0% found this document useful (0 votes)
1K views7 pages

DL - Assignment 11 Solution

This document contains a 10 question multiple choice quiz about concepts in deep learning and computer vision. The questions cover topics like skip connections in image denoising networks, calculating dice coefficients, advantages of fully convolutional networks, challenges in facial recognition, transpose convolutions, semantic segmentation output formats, and max unpooling. For each question there is a multiple choice answer option and a detailed solution explaining the reasoning.

Uploaded by

swathisreejith6
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
1K views7 pages

DL - Assignment 11 Solution

This document contains a 10 question multiple choice quiz about concepts in deep learning and computer vision. The questions cover topics like skip connections in image denoising networks, calculating dice coefficients, advantages of fully convolutional networks, challenges in facial recognition, transpose convolutions, semantic segmentation output formats, and max unpooling. For each question there is a multiple choice answer option and a detailed solution explaining the reasoning.

Uploaded by

swathisreejith6
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

NPTEL Online Certification Courses

Indian Institute of Technology Kharagpur

Deep Learning
Assignment- Week 11
TYPE OF QUESTION: MCQ/MSQ
Number of questions: 10 Total mark: 10 X 1 = 10
______________________________________________________________________________

QUESTION 1:
What is the use of Skip Connection in image denoising networks?

a. Helping de-convolution layer to recover an improved clean version of image.


b. Back propagating the gradient to bottom layers, which makes the training easy.
c. To create the direct path between convolution layer and the corresponding
mirror de-convolution layer.
d. All of the above.

Correct Answer: d

Detailed Solution:

Please refer to lecture of week 11.


__________________________________________________________________________

QUESTION 2:
What will be the value of dice coefficient between A and B?

(Consider, |A|= sum of all elements.)


a. 0.23
b. 0.77
c. 0.11
d. 0.89
Correct Answer: a
NPTEL Online Certification Courses
Indian Institute of Technology Kharagpur

Detailed Solution:

|A|=7.82

|B|=8

A∩B = |A∩B|=7.42

Dice Coefficient = 2*|A∩B|/(|A|+|B|)=2*7.42/ (8+7.82) =0.23

QUESTION 3:
What will be the dice coefficient of following two one hot encoded vector? (|A|=no of 1 bit)

A 1 0 1 0 0 0 1 1 1 0 0 1 0 1
B 1 0 0 0 0 1 1 1 0 0 0 1 0 0

a. 0.83
b. 0.41
c. 0.67
d. 0.90

Correct Answer: c

Detailed Solution:
No of 1 bit in A=7
No of 1 bit in B =5
Overlapping 1 bit =4
2∗|𝐴∩𝐵| 2∗4
Dice Coefficient = |𝐴|+|𝐵|
= 5+7 = 0.67
______________________________________________________________________________

QUESTION 4:
Fully Connected Convolutional network or FCN became one of the major successful network
architectures. Can you identify what are the advantages of FCN which makes it a successful
architecture for semantic segmentation?

a. Larger Receptive Field


b. Mixing of global feature
c. Lesser computation required
d. All of the above
NPTEL Online Certification Courses
Indian Institute of Technology Kharagpur

Correct Answer: d

Detailed Solution:
______________________________________________________________________________
QUESTION 5:
What are the different challenges one face while creating a facial recognition system?

a. Different illumination condition


b. Different pose and orientation of face images
c. Limited dataset for training
d. All of the above

Correct Answer: d

Detailed Solution:
Please refer to the lecture of week 11.
____________________________________________________________________________

QUESTION 6:
Suppose you have a 1𝐷 signal 𝑥 = [5,4,3,2,1] and a filter 𝑓 = [1,2,3,4], and you perform stride
2 transpose convolution on the signal 𝑥 by the filter 𝑓 to get the signal 𝑦. What will be the
signal 𝑦 if we don’t perform cropping?

