0.25
0.5
0.75
1.25
1.5
1.75
2
Trainable visual models for object classification
Published on Feb 25, 200715424 Views
The general theme of the tutorial will be 'trainable visual models for object classification'. I will cover: the difficulty of the problem a few approaches Perona and Welling Pictorial structures of F
Related categories
Chapter list
Trainable visual models for object class recognition00:01
Objectives00:32
Recognition00:54
(Semi) Unsupervised learning01:36
Some object classes02:11
Class of model: Pictorial Structure03:30
Representation: Parts and Structure04:20
Deformations05:08
Presence / Absence of Features05:24
Main issues:05:39
Outline06:36
1. Models that learn parts, then add structure07:22
Learning parts by clustering - 107:31
Learning parts by clustering - 208:19
Learning parts by clustering - 308:59
Detecting part positions09:03
Leibe & Schiele 2003/200409:57
Visual Vocabulary (Codebook Entries)10:53
Structure: Generalized Hough Transform12:41
Probabilistic Formulation13:45
Object Categorization Procedure14:08
Detection Results16:43
Agarwal & Roth 200216:55
Learning: Structure17:22
Recognition19:14
Object Categorization Procedure23:04
Recognition23:38
Borenstein & Ullman 200223:44
Class-based Recognition/Segmentation24:53
Structure: jigsaw puzzle approach25:13
Summary26:59
So far …..27:35
Search over scale28:11
2. Models for which the structure model is primary29:03
New ideas29:15
Pictorial Structure Models for Object Recognition29:27
Goal29:36
Matching Pictorial Structures29:56
Example: Generic Person Model30:09
Learning31:01
Recognition31:19
Example: Recognizing People32:22
Variety of Poses32:53
Variety of Poses32:59
Pictorial structures for tracking34:20
Learning articulated pictorial structures using temporal coherence35:16
Results36:06
3. Models that learn parts and structure simultaneously37:02
New ideas37:25
Detect region for candidate parts38:04
Representation of regions38:57
Generative probabilistic model39:50
Example – Learnt Motorbike Model41:26
Learning42:44
Learning procedure43:22
Recognition45:57
Experiments46:19
Experimental procedure46:21
Recognized Motorbikes 47:09
Background images evaluated with motorbike model47:20
Frontal faces47:52
Airplanes48:07
Spotted cats48:24
Sampling from models48:50
Comparison to other methods49:10
“Brain damaged” Constellation model50:05
Constellation Model Generalization 1: Conditionally independent model53:11
Shape model53:23
Spotted Cats58:01
Spotted Cats58:27
Constellation Model Generalization 2: Heterogeneous parts58:41
Variety of feature types59:00
Airplanes – Kadir & Brady operator59:50
Airplanes – Curves59:59
Airplanes – multi-scale Harris operator01:00:10
Fitting the heterogeneous model01:00:27
Motorbikes01:00:53
Motorbike Patch and Curve model01:01:09
Motorbike results using curve and patch model01:01:17
Spotted cats01:01:31
Spotted cats combination model01:01:45
Spotted cats results using combination model01:01:47
4. Summary and open challenges01:01:49
Single/Multiple visual aspects01:01:58
Open Research Areas01:03:20
Pascal Challenge: 101 Object Classes01:04:38
Learning from contaminated data01:06:30
Learning from contaminated data01:06:35
Learning from contaminated data01:07:19
Robust line estimation - RANSAC01:07:58
RANSAC robust line estimation01:08:23
RANSAC Scoring Function01:08:31
Camel curve model01:08:39
Camel curve model01:08:40
Raw Camel images & 10 picked01:08:41
Camel RPC curves01:08:48
Camel filtered results01:08:49
Raw Tiger images01:09:01
Tiger filtered results01:09:12
Tiger RPC curve01:09:15
Raw Bottles images01:09:15
Bottles filtered results01:09:23