6.819 / 6.869: advances in computer vision6.869.csail.mit.edu/fa15/lecture/6.869-l11-intro... ·...
TRANSCRIPT
![Page 1: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/1.jpg)
6.819 / 6.869: Advances in Computer Vision
Learning:Introduction to Machine Learning for Vision
Website:
http://6.869.csail.mit.edu/fa15/
Instructor: Yusuf Aytar
Lecture TR 9:30AM – 11:00AM (Room 34-101)
![Page 2: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/2.jpg)
Key Concepts
Pattern, Category/Class, Instance, Generalization, Classification
Feature space, Objective function, Regularization, Loss Function, OptimizationClassification
Concepts
Support Vector Machines
![Page 3: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/3.jpg)
Where is the bottle ?
![Page 4: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/4.jpg)
4
Where is Waldo ?
![Page 5: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/5.jpg)
Task: find the most similar patch in a second image
Find the same patch
![Page 6: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/6.jpg)
?
![Page 7: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/7.jpg)
?
![Page 8: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/8.jpg)
Computers are good with patterns and
We are good with categories
Pattern vs Category
… but computers are also getting better with categories
![Page 9: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/9.jpg)
Pattern Category
Patterns have discriminative representations
with less variation
Categories also have discriminative representations,
but with great variations
Oliva & Torralba (2007) TICS
![Page 10: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/10.jpg)
Find a bottle:CategoriesInstances Find these two objects
Instance vs Category
![Page 11: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/11.jpg)
Generalization: Extracting the essence of a concept based on its analysis of similarities from many discrete objects.
http://en.wikipedia.org/wiki/Generalization
Generalization
instances
category
![Page 12: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/12.jpg)
Challenges of Generalization
A successful object category detector should be invariant to changes in illumination, occlusion, background clutter, scale,
viewpoint, deformation and intra-class variance.
![Page 13: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/13.jpg)
Object Instance Detection
Find the Object
illumination, occlusion, background clutter, scale, viewpoint, deformation and intra-class variance.
Which of the invariances below apply for the given object instance detection problem?
![Page 14: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/14.jpg)
Classification vs. Detection
Where is the …? Localize the object.Is this a … image ?
TableKitchen HorseWaldo Car
Detection can be performed through a classifier, i.e. sliding window search
![Page 15: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/15.jpg)
Every training sample is represented as a point in the feature space
Feature Space
green
red+
blue
![Page 16: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/16.jpg)
SIFT: Scale-Invariant Feature Transform
(Lowe, 1999)
HOG: Histograms of oriented gradients(Dalal & Triggs CVPR 05)
Gist: Grid of gabors
(Oliva & Torralba, 2001)
Example Feature Spaces
![Page 17: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/17.jpg)
Machine Learning Methods
![Page 18: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/18.jpg)
Generative vs Discriminative
• Represent both the data and the labels
• Often, makes use of conditional independence and priors
• Examples• Naïve Bayes classifier• Bayesian network
• Models of data may apply to future prediction problems
• Learn to directly predict the labels from the data
• Often, assume a simple boundary (e.g., linear)
• Examples• Logistic regression• SVM• Boosted decision trees
• Often easier to predict a label from the data than to model the data
![Page 19: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/19.jpg)
Discriminative Models
Source: Vittorio Ferrari, Kristen Grauman, Antonio Torralba, Fei-Fei Li
![Page 20: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/20.jpg)
• Apply a prediction function to a feature representation of the image to get the desired output:
f( ) = “apple”f( ) = “tomato”f( ) = “cow”
Slide credit: L. Lazebnik
Classification
![Page 21: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/21.jpg)
y = f(x)
• Training: given a training set of labeled examples {(x1,y1), …, (xN,yN)}, estimate the prediction function f by minimizing the prediction error on the training set
• Testing: apply f to a never before seen test example x and output the predicted value y = f(x)
output prediction function image feature
Slide credit: L. Lazebnik
Classification Formulation
![Page 22: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/22.jpg)
Prediction
Training
Image Features
Testing
Test Image
Learned model
Slide credit: D. Hoiem and L. Lazebnik
Training LabelsTraining
Images
TrainingImage Features
Learned model
Learning Framework
![Page 23: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/23.jpg)
f(x) = label of the training example nearest to x in the feature space
• All we need is a distance function for our inputs • No training required!
