Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Ensemble of Exemplar-SVMs for Object Detection and Beyond
Search
Yasser Souri
December 08, 2012
Programming
0
140
Ensemble of Exemplar-SVMs for Object Detection and Beyond
Yasser Souri
December 08, 2012
Tweet
Share
More Decks by Yasser Souri
See All by Yasser Souri
Intro to Variational AutoEncoder
yassersouri
0
59
Deep Learning Talk - Saverin
yassersouri
0
49
Deep Relative Attribute
yassersouri
1
51
Fine-grained Image Classification
yassersouri
1
70
Image Classification Intro
yassersouri
1
110
Real-time tracking of sports pitch markings
yassersouri
1
44
Other Decks in Programming
See All in Programming
if constexpr文はテンプレート世界のラムダ式である
faithandbrave
3
670
業務ツールとして使うPostman
msys75
0
100
見た目から始める生産性向上
ikumatadokoro
10
1.3k
Ruby Function Composition
bkuhlmann
1
340
Milestoner
bkuhlmann
1
410
Three ways to use AI on Android: The Good, the Bad and the Ugly
marxallski
0
110
Behind VS Code Extensions for JavaScript / TypeScript Linnting and Formatting
unvalley
6
1.2k
SIMD Parallel Programming with the Vector API
josepaumard
0
230
効率化に挑戦してみたらモバイル開発が少し快適になった話
ryunakayama
0
140
Java 22 Overview
kishida
1
190
Micro Frontends for Java Microservices - Utah JUG 2024
mraible
PRO
1
110
スキーマ駆動開発による品質とスピードの両立 - 私達は何故、スキーマを書くのか
kentaroutakeda
0
180
Featured
See All Featured
The Cult of Friendly URLs
andyhume
74
5.7k
How to train your dragon (web standard)
notwaldorf
75
5.2k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
660
120k
How to name files
jennybc
65
93k
Web Components: a chance to create the future
zenorocha
306
41k
RailsConf 2023
tenderlove
8
550
5 minutes of I Can Smell Your CMS
philhawksworth
199
19k
A Modern Web Designer's Workflow
chriscoyier
689
190k
Building Effective Engineering Teams - LeadDev
addyosmani
32
1.9k
YesSQL, Process and Tooling at Scale
rocio
165
13k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
188
16k
[RailsConf 2023 Opening Keynote] The Magic of Rails
eileencodes
13
8.3k
Transcript
Ensemble of Exemplar- SVMs for Object Detection and Beyond Tomasz
Malisiewicz, Abhinav Gupta and Alexei A. Efros ICCV, 2011
Abstract
Abstract • Object Detection
Abstract • Object Detection • On par with state of
the art
Abstract • Object Detection • On par with state of
the art • Much simpler
Abstract • Object Detection • On par with state of
the art • Much simpler • At only a modest computational cost
Abstract • Object Detection • On par with state of
the art • Much simpler • At only a modest computational cost • Central benefit: explicit association between each detection and one training example
Motivation
Motivation • Common Computer Vision tasks:
Motivation • Common Computer Vision tasks: • Image classification
Motivation • Common Computer Vision tasks: • Image classification •
Object detection
Motivation • Common Computer Vision tasks: • Image classification •
Object detection • bounding box
Motivation - Object Detection • Can we reason with bounding
box? BUS
Motivation - How can we reason?
