Lecture 15 - Stanford Vision Lab

More documents

Recommendations

Info

Recognition 1. Run part detectors exhaustively over image 1 1 2 3 1 2 2 1 2 3 4 3 h 0N 0N 0N 0N 1 2 3 4 e.g. h 2 3 0 2 2. Try different combinations of detections in model - Allow detections to be missing (occlusion) 3. Pick hypothesis which maximizes: p( Data | Object, Hyp) p( Data | Clutter, Hyp) 4. If ratio is above threshold then, instance detected Fei-Fei Li Lecture 15 - 66 14‐Nov‐11
So far….. • Representation – Joint model of part locations – Ability to deal with background clutter and occlusions • Learning – Manual construction of part detectors – Estimate parameters of shape density • Recognition – Run part detectors over image – Try combinations of features in model – Use efficient search techniques to make fast Fei-Fei Li Lecture 15 - 67 14‐Nov‐11
Page 1 and 2:
Lecture 15: Object recognition: Bag
Page 3 and 4:
What we will learn today? • Bag o
Page 5 and 6:
Related works • Early “bag of w
Page 7 and 8:
Analogy to documents Of all the sen
Page 9 and 10:
definition of “BoW” - Independe
Page 11 and 12:
1.Feature detection and representat
Page 13 and 14:
1.Feature detection and representat
Page 15 and 16: 1.Feature detection and representat
Page 17 and 18: 2. Codewords dictionary formation
Page 19 and 20: 2. Codewords dictionary formation F
Page 21 and 22: Visual vocabularies: Issues • How
Page 23 and 24: 3. Bag of word representation frequ
Page 25 and 26: Learning and Recognition codewords
Page 27 and 28: Discriminative classifiers category
Page 29 and 30: Nearest Neighbors classifier Query
Page 31 and 32: K- Nearest Neighbors classifier fro
Page 33 and 34: Learning and Recognition 1. Discrim
Page 35 and 36: Support vector machines • Find hy
Page 37 and 38: Nonlinear SVMs •Datasets that are
Page 39 and 40: Nonlinear SVMs • Nonlinear decisi
Page 41 and 42: Pyramid match kernel • Fast appro
Page 43 and 44: What about multi‐class SVMs? •
Page 45 and 46: Object recognition results • ETH
Page 47 and 48: Learning and Recognition 1. Discrim
Page 49 and 50: Some notations • w: a collection
Page 51 and 52: the Naïve Bayes model c w N Graphi
Page 53 and 54: Csurka et al. 2004 Fei-Fei Li Lectu
Page 55 and 56: Generative vs discriminative • Di
Page 57 and 58: What we will learn today? • Bag o
Page 59 and 60: Parts and Structure Literature •
Page 61 and 62: Deformations A B C D Fei-Fei Li Lec
Page 63 and 64: Background clutter Fei-Fei Li Lectu
Page 65: Learning Models `Manually’ • Ob
Page 69 and 70: (Semi) Unsupervised learning •Kno
Page 71 and 72: Unsupervised detector training - 2
Page 73 and 74: Learning • Take training images.
Page 75 and 76: Detector Selection •Try out diffe
Page 77 and 78: Learned face model Pre-selected Par
Page 79 and 80: Background images Fei-Fei Li Lectur
Page 81 and 82: Detections of Cars Fei-Fei Li Lectu
Page 83 and 84: 3D Object recognition - Multiple mi
Page 85 and 86: So far (2)….. • Representation
Page 87 and 88: Object categorization Fergus et. al
Page 89 and 90: •Kadir & Brady saliency region de
Page 91 and 92: Motorbikes Samples from appearance
Page 93 and 94: Background images evaluated with mo
Page 95 and 96: Airplanes Fei-Fei Li Lecture 15 - 9
Page 97 and 98: Summary of results Dataset Fixed sc
Page 99 and 100: Why this design? • Generic featur
Page 101 and 102: Supplementary materials • One‐S
Page 103 and 104: Fei-Fei Li Lecture 15 - 103 14‐No
Page 105 and 106: Algorithm Burl, et al. Weber, et al
Page 107 and 108: How do we do better than what stati
Page 109 and 110: Bayesian framework P(object | test,
Page 111 and 112: Bayesian framework P(object | test,
Page 113 and 114: model () space Model Structure Each
Page 115 and 116: Variational EM Random initializatio
Page 117 and 118:
Faces Motorbikes Airplanes Spotted
Page 119 and 120:
Experiments: obtaining priors airpl
Page 121 and 122:
Number of training examples Fei-Fei
Page 123 and 124:
Number of training examples Fei-Fei
show all

Lecture 15 - Stanford Vision Lab

Create successful ePaper yourself

Delete template?

Save as template?