a. 𝑦 = [1,2,5,8,9,14,13,20,19,26,3,4]
b. 𝑦 = [5, 10, 19, 28, 15, 22, 11, 16, 7, 10, 3, 4]
c. 𝑦 = [1,2,5,8,9,14,13,20,17,26,15,20]
d. 𝑦 = [0,0,5,8,9,14,13,20,19,26,0,0]

Correct Answer: b

Detailed Solution:
5 4 3 2 1

1*5=5 1

2*5=10 2

3*5+1*4=19 3 1

4*5+2*4=28 4 2

3*4+1*3=15 3 1
NPTEL Online Certification Courses
Indian Institute of Technology Kharagpur

4*4+2*3=22 4 2

3*3+1*2=11 3 1

4*3+2*2=16 4 2

3*2+1*1=7 3 1

4*2+2*1=10 4 2

3*1=3 3

4*1=4 4

______________________________________________________________________________

____________________________________________________________________________

QUESTION 7:
Which of following can be a target output of semantic segmentation problem with 4 class?

a.
0 1 0 1 1 0 0 0 1 0 0 0
0 1 0 0 0 0 1 0 0 1 0 0
1 0 0 0 0 0 0 0 0 0 0 0
I II III IV
b.
0 1 0 1 0 0 0 0 1 0 1 0
0 1 0 0 1 0 1 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 1 1
I II III IV
c.
0 1 0 1 0 0 0 0 1 0 0 0
0 1 0 0 0 1 1 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 1 1
I II III IV
d.
0 1 0 1 0 0 0 0 1 0 0 0
0 1 0 1 0 0 1 0 0 0 0 0
1 1 0 0 1 0 0 0 1 0 1 1
I II III IV

Correct Answer: c
NPTEL Online Certification Courses
Indian Institute of Technology Kharagpur

Detailed Solution:
Target output should be one hot encoded vector at every pixel location. It should one if the pixel
belongs to that particular class otherwise 0.

______________________________________________________________________________

QUESTION 8:
Which of the following is true for semantic segmentation?

a. Semantic Segmentation can be considered as pixel wise classification problem.


b. Semantic Segmented output has same dimension as the input image dimension.
c. It has application in Autonomous driving, Industrial inspection, and Medical
imaging analysis.
d. All of the above

Correct Answer: d

Detailed Solution:

Please refer to the lectures of week 11.

____________________________________________________________________________
NPTEL Online Certification Courses
Indian Institute of Technology Kharagpur

QUESTION 9:
In a Deep CNN architecture, the feature map before applying a max pool layer with (2x2) kernel
is given bellow.
50 30 3 4
14 12 7 10
44 1 14 19
2 5 16 20
After few successive convolution layers, the feature map is again up-sampled using Max Un-
pooling. If the following feature map is present before Max-Unpooling layer, what will be the
output of the Max-Unpooling layer?
50 10
44 20

a.
50 0 0 0
0 0 0 10
44 0 0 0
0 0 0 20
b.
50 0 10 0
0 0 0 0
44 0 20 0
0 0 0 0
c.
50 0 0 10
0 0 0 0
44 0 0 20
0 0 0 0
d.
0 0 0 0
0 50 0 10
0 0 0 0
0 44 0 20
NPTEL Online Certification Courses
Indian Institute of Technology Kharagpur

Correct Answer: a
Detailed Solution:
Please refer to lectures of week 11.
____________________________________________________________________________

QUESTION 10:
Which of the following operation reduces spatial dimension of features?

a. Max un-Pooling
b. Convolution with 3 × 3 Kernel, Stride=2, Padding all sides = 1
c. Convolution with 3 × 3 Kernel, Stride=1, Padding all sides = 1
d. Transposed convolution

Correct Answer: b
Detailed Solution:

Stride of 2 reduces the spatial dimension by a factor of 2.

____________________________________________________________________________

______________________________________________________________________________

************END*******

You might also like