Pattern Recognition and Machine Learning in Computer Vision Workshop

Trainable visual models for object classification

author: Andrew Zisserman, University of Oxford

Description

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 Felzenszwalb and Huttenlocher Borenstein and Ullman Agarwal and Roth Leibe and Schiele covering the method, invariance, data preparation and then go into detail on the constellation model end with research challenges

Categories

Top: Computer Science: Image Analysis

You might be experiencing some problems with Your Video player.

*Slides*
0:01	Trainable visual models for object class recognition
0:32	Objectives
0:54	Recognition
1:36	(Semi) Unsupervised learning
2:11	Some object classes
3:30	Class of model: Pictorial Structure
4:20	Representation: Parts and Structure
5:08	Deformations
5:24	Presence / Absence of Features
5:39	Main issues:
6:36	Outline
7:22	1. Models that learn parts, then add structure
7:31	Learning parts by clustering - 1
8:19	Learning parts by clustering - 2
8:59	Learning parts by clustering - 3
9:03	Detecting part positions
9:57	Leibe & Schiele 2003/2004
10:53	Visual Vocabulary (Codebook Entries)
12:41	Structure: Generalized Hough Transform
13:45	Probabilistic Formulation
14:08	Object Categorization Procedure
16:43	Detection Results
16:55	Agarwal & Roth 2002
17:22	Learning: Structure
19:14	Recognition
23:04	Object Categorization Procedure
23:38	Recognition
23:44	Borenstein & Ullman 2002
24:53	Class-based Recognition/Segmentation
25:13	Structure: jigsaw puzzle approach
26:59	Summary
27:35	So far …..
28:11	Search over scale
29:03	2. Models for which the structure model is primary
29:15	New ideas
29:27	Pictorial Structure Models for Object Recognition
29:36	Goal
29:56	Matching Pictorial Structures
30:09	Example: Generic Person Model
31:01	Learning
31:19	Recognition
32:22	Example: Recognizing People
32:53	Variety of Poses
32:59	Variety of Poses
34:20	Pictorial structures for tracking
35:16	Learning articulated pictorial structures using temporal coherence
36:06	Results
37:02	3. Models that learn parts and structure simultaneously
37:25	New ideas
38:04	Detect region for candidate parts
38:57	Representation of regions
39:50	Generative probabilistic model
41:26	Example – Learnt Motorbike Model
42:44	Learning
43:22	Learning procedure
45:57	Recognition
46:19	Experiments
46:21	Experimental procedure
47:09	Recognized Motorbikes
47:20	Background images evaluated with motorbike model
47:52	Frontal faces
48:07	Airplanes
48:24	Spotted cats
48:50	Sampling from models
49:10	Comparison to other methods
50:05	“Brain damaged” Constellation model
53:11	Constellation Model Generalization 1: Conditionally independent model
53:23	Shape model
58:01	Spotted Cats
58:27	Spotted Cats
58:41	Constellation Model Generalization 2: Heterogeneous parts
59:00	Variety of feature types
59:50	Airplanes – Kadir & Brady operator
59:59	Airplanes – Curves
60:10	Airplanes – multi-scale Harris operator
60:27	Fitting the heterogeneous model
60:53	Motorbikes
61:09	Motorbike Patch and Curve model
61:17	Motorbike results using curve and patch model
61:31	Spotted cats
61:45	Spotted cats combination model
61:47	Spotted cats results using combination model
61:49	4. Summary and open challenges
61:58	Single/Multiple visual aspects
63:20	Open Research Areas
64:38	Pascal Challenge: 101 Object Classes
66:30	Learning from contaminated data
66:35	Learning from contaminated data
67:19	Learning from contaminated data
67:58	Robust line estimation - RANSAC
68:23	RANSAC robust line estimation
68:31	RANSAC Scoring Function
68:39	Camel curve model
68:40	Camel curve model
68:41	Raw Camel images & 10 picked
68:48	Camel RPC curves
68:49	Camel filtered results
69:01	Raw Tiger images
69:12	Tiger filtered results
69:15	Tiger RPC curve
69:15	Raw Bottles images
69:23	Bottles filtered results

Lecture rating

People found this lecture:
Worth seeing
because it is:
Valuable and informative
Well presented
Easily understandable
Acceptably recorded
You need to login to cast your vote.

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.