AI Computer Vision - Practice Questions 2026

Core Concepts: Covers the essential building blocks of modern CV, including convolutional neural networks (CNNs), pooling layers, and activation functions. You will be tested on how these components interact to extract features.

Intermediate Concepts: Moves into specialized architectures. Expect questions on learn object detection from zero to hero frameworks (YOLO, SSD), image segmentation (U-Net, Mask R-CNN), and the mechanics of transfer learning.

Advanced Concepts: Dives into complex topics such as Generative Adversarial Networks (GANs), Vision Transformers (ViTs), and 3D computer vision. This section challenges your understanding of state-of-the-art research.

Real-world Scenarios: Focuses on deployment and optimization. You will solve problems related to model quantization, edge computing, and handling imbalanced datasets in production.

Mixed Revision / Final Test: A comprehensive simulation of a high-pressure exam environment, pulling questions from all previous sections to test your retention and speed.

Option 2: Max Pooling

Option 3: Batch Normalization

Option 4: Zero Padding

Option 5: Dropout

CORRECT ANSWER: Option 2

CORRECT ANSWER EXPLANATION: Max Pooling is a down-sampling strategy that selects the maximum value from a defined window. This reduces the computational load and provides a form of translation invariance by highlighting the most prominent features in a local region.

WRONG ANSWERS EXPLANATION:

Option 3: Batch Normalization stabilizes the learning process by re-centering and re-scaling inputs; it maintains the dimensions of the feature map.

Option 4: Zero Padding actually increases or maintains the spatial dimensions by adding zeros around the border.

Option 5: Dropout is a regularization technique that randomly sets input units to 0 during training to prevent overfitting; it does not change the shape of the feature map.

Option 2: The number of layers in the backbone network

Option 3: The overlap between the predicted bounding box and the ground truth

Option 4: The learning rate decay schedule

Option 5: The color histogram similarity

CORRECT ANSWER: Option 3

CORRECT ANSWER EXPLANATION: IoU is the standard metric for measuring the accuracy of an object detector. It is calculated by dividing the area of overlap between the predicted box and the ground truth box by the area of their union.

WRONG ANSWERS EXPLANATION:

Option 2: This refers to the architecture complexity, which is independent of the box overlap calculation.

Option 4: Learning rate decay is an optimization hyperparameter, not an evaluation metric for localization.

Option 5: Color histogram similarity is a traditional image retrieval technique, not a spatial overlap metric for bounding boxes.

Option 2: Data Augmentation

Option 3: Overfitting

Option 4: Vanishing Gradient

Option 5: Internal Covariate Shift

CORRECT ANSWER: Option 3

CORRECT ANSWER EXPLANATION: Overfitting occurs when a model captures the "noise" or random fluctuations in the training data rather than the underlying pattern. This leads to high training accuracy but low validation/test accuracy.

WRONG ANSWERS EXPLANATION:

Option 2: Data Augmentation is a technique used to prevent overfitting by artificially increasing the dataset size.

Option 4: Vanishing Gradient is a training issue where gradients become too small for the weights to update effectively in deep networks.

Option 5: Internal Covariate Shift refers to the change in the distribution of network activations during training, which Batch Normalization aims to solve.

Access to a huge, original question bank that covers the latest industry trends.

Direct support from instructors to clear up any confusion on complex topics.

cissp exam tests 500 questions detailed explanations 2026 for every single question, including why wrong answers are incorrect.

Full mobile compatibility via the Udemy app, allowing you to study on the go.

A 30-day money-back guarantee—if you are not satisfied, you can request a refund with no questions asked.