Download - Computer Vision - University of Michigan
![Page 1: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/1.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
EECS 442Computer Vision
Justin Johnson & David Fouhey
Winter 2021https://web.eecs.umich.edu/~justincj/teaching/eecs442/WI2021/
1
![Page 2: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/2.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Lecture 1: Introduction
2
![Page 3: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/3.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Goals of Computer Vision
Get a computer to understand
3
![Page 4: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/4.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Goals of Computer Vision
Goal: Naming
4
![Page 5: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/5.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Goals of Computer Vision
Goal: Naming
5
![Page 6: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/6.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Goals of Computer Vision
Goal: Naming
6
![Page 7: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/7.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Goals of Computer Vision
Goal: 3D Structure
7
![Page 8: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/8.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Goals of Computer Vision
Goal: Actions
8
![Page 9: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/9.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Goals of Computer Vision
Goal: Matching
9
![Page 10: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/10.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Something I’m excited about…
10
![Page 11: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/11.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Generating Images
Karras et al, “Progressive Growing of GANs for Improved Quality, Stability, and Variation”, ICLR 2018
11
![Page 12: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/12.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Generating Images from Language
12
Ramesh et al, “DALL·E: Creating Images from Text”, 2021. https://openai.com/blog/dall-e/
![Page 13: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/13.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Generating Images from Language
13
Ramesh et al, “DALL·E: Creating Images from Text”, 2021. https://openai.com/blog/dall-e/
![Page 14: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/14.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Reconstructing the 3D World
14
![Page 15: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/15.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Giving People Tools
15
![Page 16: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/16.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Why is this hard?
All of this stuff seems obvious and effortless!
Why do we need a whole subfield of CS for this?
Key Concept: We see with both our eyes and our brain
16
![Page 17: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/17.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Why is this hard?
17
![Page 18: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/18.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Why is this hard?
18
![Page 19: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/19.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Why is this hard?
19
![Page 20: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/20.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Why is this hard?
Problem: Semantic Gap
20
![Page 21: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/21.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Why is it important?
This image by David Wilson is licensed under CC BY 2.0. No changes made.
Understand Web Data
Instagram: 100 million photos and videos uploaded per day
YouTube: 300 hours of video uploaded every minute
Looking at all content created in one day would take >50 years
21
![Page 22: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/22.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Why is it important?
Autonomous Vehicles
This image is licensed under CC BY-SA 4.0. No changes made.
22
![Page 23: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/23.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Why is it important?
Robotics
This image is licensed under a CC BY 2.0 license. No changes made. Image source
23
![Page 24: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/24.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Why is it important?
Augmented / Virtual Reality
This image is licensed under a CC BY-SA 4.0 license. No changes made. This image is licensed under a CC BY-SA 3.0 IGO license. No changes made.
24
![Page 25: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/25.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Why is it important?Science and Medicine
Medical Imaging
Levy et al, 2016 Figure reproduced with permission
Galaxy Classification
Dieleman et al, 2014 From left to right: public domain by NASA, usage permitted by ESA/Hubble, public
domain by NASA, and public domain.
Whale recognition
This image by Christin Khan is in the public domain and originally came from the U.S. NOAA.Kaggle Challenge
25
![Page 26: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/26.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
We have made great progress
Enter Deep Learning
26
![Page 27: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/27.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Explosion of Computer VisionPublications at top Computer Vision conference:Number of submitted and accepted papers at CVPR
27
71
20
64
24
51
65
33
09
26
80
21
45
21
23
18
07
17
98
19
33
16
77
17
24
14
64
15
93
12
50
11
31
11
60
87
39
05
92
04
66
50
34
53
54
45
51
47
24
31
35
74
91
32
73
00
20
02
40
0
1000
2000
3000
4000
5000
6000
7000
8000
20
21
20
20
20
19
20
18
20
17
20
16
20
15
20
14
20
13
20
12
20
11
20
10
20
09
20
08
20
07
20
06
20
05
20
04
20
03
20
02
20
01
20
00
19
99
19
98
19
97
19
96
19
95
19
94
19
93
19
92
19
91
19
90
19
89
19
88
19
87
19
86
19
85
Submitted
Accepted
![Page 28: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/28.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Despite our success, computer vision still has a long way to go.
Computer vision is far from solved!
