Home
Subjects
Textbook solutions
Create
Study sets, textbooks, questions
Log in
Sign up
Upgrade to remove ads
Only $35.99/year
Lecture 9 - Discriminative Correlation Filters for Visual Tracking
STUDY
Flashcards
Learn
Write
Spell
Test
PLAY
Match
Gravity
Terms in this set (5)
What does DCF stand for?
Discriminative Correlation Filter
What are the basic steps of MOSSE?
### Preprocessing ###
1) Transform pixel-values by log function. Help with low contrast light
2) Normalize pixel-values (mean = 0, norm = 1)
3) Multiply by Cosine-window to reduce pixel values near edge to zero
### MOSSE-algorithm ###
4) Generate 8 small perturbations of the tracing window in the initial frame
5) Initialize first filter using the perturbations
6) Calculate correlation between filter and frame (input)
7) Find the peak correlation and create g_{i+1} as a 2D Gaussian, with mean in peak and with standard deviation of 2.
8) Calculate new filter H_{i+1} by:
H_{t} = A_{t}/B_{t}
A_{t} = η
G_{t} ∘ F_{t}
t}* + (1 - η)*A_{1}
B_{t} = ηF_{t} ∘ F_{t}
+ (1 - η)
η)*B_{t}
9) Redo step 6-8 until no more frames
What's the difference between MOSSE and Regularized MOSSE?
A penalization is added to the L2 norm (Ridge regression).
This leads to adding the penalizing constant λ to the denominator when calculating the new filter:
H_{t} = A_{t}/(λ + B_{t})
Explain Multi-channel MOSSE
Similar to MOSSE, but instead of just having one input (e.g. grey scale), we have multiple inputs, e.g. RGB (3 inputs) and grey scale, a total of 4 inputs.
Different correlations are then calculated from the different inputs. The correlations are then summed up element wise. The maximum correlation is then picked as the center of the next 2D gaussian filter.
List some suitable features as input to MOSSE
1) Histogram of oriented gradients (HOG)
2) Color names (CN)
- With or without HOG
3) RGB
4) Deep network features
- Both shallow and deep features
Sets with similar terms
GIS Final
50 terms
GIS Final Review
68 terms
GIS 2
27 terms
NR-318
55 terms
Sets found in the same folder
Lecture 1
14 terms
Lecture 2 - Feature Descriptors
43 terms
Lecture 3 - Convolutional Neural Networks
31 terms
Lecture 4 - Image classification with convolutiona…
30 terms
Other sets by this creator
Lecture 8
14 terms
Lecture 7
18 terms
Lecture 6
8 terms
Lecture 5
18 terms
Other Quizlet sets
Investment Final Exam
76 terms
Microbiology Chapter 18- Part 1
21 terms
Professional Helping Skills & Techniques CHS 204-…
19 terms
Patho Exam 1 Study Objectives
61 terms