Singular Vector Decomposition (SVD), SVM, HOG-Detector
|
00:00:00
|
Lecture 10: Sliding-Wnidow based Object Detection |
|
00:00:06
|
Course Outline |
|
00:01:13
|
Recap: Subspace Methods |
|
00:02:18
|
Recap: Obj. Detection by Distance TO Eigenspace |
|
00:04:59
|
Recap: Obj. Detection by Distance IN Eigenspace |
|
00:07:45
|
Recap: Eigenfaces |
|
00:09:27
|
Recap: Obj. Detection by Distance TO Eigenspace |
|
00:11:04
|
Important Footnote |
|
00:13:08
|
Singular Value Decomposition (SVD) |
|
00:15:25
|
Performing PCA with SVD |
|
00:18:24
|
SVD Properties |
|
00:21:29
|
Limitations (1) |
|
00:22:42
|
Limitations (2) |
|
00:24:01
|
Restrictions of PCA |
|
00:29:49
|
Topics of This Lecture |
|
00:31:11
|
Identification vs. Categorization (1) |
|
00:34:04
|
Identification vs. Categorization (2) |
|
00:35:09
|
Object Categorization - Potential Applications |
|
00:35:54
|
How many object categories are there? |
|
00:37:19
|
~10,000 to 30,000 |
|
00:38:42
|
Challenges: Robustness (1) |
|
00:39:45
|
Challenges: Robustness (2) |
|
00:40:38
|
Topics of This Lecture |
|
00:41:02
|
Detection via Classification: Main Idea (1) |
|
00:41:50
|
Detection via Classification: Main Idea (2) |
|
00:44:18
|
What is a Sliding Window Approach? |
|
00:45:46
|
Detection via Classification: Main Idea (3) |
|
00:47:22
|
Feature extraction: Global Appearance |
|
00:50:11
|
Eigenfaces: Global Appearance Description |
|
00:50:41
|
Feature Extraction: Global Appearance |
|
00:52:34
|
Gradient-based Representations |
|
00:53:36
|
Gradient-based Representations: histogram |
|
00:56:25
|
Gradient-based Representations: Histograms of Oriented Gradients (HoG) |
|
00:56:57
|
Classifier Construction |
|
00:58:24
|
Discriminative Methods |
|
00:59:10
|
Classifier Construction: Many Choices... |
|
01:00:39
|
Linear Classifiers (1) |
|
01:02:56
|
Linear Classifiers (2) |
|
01:05:54
|
Support Vector Machines (SVMs) |
|
01:07:00
|
Support Vector Machines: margin |
|
01:11:46
|
Finding the Maximum Margin Line |
|
01:14:18
|
Questions |
|
01:14:45
|
Non-Linear SVMs: Feature Spaces |
|
01:18:38
|
Nonlinear SVMs |
|
01:19:10
|
Some Often-Used Kernel Functions |
|
01:19:59
|
Questions |
|
01:20:02
|
Multi-Class SVMs |
|
01:20:05
|
SVMs for Recognition |
|
01:21:35
|
Pedestrian Detection |
|
01:21:50
|
HOG Descriptor Processing Chain |
|
01:22:13
|
HOG Descriptor Processing Chain: Gamma compression |
|
01:22:46
|
HOG Descriptor Processing Chain: Gradient computation |
|
01:23:17
|
HOG Descriptor Processing Chain: Spatial/Orientation binning |
|
01:23:47
|
HOG Cell Computation Details |
|
01:25:21
|
HOG Cell Computation Details (2) |
|
01:27:05
|
HOG Descriptor Processing Chain: 2-Stage contrast normalization |
|
01:28:05
|
HOG Descriptor Processing Chain: Feature vector construction |
|
01:28:16
|
HOG Descriptor Processing Chain: SVM Classification |
|
01:28:53
|
Pedestrian Detection with HOG |
|
01:29:30
|
Non-Maximum Suppression |
|
01:30:22
|
Pedestrian detection with HoGs & SVMs |
|
01:31:44
|
References and Further Reading |