Scribd
Upload
12 views

Week 10

Uploaded by

yanwang2693
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
12 views

Week 10

Uploaded by

yanwang2693
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/ 31

Data Visualization

Angela Lu

Week 10
Information Systems, CB
City University of Hong Kong
Recap
• Network visualization
• Today: Computer vision

2
What is computer vision?
• A field of artificial intelligence (AI) that enables computers
and systems to derive meaningful information from digital
images, videos and other visual inputs — and take actions or
make recommendations based on that information
• The search of the fundamental visual features, maps a 2D image
into a vector or a point
• Applications of reconstruction and recognition
An Interdisciplinary Area

4
CV Applications
Facial recognition
Augmented reality
Autonomous driving
Deep fake
Artwork with GANs …
6
Visual Recognition Problems
• Image classification, such as
• Object/action detection
• Image captioning
• Semantic segmentation
• Visual question answering
• Visual instruction navigation
• Scene graph generation
•…

7
Object Detection

8
Action Detection

9
Image Captioning

10
Sematic Segmentation

11
Visual Question Answering

12
Scene Graph

13
GANs Artwork

14
Image Classification: A Core Task

cat

15
How the Computer Sees it?

What the computer sees

An image is a tensor of integers between [0, 255]:


e.g. 800 x 600 x 3 (3 channels RGB)

16
Other Challenges
• Viewpoint change
• Background
• Illumination
• Occlusion
• Deformation etc…
So how to classify?
17
K-nearest Neighbor Classifier
• K-Nearest Neighbors
• Like nearest
neighbors, but instead
of calculating the
single nearest
neighbor, it takes a
majority vote based
on multiple (K)
nearest neighbors

18
K-nearest Neighbor Classifier
cat

d(I1, I2)

chair
Min(Distance)?
d(I1, I3)

fish
???

Start with a training set of images and labels, and predict d(I1, I4)
labels on the test set 19
Distance Metric
• L1 (Manhattan) distance:

20
K-nearest Neighbor Classifier
• Predicts labels based on the K
nearest training examples,
based on distance metric and
K
• But how to choose K?
• E.g. cross-validation for the
value of k
• (Seems that k ~= 12 works best
for this data)

21
Linear Classifier

f(x, W) = Wx + b

W: weights
x: image pixels
b: vector of bias

A linear regression is a simple


case

22
Linear Classifier: Parameters

3072X1
3x1
f(x, W) = Wx + b 3 numbers giving
class scores
Array of 32x32x3 numbers
(3072 numbers) Weights, 3x3072

Classify to 3 classes:
cat, dog, panda

23
Linear Classifier: Example

b
W

x
24
Classification Results

Cat 0.51 1.01 0.05


Dog -1.09 2.05 0.20
Panda 0.02 0.05 1.54

25
Linear Classifier
• f(x, W) = Wx + b
• How to come up with a good W?
• Start with random W
• Define a loss function
W?
• Quantifies training data inaccuracy
• Find W that minimizes loss (optimization)

26
Loss Function
• SVM loss

• Softmax

• Regularization loss

27
Loss Function: SVM vs. Softmax

28
Optimization: Gradient Descent

29
Stochastic Gradient Descent

• Full sum expensive when N is large!


• Approximate sum using a minibatch of
examples (32/64/128 etc.)

30
Next time…
• Deep learning & neural networks

31

You might also like