Additional Exam Notes

Question 1

Explain the term Epipolar Plane

Answer 1

It’s a combination of the image feature and two optical centres, all of which define a plane of interpretation.
The world feature generating the known image feature and the corresponding feature in the other view must lie in this plane.

Question 2

Why does the epipolar plane simplify the stereo correspondence problem?

Answer 2

• The intersection of the plane of interpretation and the second image plane is a straight line (called an epipolar line)
  The epipolar plane determines the epipolar lines. Given a feature extracted from one view, the corresponding feature must lie on the corresponding epipolar line in the other.
• Search space is reduced from two dimensions to one

Question 3

How is the Integral Image calculated?

Answer 3

• Computes a value at each pixel, that is the sum of the pixel values above and to the left of the source pixel
• Cumulative row sum: s(x, y) = s(x-1, y) + i(x, y)
• Integral image/ii(x, y) = ii(x, y-1) + s(x, y)

Question 4

What benefit does using Integral images bring?

Answer 4

• Advantages of speed i.e. more efficient calculations. Only 4 numbers are needed to calculate the sum of intensity values for each rectangle

Question 5

What are advantages of a pinhole camera?

Answer 5

• Simple to understand
• No lens distortion due to lack of lens
• Infinite depth of field: No depth of field effect distorting the image

Question 6

What is a one line description for segmentation?

Answer 6

Assign all pixels to objects

Question 7

What is a one line description for Recognition?

Answer 7

Identify the main object in the image

Question 8

What is a one line description for Detection?

Answer 8

Find the location of all objects

Question 9

What is a one line description for Pose?

Answer 9

Find all of the object parts

Question 10

What are characteristics of a good feature?

Answer 10

• Invariant to scale and rotation
• Reflect useful object properties
• Unique and repeatable

Question 11

What are example types of object recognition tasks?

Answer 11

• Segmentation
• Pose Estimation
• Detection

Question 12

Why are Diffusion Models preferred over GANs?

Answer 12

• Easier to train
• Generate high-quality images
• Avoid GAN’s issues such as mode collapse or tough to train due to discriminator balance

Question 13

What are important concepts in Background Subtraction?

Answer 13

• Gaussian Mixture Models
• Thresholding

Question 14

What features are commonly used in Classical Computer Vision?

Answer 14

• HOG descriptors
• Colour histograms
• Texture Features

Question 15

Describe the structure of a pinhole camera

Answer 15

• The pinhole camera’s optical point/centre is located on the principal axis / Z axis
• The 3D scene is projected onto the image plane through the pinhole located at the optical point
• The focal length narrows or widens the field of view by increasing or decreasing in length respectively.

Question 16

Explain the forward process in a diffusion model

Answer 16

• Noise is added in a Markov Chain
• Each step’s noise is only dependent on the previous step
• This is a fixed process, containing no learned parameters.

Question 17

Explain the reverse process in a Diffusion model

Answer 17

• Train and use a U-net style model that learns the reverse diffusion process
• This is trained by predicting the noise in the image given the current timestep
• By removing predicted noise from the image, we get an approximation of the original image

Question 18

Describe the process of object detection for classification

Answer 18

• Features are extracted using feature extraction methods
• They are then used to train a classifier model, which identifies regions in test images likely to contain objects and classifies them accordingly

Question 19

Explain how a particle filter updates object tracking

Answer 19

• Each particle represents a hypothesis
• Predictions are made based on a motion model
• Measured positions are then used to update the particle weights based on how well they match the observation

Question 20

Describe the architecture of a U-net model?

Answer 20

• U-net uses an encoder to compress features
• Followed by a decoder to reconstruct segmentation maps
• Skip connections between matching levels of encoder and decoder to preserve spatial information

Question 21

Explain both semantic and instance segmentation

Answer 21

• Semantic labels every pixel by class
• Instance distinguishes individual objects of the same class

Question 22

What are the challenges of using a single Gaussian Background Model?

Answer 22

• Cannot model dynamic backgrounds
• Sensitive to lighting changes
• Cannot represent multiple background modes at a pixel

Question 23

What’s an advantage and disadvantage of using a Particle Filter over a Kalman Filter?

Answer 23

Advantage - Particle filters handle multi-modal distributions
Disadvantage - Requires more computation

Question 24

Why is SIFT useful?

Answer 24

• Invariant to scale, rotation and illumination
• It’s robust in regards to matching across different views

Question 25

What is the importance of illumination modelling?

Answer 25

Helps accurately interpret scene radiance, ensuring features and colours are not misinterpreted due to lighting changes

Question 26

What factors affect the quality of images captured in digital photography?

Answer 26

- Shutter speed - Aperture - ISO Setting

Question 27

What are valid applications of saliency prediction in Computer Vision?

Answer 27

- Smart image cropping - Object Detection - Content-aware compression

Question 28

In Stereo vision, what helps estimate depth from two images?

Answer 28

- Epipolar geometry: Contains correspondence search - Disparity maps: Allow depth calculation

Question 29

What are components of a Kalman filter used in tracking?

Answer 29

- Motion model - Measurement Update - State Prediction

Question 30

Explain how ISO, Shutter Speed and Aperture interact to control image brightness?

Answer 30

- ISO controls sensor sensitivity - Aperture determines how much light enters the lens - Shutter Speed determines how long the light is exposed for Increasing one typically requires reducing another to maintain consistent exposure

Question 31

Describe how disparity maps are used in Stereo Vision to infer depth

Answer 31

Disparity maps represent pixel shifts between left and right images. Depth is inversely proportional to disparity. The larger the disparity, the closer the object is to the camera that took the image.

Question 32

How does a Kalman Filter handle uncertainty during object tracking?

Answer 32

Kalman filters maintain a mean and a covariance for state estimates. Prediction steps add uncertainty and measurement updates reduce it

Question 33

What is the formal way of explaining the role of Pooling layers?

Answer 33

Pooling layers downsample feature maps to reduce computation and enforce spatial hierarchy.

Question 34

What is the formal way of explaining the role of Activation Functions in CNNs?

Answer 34

Activation functions introduce non-linearity, enabling complex pattern learning.

Question 35

What is the function of a Disparity Map in Stereo Vision?

Answer 35

Disparity maps capture the shift of corresponding pixels between stereo image pairs. This disparity allows calculation of depth via triangulation

Question 36

Compare background subtraction using a single Gaussian model against a Gaussian Mixed Model

Answer 36

Single Gaussian: - Assumes a static background Mixed Gaussian: - Can represent dynamic backgrounds like moving leaves by modelling multiple modes per pixel

Question 37

How does Motion Difference capture movement?

Answer 37

Motion difference compares sequential frames to detect pixel intensity changes, indicating movement

Question 38

What is a major challenge of evaluating saliency maps?

Answer 38

Human attention is subjective, making ground truths hard to define. Multiple metrics exist, but interpreting results remains complex

Question 39

What is the main advantage of using U-Net in image segmentation tasks?

Answer 39

U-Net preserves spatial resolution through skip connections and so they are very efficient for small dataset segmentation tasks.