Question 1
1→ ii, 2→ ii, 3 → iv, 4→ i
1→ ii, 2→iv, 3 → i, 4→ iii
1→ ii, 2→ ii, 3 → iv, 4→ iii
1→ ii, 2→ iv, 3 → v, 4→ iii
Deep Learning for Computer Vision Quiz 1: 25 February 2024 (January 2024 term)
The IIT Madras BS Deep Learning for Computer Vision (Deep Learning for Computer Vision) Quiz 1 paper sat on 25 Feb 2024, in the January 2024 term: 23 questions for 36 marks in 120 minutes. Every question is below with its answer. Take it as a timed mock test to be marked, or read it through first.
- 23
- 36
- 120 min
- 8
- 2
- 11
- 2
Show answer
Correct answer
1→ ii, 2→ iv, 3 → v, 4→ iii
Question 2
Which of the following statements is True
Gaussian filter is Separable filter because it is non linear.
Median filter is a non separable filter
Gaussian filter is a High Pass filter.
Mean filter is a non separable filter because it is non linear
Show answer
Correct answer
Median filter is a non separable filter
Question 3
Which of the following statements is false regarding SIFT and SURF?
SIFT and SURF are two most useful approaches to detect and match features because they are invariant to scale, rotation, translation, illumination, and blur
SURF is better than SIFT in rotation invariance, blur and warp transform
SURF is 3 times faster than SIFT because of its use of integral image and box filter
descriptors with dimensions lower than 128 perform pretty well across a range of matching tasks
Show answer
Correct answer
descriptors with dimensions lower than 128 perform pretty well across a range of matching tasks
Question 4
Identify the correct sequence of steps in a Canny edge detection pipeline. Steps listed below: 1. Compute gradient magnitude and direction
2. Connect individual components
3. Smoothen the image
4. Non-maximum suppression along the edge
5. Threshold into strong, weak or no edge
6. Non-maximum suppression perpendicular to the edge
3→ 1→ 4→ 5 → 2
3→ 5→ 1→ 4 → 2
3→ 1→ 6→ 5 → 2
3→ 1→ 5→ 6 → 2
Show answer
Correct answer
3→ 1→ 6→ 5 → 2
Question 5
1→iii, 2 → iv, 3→ i,4→ v
1→ii, 2→i, 3→iv,4→iii
1→iii, 2→i, 3→iv,4→ii
1→iv, 2→i, 3→iii,4→ii
Show answer
Correct answer
1→iv, 2→i, 3→iii,4→ii
Question 6
In RANSAC, what is the probability that the algorithm never selects a whole set of inliers from 20 data points (which has 5 inliers in total) for k = 2 time steps? Here, n = 2 is the minimum number of samples to fit a model.
0.2
0.93
0.87
0
Show answer
Correct answer
0.87
Question 7
Suppose we have a codebook of 2048 SIFT visual words. We extract 1000 SIFT descriptors (SIFT is a 128-dimensional feature) from a new image. What is the dimensionality of the BoW (Bag-of-Words) descriptor?
1000
128
128 × 2048
2048
Show answer
Correct answer
2048
Question 8
1→ iii, 2→ ii, 3 → iv, 4→ i
1→ v, 2→i, 3 → iv, 4→ iii
1→ iii, 2→ i, 3 → ii, 4 →v
1→ ii, 2→ v, 3 → iv, 4→ iii
Show answer
Correct answer
1→ ii, 2→ v, 3 → iv, 4→ iii
Question 9
Which of the following linear filters will leave the central value of the given matrix unchanged.
Show answer
Correct answers
Question 10
Which of the following statements are True?
LoG Filter is not appropriate to be used for blob detection.
Image patches with large contrast changes (gradients) are easier to localize.
Gradients in at least two (significantly) different orientations are the hardest (e.g., corners) to detect.
In case of ‘Hysteresis Thresholding’, if gradient at a pixel is greater than high threshold that means it is an edge pixel.
Show answer
Correct answers
Image patches with large contrast changes (gradients) are easier to localize.
In case of ‘Hysteresis Thresholding’, if gradient at a pixel is greater than high threshold that means it is an edge pixel.
Question 11
Show answer
Correct answer: 0
Question 12
Consider the grayscale image shown below:
If a median filter with a neighborhood is applied to this image with a stride of 1 and no padding, __________ will be the value at position in the filtered image? (Assume that the top-left pixel coordinate of output image is addressed as )?
Show answer
Correct answer: 30
Question 13
Show answer
Correct answer: 143
Question 14
Show answer
Correct answer: -0.35 (accepted within ±0.05)
Question 15
Show answer
Correct answer: 4
Question 16
In case of template matching by using correlation, even if template is not identical to some subimage in the scene, match can be meaningful, if (a) _____________ , (b) _____________ and general appearance is right.
Based on the above data, answer the given subquestions.
Enter the correct answer for Blank (a) ______________ .
Show answer
Correct answer: scale or orientation
Question 17
In case of template matching by using correlation, even if template is not identical to some subimage in the scene, match can be meaningful, if (a) _____________ , (b) _____________ and general appearance is right.
Based on the above data, answer the given subquestions.
Enter the correct answer for Blank (b) ______________ .
Show answer
Correct answer: orientation or scale
Question 18
Based on the above data, answer the given subquestions.
a11 = ____________
Show answer
Correct answer: 1
Question 19
Based on the above data, answer the given subquestions.
a12 = ____________
Show answer
Correct answer: 0.5
Question 20
Based on the above data, answer the given subquestions.
a21 = ____________
Show answer
Correct answer: 0
Question 21
Based on the above data, answer the given subquestions.
a22 = ____________
Show answer
Correct answer: 1
Question 22
The degrees of freedom for translation and affine transformations are (a) __________ and (b) ___________ respectively.
Based on the above data, answer the given subquestions.
Enter the correct answer for blank (a) __________.
Show answer
Correct answer: 2
Question 23
The degrees of freedom for translation and affine transformations are (a) __________ and (b) ___________ respectively.
Based on the above data, answer the given subquestions.
Enter the correct answer for blank (b) __________.
Show answer
Correct answer: 6