Test example
Training examples
from class 1
Training examples
from class 2
Slide credit: L. Lazebnik
Nearest neighbor classification
![Page 24: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/24.jpg)
K-Nearest neighbor classification
Test example
![Page 25: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/25.jpg)
3-Nearest neighbor classification
Test example
![Page 26: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/26.jpg)
5-Nearest neighbor classification
Test example
Simple, a good one to try first
![Page 27: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/27.jpg)
Binary classification can be viewed as the task of separating classes in feature space:
wTx + b = 0
wTx + b < 0wTx + b > 0
f(x) = sign(wTx + b)
Linear Classifiers: Perceptron
![Page 28: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/28.jpg)
Linear Classifiers
A B C
Which one is a better classifier?
f(xi) = sign( )
![Page 29: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/29.jpg)
Which one is a better classifier?
Linear Classifiersf(xi) = sign( )
![Page 30: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/30.jpg)
Support Vector Machines (Intuition)
![Page 31: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/31.jpg)
■ The linear discriminant function (classifier) with the maximum margin is the best
■ Margin is defined as the width that the boundary could be increased by before hitting a data point
■ Why it is the best? • Robust to outliners and thus
strong generalization ability
“safe zone”Margin
x1
x2
denotes +1
denotes -1
Support Vector Machines
Slide Credit: Jinwei Gu
![Page 32: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/32.jpg)
■ Given a set of data points:
■ With a scale transformation on both w and b, the above is equivalent to
For 1, 0
For 1, 0
Ti i
Ti i
y by b= + + >
= − + <
w x
w x
, where
For 1, 1
For 1, 1
Ti i
Ti i
y by b= + + ≥
= − + ≤ −
w x
w x
“safe zone”Margin
x1
x2
denotes +1
denotes -1{( , )}, 1, 2, ,i iy i n=x !…
Support Vector Machines
Slide Credit: Jinwei Gu
![Page 33: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/33.jpg)
Loss Function (Hinge Loss)Regularizer
Support Vector Machines
![Page 34: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/34.jpg)
Loss Function (Hinge Loss)Regularizer
Support Vector Machines
Objective Function
y = sign( )Prediction Function Learning: Convex Optimization
gradient
objective (energy)
![Page 35: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/35.jpg)
• Datasets that are linearly separable work out great:
0 x
Slide credit: Andrew Moore
Non-Linear SVMs
0 x
• But what if the dataset is just too hard?
0 x
x2
• We can map it to a higher-dimensional space:
![Page 36: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/36.jpg)
Φ: x → φ(x)
• General idea: the original input space can always be mapped to some higher-dimensional feature space where the training set is separable:
Slide credit: Andrew Moore
Non-Linear SVMs
![Page 37: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/37.jpg)
• Consider the mapping ),()( 2xxx =ϕ
22
2222
),(),(),()()(yxxyyxK
yxxyyyxxyx
+=
+=⋅=⋅ϕϕ
x2
Non-Linear Kernel: Example
![Page 38: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/38.jpg)
• The kernel trick: instead of explicitly computing the lifting transformation φ(x), define a kernel function K such that K(xi , xj) = φ(xi ) · φ(xj)
• This gives a nonlinear decision boundary in the original feature space:
bKybyi
iiii
iii +=+⋅ ∑∑ ),()()( xxxx αϕϕα
C. Burges, A Tutorial on Support Vector Machines for Pattern Recognition, Data Mining and Knowledge Discovery, 1998
Non-Linear SVMs
![Page 39: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/39.jpg)
❑ Linear kernel:
2
2( , ) exp( )2i j
i jKσ
−= −
x xx x
( , ) Ti j i jK =x x x x
( , ) (1 )T pi j i jK = +x x x x
0 1( , ) tanh( )Ti j i jK β β= +x x x x
❑ Polynomial kernel:
❑ Gaussian (Radial-Basis Function (RBF) ) kernel:
❑ Sigmoid:
■ In general, functions that satisfy Mercer’s condition can be kernel functions.
