Exam

Question 1

Describe the steps of Image classification

Answer 1

1) Feature extraction
2) Feature description
3) Classification

Question 2

Describe the typical pre-processing of digit recognition in general images

Answer 2

1. Detect the digits in the large image
2. Normalize the size of the digit, for example, to 28x,28 pixels
3. Normalize the location, place mass center in the middle
4. “Slant” Make the orientation canonical

Question 3

Name some advantages and disadvantages of K-Nearest Neighbour

Answer 3

1. It works reasonably well
2. No training required
3. Nonlinear decision boundaries
4. Multi-class
5. All training data must be stored in memory
6. Long evaluation time

Question 4

Name the three conditions Canny proposed for a good edge detector

Answer 4

1) Good detection; should detect all edges
2) Good localization; should detect edges where they are
3) A single response, should only detect edges where they are

Question 5

Describe Canny’s algorithm for edge detection

Answer 5

1) Gaussian filtering
2) Calculate gradient magnitude and direction
3) Perform non-maximum repression
4) Perform hysteresis thresholding

Question 6

How can we approximate E(u,v) the error when shifting the neighborhood by u,v pixels?

Answer 6

E(u,v) = [u,v] M [u, v]^T where M is the matrix:

M = [Ix^2 Ixy
Iyx Iy^2]

and Ix, Iyx, Iy are the derivatives.

Question 7

Given the M matrix, what metrics can be used to determine if we have a corner?

Answer 7

1) R = min(lambda_1, lambda_2)
2) R = lambda_1lambda_2 /(lambda_1 + lambda_2 + epsilon)
3) R = lambda_1lambda_2 - k(lambda_1 + lambda_2)^2
4) R= det(M) - k*trace(M)^2

where lambda_1 and lambda_2 are the eigenvalues

Question 8

Describe the Harris Corner Detector

Answer 8

1) compute Ix and Iy
2) Create M
3) calculate eigenvalues lambda_1, lambda_2
4) Calculate the Respons R = lambda_1lambda_2 + k(lambda_1 + lambda_2)^2
5) Threshold and Non-Maxima repression with respect to R

Question 9

How are DoG and LoG filters correlated

Answer 9

DoG(x,y,k,sigma) = I * G(ksigma) - I *G(sigma) 
approx= (k-1)sigma^2Log(x,y,sigma)

Question 10

Describe the SIFT algorithm for keypoint detection

Answer 10

1) Find scale/ space extrema using DoG response
2) Fit a quadratic function over space to the extremes and estimate a refined new keypoint (can be in between pixels…)
3) Threshold keypoint responses

Question 11

Describe how we can determine the canonical orientation of each patch in the SIFT algorithm

Answer 11

1) Create a histogram of 36 bins for degrees from 0-360
2) Each pixel in a neighborhood vote for an orientation weighted by gradient magnitude.
3) The keypoint is assigned an orientation corresponding to the largest bin.

Question 12

How is the SIFT descriptor created?

Answer 12

1) Take a small window around the keypoint
2) weigh gradients near the center more.
3) Calculate the gradient orientation and magnitude after using a Gaussian filter.
4) Calculate the canonical orientation
5) Rotate all gradient directions relative to the canonical orientation
6) Create a histogram of gradient orientations for each subregion in the local window (Originally 16 subregions and 8 direction histogram).
7) These 16 histograms form a 128-feature vector and are used as a descriptor.

Question 13

Describe the main advantages of SIFT

Answer 13

1) Robust to intensity changes
2) Invariant to scale
3) Invariant to rotation

Question 14

What is the main difference between SURF and SIFT?

Answer 14

SURF is faster since it only considers horizontal and vertical gradients using the Haar wavelet. It uses 4 descriptors for each subregion, (sum dx, sum dy, sum |dx| , sum |dy|) and the total descriptor is of length 16*4=64.

Question 15

Name some improvements to neural networks the past 30 years

Answer 15

1 ) Better hardware

2) Deeper networks
3) Larger datasetets
4) Other changes, better activation funcitons, different layers…

Question 16

How can we adapt gradient descent to fix the vanishing and exploding gradient problems?

Answer 16

1) We can use adaptiv stepsizes.

2) We can Clip the gradient using thresholding or L2 norm.

Question 17

What is the R_CNN method?

Answer 17

Uses selective search for region proposal, SVM for classification and linear regression for localization. Both use CNN features.

Question 18

What is the fast R_CNN method

Answer 18

Uses selective search for proposal, CNN for localization and classification

Question 19

What is a RoI pooling layer?

Answer 19

Converts convo feature maps into a fixed size. Used because region proposals can be of arbitary size.

Question 20

What is the main difference between fast R_CNN and faster R_CNN?

Answer 20

Faster R_CNN uses a a proposal CNN. The proposal and and detection networks share feature maps.

Question 21

How does the Region Proposal Network work in faster R_CNN?

Answer 21

It uses a 3x3 sliding window On the convo feature map and proposes a number of bounding boxes with different scales, aspects and anchors.

Question 22

Name som advantages/ disadvantages of K-Means clustering for segementation

Answer 22

+ Fast and easy to implement

- No semantics and supervised methods often perform better

Question 23

How can we use K-means clustering for image segmentation?

Answer 23

1) On the histogram of the pixel intensities
2) Colour similarity
3) position and colour similarity
4) Others…

Question 24

How can we use a classification convo net for segmentation?

Answer 24

Change the fully connected parts at the end to convo. This gives a heat map of class probabilities. Use tranposed convolayer to upsample this heat map.

Question 25

What are the basic assumptions behind optical flow?

Answer 25

Brightness constancy and small displasement

Question 26

What do we know about I(x+u, y+v, t+1) given displacement (u,v) and the brigthness constancy assumption?

Answer 26

It is the same as I(x,y,t)

Question 27

What is the optical flow constraint?

Answer 27

dI/du * u + DI/dv * v + dI/Dt = 0

Question 28

What assumption does the the Lucas Kanade method make?

Answer 28

Flow, or displacement, is constant in a local neighbourhood

Question 29

What is the condition number of a 2-d matrix?

Answer 29

abs(lambda_max) / abs(lambda_min)

Question 30

How can the condition number be interpreted?

Answer 30

Large conditon number makes the matrix inverse sensitive to noise and small changes in values.

Question 31

What is the idea behind the Horn- Schunk method?

Answer 31

Define a global energy function:

Integral_(x,y) ((Ix*u) + (Iy * v) + I_t)^2 +
alpha(abs(gradient(u))^2 + abs(gradient(u))^2) dxdy

And minimize the energy with respect to u(x,y), v(x,y).

Question 32

What can we do if the small displacement condition does not hold?

Answer 32

Use spatial pyramids of downsampled images. Estimate motion from the coarsest (smallest) image first and iteratively to larger versions.

Question 33

Derive the optical flow constraint

Answer 33

Constant brightness: I(x+u, y+v, t+1) = I(x,y,t)
Taylor: I(x+u, y+v, t+1) = I(x,y,t) + Ixu + Iyu + It
Combining: Ix + Iy + It = 0

Question 34

What is the:

1) formula for calculating output sizes of convo layers
2) formula for calculating k for dilation layers

Answer 34

1) (Xin + 2p - k) / s + 1

2) k + (k-1)(d-1)