Motivation - How can we reason? • Obtain Association with
a very similar exemplar from training
Motivation - How can we reason? • Obtain Association with
a very similar exemplar from training • This is what mind does
Motivation - How can we reason? • Obtain Association with
a very similar exemplar from training • This is what mind does • Enough data is currently available
Motivation - How can we reason? • Obtain Association with
a very similar exemplar from training • This is what mind does • Enough data is currently available • Any kind of meta data could be transferred
Exemplars
Motivation - Exemplar Theory
Motivation - Exemplar Theory • Associating a new instance with
something seen in the past
Motivation - Exemplar Theory • Associating a new instance with
something seen in the past • Exemplar theory in cognitive psychology
Motivation - Exemplar Theory • Associating a new instance with
something seen in the past • Exemplar theory in cognitive psychology • Case-based reasoning in AI
Motivation - Exemplar Theory • Associating a new instance with
something seen in the past • Exemplar theory in cognitive psychology • Case-based reasoning in AI • Instance-based learning in ML
Exemplar Reasoning is Non-parametric
Exemplar Reasoning is Non-parametric KNN: non-parametric
Exemplar Reasoning is Non-parametric KNN: non-parametric SVM: parametric
Exemplar Theory in Computer Vision
Exemplar Theory in Computer Vision • Object Alignment • Scene
Recognition • Image Parsing • Object Detection (not successful)
Non-parametric Object Detection
Non-parametric Object Detection • has not been competitive against discriminative
approaches
Non-parametric Object Detection • has not been competitive against discriminative
approaches • Why?
Non-parametric Object Detection • has not been competitive against discriminative
approaches • Why? • Massive Amount of Negative data
Non-parametric Object Detection • has not been competitive against discriminative
approaches • Why? • Massive Amount of Negative data • Classification vs Detection and KNN
Motivation - Negative Data
Motivation - Negative Data • Non-parametric methods are not suitable
Motivation - Negative Data • Non-parametric methods are not suitable
• Parametric methods handle large amount of negative data very well
Motivation - Negative Data • Non-parametric methods are not suitable
• Parametric methods handle large amount of negative data very well • HOG
Motivation - Negative Data • Non-parametric methods are not suitable
• Parametric methods handle large amount of negative data very well • HOG • DPM
Motivation - Negative Data • Non-parametric methods are not suitable
• Parametric methods handle large amount of negative data very well • HOG • DPM
Motivation - Negative Data
Motivation - Negative Data • SVM can handle negative data
parametrically
Motivation - Negative Data • SVM can handle negative data
parametrically • No negative data is stored (vs KNN)
Motivation - Negative Data • SVM can handle negative data
parametrically • No negative data is stored (vs KNN) • Used by HOG
Parametric Approach
Parametric Approach • Very good representation of negative data
Parametric Approach • Very good representation of negative data •
What about positive data?
Parametric Approach • Very good representation of negative data •
What about positive data? • implicit assumption that all positive examples are visually related
None
Parametric Approach • Very good representation of negative data •
What about positive data? • implicit assumption that all positive examples are visually related • results in over generalized models
Desirable Approach
Desirable Approach • All strengths of HOG/DPM
Desirable Approach • All strengths of HOG/DPM • powerful descriptor
Desirable Approach • All strengths of HOG/DPM • powerful descriptor
• discriminative framework
Desirable Approach • All strengths of HOG/DPM • powerful descriptor
• discriminative framework • handle massive amount of negatives
Desirable Approach • All strengths of HOG/DPM • powerful descriptor
• discriminative framework • handle massive amount of negatives • Not rigidly representing positives
Desirable Approach • All strengths of HOG/DPM • powerful descriptor
• discriminative framework • handle massive amount of negatives • Not rigidly representing positives • Good Association for meta-data transfer
Desirable Approach • All strengths of HOG/DPM • powerful descriptor
• discriminative framework • handle massive amount of negatives • Not rigidly representing positives • Good Association for meta-data transfer Parametric Negatives Non-parametric Positives
Exemplar-SVMs • Learn a model for each positive example •
HOG features • linear SVM classifier
Exemplar-SVMs • Learn a model for each positive example •
HOG features • linear SVM classifier
Exemplar-SVMs • Training • Single Positive example • Millions of
negative examples (sliding windows) - from images not containing any in-class instances
Large Scale Training • Use parallel Training on clusters
Exemplar-SVMs • Testing • Each sliding window is given to
all Exemplar-SVMs • Highest score is the detection
Qualitative Examples
None
None
None
None
None
None
None
Meta-Data Transfer
None
None
None
None
None
Thank You Any Questions?