DOMAIN ADAPTATION FOR IMAGE CLASSIFICATION WITH CLASS PRIORS
First Claim
1. A labeling system comprising:
- an electronic data processing device configured to label an image to be labeled belonging to a target domain by operations including;
training a boost classifier ƒ
T(x)=Σ
r=1Mβ
rhr(x) to classify an image belonging to the target domain and represented by a feature vector x, the training using a target domain training set DT comprising labeled feature vectors representing images belonging to the target domain and a plurality of source domain training sets DS1, . . . , DSN where N≧
2 comprising labeled feature vectors representing images belonging to source domains S1, . . . , SN respectively, the training comprising applying an adaptive boosting (AdaBoost) algorithm to generate the base classifiers hr(x) and the base classifier weights β
r of the boost classifier ƒ
T(x), wherein the rth iteration of the AdaBoost algorithm includes (i) performing N sub-iterations in which the kth sub-iteration trains a candidate base classifier hrk(x) on a training set combining the target domain training set DT and the source domain training set DSk and (ii) selecting hr(x) as the candidate base classifier with lowest error for the target domain training set DT;
computing a feature vector xin representing the image to be labeled; and
generating a label for the image to be labeled by operations including evaluating ƒ
T(xin)=Σ
r=1Mβ
rhr(xin).
7 Assignments
0 Petitions
Accused Products
Abstract
In camera-based object labeling, boost classifier ƒT(x)=Σr=1Mβrhr(x) is trained to classify an image represented by feature vector x using a target domain training set DT of labeled feature vectors representing images acquired by the same camera and a plurality of source domain training sets DS
-
Citations
24 Claims
-
1. A labeling system comprising:
-
an electronic data processing device configured to label an image to be labeled belonging to a target domain by operations including; training a boost classifier ƒ
T(x)=Σ
r=1Mβ
rhr(x) to classify an image belonging to the target domain and represented by a feature vector x, the training using a target domain training set DT comprising labeled feature vectors representing images belonging to the target domain and a plurality of source domain training sets DS1 , . . . , DSN where N≧
2 comprising labeled feature vectors representing images belonging to source domains S1, . . . , SN respectively, the training comprising applying an adaptive boosting (AdaBoost) algorithm to generate the base classifiers hr(x) and the base classifier weights β
r of the boost classifier ƒ
T(x), wherein the rth iteration of the AdaBoost algorithm includes (i) performing N sub-iterations in which the kth sub-iteration trains a candidate base classifier hrk(x) on a training set combining the target domain training set DT and the source domain training set DSk and (ii) selecting hr(x) as the candidate base classifier with lowest error for the target domain training set DT;computing a feature vector xin representing the image to be labeled; and generating a label for the image to be labeled by operations including evaluating ƒ
T(xin)=Σ
r=1Mβ
rhr(xin). - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9)
-
-
10. A labeling method for labeling an image to be labeled belonging to a target domain, the image labeling method comprising:
-
computing feature vectors representing target domain training images belonging to the target domain; labeling the target domain training images using labels selected from a set of labels to generate a target domain training set DT comprising labeled feature vectors representing the target domain training images; receiving a plurality of source domain training sets DS i , . . . , DSN where N≧
1 comprising feature vectors representing images belonging to source domains different from the target domain that are labeled using labels selected from the set of labels;performing unsupervised source-target domain alignment to align the target domain training set DT and the source training sets DS k , k=1, . . . , N;training a boost classifier ƒ
T(x)=Σ
r=1Mβ
rhr(x) to classify an image belonging to the target domain and represented by a feature vector x, the training using the aligned target domain training set DT and plurality of source domain training sets DS1 , . . . , DSN , the training comprising applying an adaptive boosting (AdaBoost) algorithm to generate the base classifiers hr(x) and the base classifier weights β
r of the boost classifier ƒ
T (x), where r=1, M; andcomputing a feature vector xin representing the image to be labeled; and generating a label for the image to be labeled by operations including evaluating ƒ
T(xin)=Σ
r=1Mβ
rhr(xin);wherein the feature vector computing operations, the training operation, and the generating operation are performed by an electronic data processing device. - View Dependent Claims (11, 12, 13, 14, 15, 16, 17, 18, 19, 20)
-
-
21. A non-transitory storage medium storing instructions executable by an electronic data processing device to perform a camera-based object labeling method to label an object based on an image of the object acquired using a target camera, the camera-based object labeling method including the operations of:
-
training a boost classifier ƒ
T(x)=Σ
r=1Mβ
rhr(x) to classify an image acquired by the target camera and represented by a feature vector x, the training using a target domain training set DT comprising labeled feature vectors representing images acquired by the target camera and a plurality of source domain training sets DS1 , . . . , DSN where N≧
2 comprising labeled feature vectors representing images of objects acquired by cameras other than the target camera, the training comprising applying an adaptive boosting (AdaBoost) algorithm to generate the base classifiers hr(x) and the base classifier weights β
r of the boost classifier ƒ
T(x), wherein the AdaBoost algorithm includes r=1, . . . , M iterations and the rth iteration includes training a plurality of candidate base classifiers hrk(x) wherein each candidate base classifier hrk(x) is trained on a training set DT∪
DSk and selecting hr(x) from a set of previously trained candidate base classifiers; andcomputing a feature vector xin representing the image of the object; and generating a label for the object by evaluating ƒ
T(xin)=Σ
r=1Mβ
rhr(xin). - View Dependent Claims (22, 23, 24)
-
Specification