Common Kernel Functions
![Page 40: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/40.jpg)
1. Pick an image representation (HoG, SIFT+BOW, etc.)2. Pick a kernel function for that representation3. Compute the matrix of kernel values between every pair of
training examples4. Feed the kernel matrix into your favorite SVM solver to obtain
support vectors and weights5. At test time: compute kernel values for your test example and
each support vector, and combine them with the learned weights to get the value of the decision function
Slide credit: L. Lazebnik
Summary: SVMs for image classification
![Page 41: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/41.jpg)
http://www.mathworks.com/help/stats/svmtrain.html
MATLAB SVM Example
http://www.vlfeat.org/overview/svm.html
![Page 42: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/42.jpg)
• Unfortunately, there is no “definitive” multi-class SVM formulation
• In practice, we have to obtain a multi-class SVM by combining multiple two-class SVMs
• One vs. others• Training: learn an SVM for each class vs. the others• Testing: apply each SVM to test example and assign to it the
class of the SVM that returns the highest decision value
• One vs. one• Training: learn an SVM for each pair of classes• Testing: each learned SVM “votes” for a class to assign to the
test example
Slide credit: L. Lazebnik
What about multi-class SVMs?
![Page 43: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/43.jpg)
• Pros• Many publicly available SVM packages (LibSVM, Liblinear,etc):
http://www.kernel-machines.org/software• Kernel-based framework is very powerful, flexible• SVMs work very well in practice, even with very small training
sample sizes
• Cons• No “direct” multi-class SVM, must combine two-class SVMs• Computation, memory
– During training time, must compute matrix of kernel values for every pair of examples
– Learning can take a very long time for large-scale problems
SVMs: Pros and cons
![Page 44: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/44.jpg)
• How well does a learned model generalize from the data it was trained on to a new test set?
Training set (labels known) Test set (labels unknown)
Slide credit: L. Lazebnik
Generalization
![Page 45: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/45.jpg)
Slide credit: D. Hoiem
Underfitting: model is too “simple” to represent all the relevant class characteristics samples
simple model
preferred model
Overfitting: model is too “complex” and fits irrelevant characteristics (noise) in the data
samples
complex model
preferred model
Overfitting vs Underfitting“Everything should be made as simple as possible, but not simpler.”
Albert Einstein
![Page 46: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/46.jpg)
Traditional Detector Training (Motorbike)
Positive Samples
Negative Samples
Positive HOGs
Negative HOGs
Motorbike Detector
[Dalal et al. CVPR’05][Felzenszwalb et al. CVPR’08]
Training Samples
Feature Extraction (Histogram of
Oriented Gradients)
HOG Features
Linear SVM
Use Case: Linear SVMs over HoG
![Page 47: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/47.jpg)
Linear SVMs over
HoG features
[Dalal &Triggs’05], [Felzenszwalb’08] [Malisiewicz’11]
Exemplar SVM
Training an SVM with a single positive and many negative samples
Use Case: Exemplar SVMs
![Page 48: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/48.jpg)
Another Classifier: Randomized Decision Forests
Shotton et.al., Real-Time Human Pose Recognition in Parts from Single Depth Images, CVPR, 2011
![Page 49: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/49.jpg)
Body Part Classification with Randomized Decision Forests
Shotton et.al., Real-Time Human Pose Recognition in Parts from Single Depth Images, CVPR, 2011
![Page 50: 6.819 / 6.869: Advances in Computer Vision6.869.csail.mit.edu/fa15/lecture/6.869-L11-Intro... · Introduction to Machine Learning for Vision ... Another Classifier: Deep Learning](https://reader033.vdocument.in/reader033/viewer/2022042312/5edace3c3ee6fd189f66e5f8/html5/thumbnails/50.jpg)
Another Classifier: Deep Learning
Next Lecture …