like flips and rotations, helping the model generalize better to unseen data.
Question 6: Which loss function is most commonly used for multi-class image classification?
- Mean squared error
- Binary cross-entropy
- Categorical cross-entropy (Correct answer)
- Hinge loss
Correct answer: Categorical cross-entropy
Categorical cross-entropy measures the difference between predicted probability distributions and true one-hot encoded class labels.
Question 7: What does 'stride' control in a convolutional layer?
- The number of filters applied
- How many pixels the filter moves at each step (Correct answer)
- The padding added around the input
- The depth of the feature maps
Correct answer: How many pixels the filter moves at each step
Stride determines how many pixels the convolutional filter shifts at each step, controlling the spatial dimensions of the output feature map.
Machine Learning Computer Vision 2
Which pooling operation retains the maximum value within each pooling window?