Learning generative visual models from few training examples: An incremental Bayesian approach tested on 101 object categories

Li Fei-Fei, Robert Fergus, Pietro Perona

Research output: Contribution to journalArticle

Abstract

Current computational approaches to learning visual object categories require thousands of training images, are slow, cannot learn in an incremental manner and cannot incorporate prior information into the learning process. In addition, no algorithm presented in the literature has been tested on more than a handful of object categories. We present an method for learning object categories from just a few training images. It is quick and it uses prior information in a principled way. We test it on a dataset composed of images of objects belonging to 101 widely varied categories. Our proposed method is based on making use of prior information, assembled from (unrelated) object categories which were previously learnt. A generative probabilistic model is used, which represents the shape and appearance of a constellation of features belonging to the object. The parameters of the model are learnt incrementally in a Bayesian manner. Our incremental algorithm is compared experimentally to an earlier batch Bayesian algorithm, as well as to one based on maximum likelihood. The incremental and batch versions have comparable classification performance on small training sets, but incremental learning is significantly faster, making real-time learning feasible. Both Bayesian methods outperform maximum likelihood on small training sets.

Original languageEnglish (US)
Pages (from-to)59-70
Number of pages12
JournalComputer Vision and Image Understanding
Volume106
Issue number1
DOIs
StatePublished - Apr 2007

Fingerprint

Maximum likelihood
Information use
Statistical Models

Keywords

  • Bayesian model
  • Categorization
  • Generative model
  • Incremental learning
  • Object recognition

ASJC Scopus subject areas

  • Computer Vision and Pattern Recognition
  • Signal Processing
  • Electrical and Electronic Engineering

Cite this

Learning generative visual models from few training examples : An incremental Bayesian approach tested on 101 object categories. / Fei-Fei, Li; Fergus, Robert; Perona, Pietro.

In: Computer Vision and Image Understanding, Vol. 106, No. 1, 04.2007, p. 59-70.

Research output: Contribution to journalArticle

@article{f27f85ef516241f39ed3a2489f8148a5,
title = "Learning generative visual models from few training examples: An incremental Bayesian approach tested on 101 object categories",
abstract = "Current computational approaches to learning visual object categories require thousands of training images, are slow, cannot learn in an incremental manner and cannot incorporate prior information into the learning process. In addition, no algorithm presented in the literature has been tested on more than a handful of object categories. We present an method for learning object categories from just a few training images. It is quick and it uses prior information in a principled way. We test it on a dataset composed of images of objects belonging to 101 widely varied categories. Our proposed method is based on making use of prior information, assembled from (unrelated) object categories which were previously learnt. A generative probabilistic model is used, which represents the shape and appearance of a constellation of features belonging to the object. The parameters of the model are learnt incrementally in a Bayesian manner. Our incremental algorithm is compared experimentally to an earlier batch Bayesian algorithm, as well as to one based on maximum likelihood. The incremental and batch versions have comparable classification performance on small training sets, but incremental learning is significantly faster, making real-time learning feasible. Both Bayesian methods outperform maximum likelihood on small training sets.",
keywords = "Bayesian model, Categorization, Generative model, Incremental learning, Object recognition",
author = "Li Fei-Fei and Robert Fergus and Pietro Perona",
year = "2007",
month = "4",
doi = "10.1016/j.cviu.2005.09.012",
language = "English (US)",
volume = "106",
pages = "59--70",
journal = "Computer Vision and Image Understanding",
issn = "1077-3142",
publisher = "Academic Press Inc.",
number = "1",

}

TY - JOUR

T1 - Learning generative visual models from few training examples

T2 - An incremental Bayesian approach tested on 101 object categories

AU - Fei-Fei, Li

AU - Fergus, Robert

AU - Perona, Pietro

PY - 2007/4

Y1 - 2007/4

N2 - Current computational approaches to learning visual object categories require thousands of training images, are slow, cannot learn in an incremental manner and cannot incorporate prior information into the learning process. In addition, no algorithm presented in the literature has been tested on more than a handful of object categories. We present an method for learning object categories from just a few training images. It is quick and it uses prior information in a principled way. We test it on a dataset composed of images of objects belonging to 101 widely varied categories. Our proposed method is based on making use of prior information, assembled from (unrelated) object categories which were previously learnt. A generative probabilistic model is used, which represents the shape and appearance of a constellation of features belonging to the object. The parameters of the model are learnt incrementally in a Bayesian manner. Our incremental algorithm is compared experimentally to an earlier batch Bayesian algorithm, as well as to one based on maximum likelihood. The incremental and batch versions have comparable classification performance on small training sets, but incremental learning is significantly faster, making real-time learning feasible. Both Bayesian methods outperform maximum likelihood on small training sets.

AB - Current computational approaches to learning visual object categories require thousands of training images, are slow, cannot learn in an incremental manner and cannot incorporate prior information into the learning process. In addition, no algorithm presented in the literature has been tested on more than a handful of object categories. We present an method for learning object categories from just a few training images. It is quick and it uses prior information in a principled way. We test it on a dataset composed of images of objects belonging to 101 widely varied categories. Our proposed method is based on making use of prior information, assembled from (unrelated) object categories which were previously learnt. A generative probabilistic model is used, which represents the shape and appearance of a constellation of features belonging to the object. The parameters of the model are learnt incrementally in a Bayesian manner. Our incremental algorithm is compared experimentally to an earlier batch Bayesian algorithm, as well as to one based on maximum likelihood. The incremental and batch versions have comparable classification performance on small training sets, but incremental learning is significantly faster, making real-time learning feasible. Both Bayesian methods outperform maximum likelihood on small training sets.

KW - Bayesian model

KW - Categorization

KW - Generative model

KW - Incremental learning

KW - Object recognition

UR - http://www.scopus.com/inward/record.url?scp=34047174674&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=34047174674&partnerID=8YFLogxK

U2 - 10.1016/j.cviu.2005.09.012

DO - 10.1016/j.cviu.2005.09.012

M3 - Article

AN - SCOPUS:34047174674

VL - 106

SP - 59

EP - 70

JO - Computer Vision and Image Understanding

JF - Computer Vision and Image Understanding

SN - 1077-3142

IS - 1

ER -