28
![Page 29: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/29.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
“The Elephant in the Room”
Modern object detectors seem to work well!
Rosenfeld et al, “The Elephant in the Room”, arXiv 2018
29
![Page 30: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/30.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
“The Elephant in the Room”
Rosenfeld et al, “The Elephant in the Room”, arXiv 2018
”Unusual” objects are often missed!
Computer Vision systems ”see” very differently than we do
30
![Page 31: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/31.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
This image is copyright-free United States government work Example credit: Andrej Karpathy
31
![Page 32: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/32.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Course Logistics
32
![Page 33: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/33.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Meeting Times
• Lecture: • Section 1: Tue / Thu 10:30am – 12:00pm on Zoom
• Section 2: Tue / Thu 12:00pm – 1:30pm on Zoom
• Office Hours• Lots of office hours! Schedule TBD
• GSI Office Hours: Questions about homework, code
• Faculty Office Hours: Prefer conceptual questions
33
![Page 34: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/34.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Prerequisites
You absolutely need:
- Programming: EECS 281
- Linear Algebra (new): Math 214, 217, 296 or 417- Reach out to us if you have equivalent experience
- We will have a one-lecture math refresher
You’ll have to learn: Numpy + PyTorch, a little tiny bit of continuous optimization
35
![Page 35: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/35.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Prerequisites
Suppose K in R3x3, x in R3 .Should know:
• How do I calculate Kx?
• When is K invertible?
• What is x if Kx = λx for some λ?
• What’s the set {y: xTy = 0} geometrically?
You should also be able to remember some notion of a derivative
36
![Page 36: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/36.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Websites
• Course website: https://web.eecs.umich.edu/~justincj/teaching/eecs442/WI2021/
• Piazza: https://piazza.com/class/umich/winter2021/eecs442
• Canvas / Autograder for code submission
• Gradescope for writeup submission
37
![Page 37: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/37.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Piazza
• Please ask questions on Piazza so we can answer the question once, officially, and quickly
• We will monitor Piazza in a systematic way, but we do not guarantee instant response times
• Please don’t ask questions about course material, homework, or logistics over email – those should all go to Piazza
• If you have a sensitive or private issue, then email me / David directly
38
![Page 38: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/38.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Textbooks
No required textbook.
Szeliski, Computer Vision: Algorithms and Applications, is a good reference and available online.
http://szeliski.org/Book/
39
![Page 39: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/39.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Evaluation
• Homework (76%)• Programming assignments in Python / numpy / Pytorch
• HW0 is a numpy crash course (6%)
• HW1-6 are guided mini-projects
• HW1-5 are 12% each
• HW6 is 10%
• Project (24%)• Project Proposal (2%)
• Virtual Project Showcase (8%)
• Project Report (14%)
40
![Page 40: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/40.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Homework Late Policy
• Penalty: 1% per hour, rounded to nearest
• Example:• Due: Midnight Mon. (1s after 11:59:59pm Mon)
• Submitted at 12:15am Tue: No penalty!
• Submitted at 6:50am Tue: 7% penalty (specifically 90% -> 83%)
• Exceptions only for exceptional circumstances.
• Everyone gets 72 free late hours, applied automatically
41
![Page 41: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/41.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Copying: There are better options
• Read the syllabus
• Copying is usually painfully obvious and I don’t have many options
• Submit it late (that’s why we have late days), half-working (that’s why we have partial credit), or take the zero on the homework – I guarantee you won’t care about one bad homework in a year
• If you’re overwhelmed, talk to us
42
![Page 42: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/42.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
Course Project
• Work in a team of 3-5 to do something cool
• There will be a piazza thread for pairing up
• Could be:• Applying vision to a problem you care about
• Independent re-implementation of a paper
• Trying to build and extend an approach
• Should be 2 homeworks worth of work per person
43
![Page 43: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/43.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
First Homework: Numpy warmup
First homework is meant as an introduction / refresher to array programming with numpy
Will be released tomorrow, due in one week on Wednesday 1/27 11:59pm
44
![Page 44: Computer Vision - University of Michigan](https://reader030.vdocuments.site/reader030/viewer/2022032402/6231d06af1162a4f1265ceec/html5/thumbnails/44.jpg)
Justin Johnson & David Fouhey January 19, 2021EECS 442 WI 2021: Lecture 1 -
And now to David:Camera Models
45