AIS412 - Lecture 1
AIS412 - Lecture 1
AIS412 - Lecture 1
Deep Learning
for Computer
Vision
AIS412
MUSTAFA ELATTAR
7
typical perception pipeline
representation
‘fancy math’
output
typical perception pipeline
representation
‘fancy math’
‘fancy math’
‘fancy math’
12
Important note:
In general, computer vision does not
work
(except in certain situations/conditions)
13
Applications of computer vision
14
Machine Vision
Age recognition
Sony Cyber-shot
Smile recognition
Face makeovers
BMW 5 series
Mission:
The mission of the school is to contribute to the development
of cultural values and to information technology-driven
economies in the region through the pursuit of education,
research, innovation and entrepreneurship at the highest
levels of excellence.
AIS412 - ILOs
Explain the basic components and workings of deep learning algorithms, including CNNs,
A1
RNNs, attention mechanisms, encoder-decoder models, and generative models.
List and describe real-life use cases of deep learning, including voice, NLP, and vision
A2
applications.
Identify the limitations of traditional machine learning and the advantages of deep learning over
A3
machine learning.
Apply deep learning algorithms to real-life scenarios, including developing and training neural
B1
networks for various applications.
Investigate and analyze the performance of deep learning models and identify areas for
B2
improvement.
Combine and test different deep learning techniques to solve complex problems and explore
B3
creative applications.
C1 Design and suggest deep learning solutions for specific use cases and evaluate their effectiveness.
Present and report on the results of deep learning projects, including analysis of performance and
C2
areas for improvement.
Link and judge the ethical and social implications of deep learning, including issues of bias and
C3
fairness in algorithmic decision-making.
D1 Collaborate with peers in developing and testing deep learning solutions.
D2 Apply entrepreneurial skills to develop and explore creative applications of deep learning.
AIS412 - Syllabus
Date Topics Lab Assignments
Topic 1: Course Introduction
30 Sep
Topic 2: Advanced Image processing - Hough transforms
Topic 2: Advanced Image processing - Hough transforms
7 Oct Lab 1
Topic 3: Feature Detection - Corner detection
Topic 3: Feature Detection - Corner detection
14 Oct Lab 2 Assignment 1
Topic 4: Feature Detection - Feature descriptors
21 Oct Topic 5: Stereo Vision Lab 3 Assignment 1 Deadline
Topic 6: Motion Tracking: Optical flow (LK, HS) - Tracking (KLT, Mean-Shift)
28 Oct Lab 4 Assignment 2
Topic 7: Image Registration: Correspondence Finding
4 Nov Topic 8: Object Features Learning: K-means, Bag of words Lab 5 Assignment 2 Deadline
11 Nov Midterm Week
Topic 9: Classification, Loss Functions and Optimization
18 Nov Topic 10: Convolutional Neural Networks Lab 6 Project registered for each five students
Topic 11: Training Neural Networks
Topic 12: CNN Architectures (VGG, GoogLeNet, ResNet, etc)
25 Nov Lab 7 Assignment 3
Topic 13: Recurrent Neural Networks
2 Dec Topic 14: Detection and Segmentation Lab 8 Assignment 3 Deadline
9 Dec Topic 15: Generative Models Lab 9
16 Dec
23 Dec Backup Week
30 Dec
ADD FOOTER HERE
Project Submission, Discussion, and Oral Presentation 36
6 Jan Study Week
AIS412 – Grading Scheme and Resources
Grading Policy:
● Attendance 3%
● 3 Assignments 18% (3*6)
● Tutorial and Lab 14% (9*1.5)
● Quizzes 5%
● Project 15%
● Midterm 15%
● Final 30%
Handouts:
● Lectures + Labs
● Textbook
● Computer Vision: Algorithms and Applications, 2nd ed. by Richard
Szeliski
● Deep Learning by Ian Goodfellow
TA:
Eng. Aly Abdelmegeid
alymohamed@nu.edu.eg
Office: 220
Office hours: Monday 11:00 to 1:00
Finding boundaries
Line fitting
Line parameterizations
Hough transform
Hough circles
TEACH A COURSE 42
Finding Boundaries
43
Where are the object boundaries?
Human annotated boundaries
Edge detection
Multi-scale edge detection
Edge strength does not necessarily correspond to our
perception of boundaries
Where are the object boundaries?
Human annotated boundaries
Edge detection
Defining boundaries
are hard for us too
Where is the boundary of the mountain top?
Applications
55
Line fitting
Given: Many pairs
Find: Parameters
Using:
Note:
Huge E!
Problems with parameterizations
Where is the line that minimizes E?
• Cost:
– It is not feasible to check all combinations of features by
fitting a model to each possible subset
61
Slope intercept form
slope y-intercept
Double intercept form
x-intercept y-intercept
Derivation:
(Similar slope)
Normal Form
Derivation:
plug into:
Hough transform
65
Hough transform
• Generic framework for detecting a parametric model
66
Image and parameter space
variables variables
parameters parameters
a line
becomes a
point
parameters
Image space
Image and parameter space
variables variables
parameters parameters
a point
becomes a
line
parameters parameters
two points
become
?
parameters parameters
two points
become
?
parameters parameters
three points
become
?
parameters parameters
three points
become
?
parameters parameters
four points
become
?
parameters parameters
four points
become
?
Al g o r i t h m:
3. Set
1 1
4 . F o r e a c h image e d g e 1
1 1
1
If l i e s on t h e l i n e :
1 1
1 1
Increment 1 1
5 . F i n d l o c a l maxima i n
Problems with parameterization
How big does the accumulator need to be for the parameterization ?
1 1
1 1
1 1
2
1 1
1 1
1 1
Problems with parameterization
How big does the accumulator need to be for the parameterization ?
1 1
1 1
1 1
2
1 1
1 1
1 1
Image Space
Hough Space Sinusoid
?
(Finite Accumulator Array Size)
Hough Space
Image and parameter space
variables parameters
parameters variables
a point
becomes?
parameters variables
a point
becomes a
wave
parameters
a line
becomes?
parameters
a line
becomes a
point
parameters
a line
becomes?
parameters
a line
becomes a
point
parameters
a line
becomes a
point
parameters
a line
becomes a
point
parameters
a line
becomes a
point
parameters
a line
becomes a
point
parameters
Wait …why is rho negative?
a line
becomes a
point
Recall:
Image and parameter space
variables
same line through the point
a parameters
line becomes a point
parameters
two points
become
?
parameters
three points
become
?
parameters
four points
become
?
15
10
5
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
Noise level
0
20 40 60 80 100 120 140 160 180 200
Number of noise points
108
Let’s assume radius known
parameters parameters
variables variables
variables variables
variables variables
parameters parameters
variables variables
parameters parameters
variables variables
parameters parameters
variables variables
What if radius is unknown?
parameters parameters
variables variables
UseAccumulator array
Edge Location
Edge Direction
variables variables
parameters parameters
variables variables
Pennie Hough detector Quarter Hough detector
Pennie Hough detector Quarter Hough detector
The Hough transform
Deals with occlusion well?
Robust to noise?
123
Detecting shape features