Image Classification: An Integration of Randomization and Discrimination
in A Dense Feature Representation

Bangpeng Yao       Aditya Khosla       Kyunghee Kim      Li Fei-Fei


Introduction
The goal of our method is to identify the discriminative fine-grained image regions that distinguish different classes. To achieve this goal we sample image regions from a dense sampling space and use a random forest algorithm with discriminative classifiers. Each node of the tree of the random forest is trained and tested with fine-grained image patches combining the information from upstream nodes. We implemented each node of the tree with a discriminative SVM classifier, which makes the node a strong classifier.

PASCAL VOC Winner Prize
Our method achieves the best performance in 6 out of the 10 classes in the PASCAL VOC action classification challenge. The table below shows the average precision of our method for each action category.

jumping phoning playing
instrument		 reading riding
bike		 riding
horse		 running taking
photo		 using
computer		 walking
CAENLEAR_DSAL
62.1
39.7
60.5
33.6
80.8
83.6
80.3
23.2
53.4
50.2
CAENLEAR_HOBJ_DSAL
71.6
50.7
77.5
37.8
86.5
89.5
83.8
25.1
58.9
59.2
MISSOURI_SSLMF
58.8
36.8
48.5
30.6
81.5
83.0
78.5
21.3
50.7
53.8
NUDT_CONTEXT
65.9
41.5
57.4
34.7
88.8
90.2
87.9
25.7
54.5
59.5
NUDT_LL_SEMANTIC
66.3
41.3
53.9
35.2
88.8
90.0
87.6
25.5
53.7
58.2
WVU_SVM-PHOW
42.5
29.5
32.1
26.7
48.5
46.3
59.2
13.5
24.3
35.6
Our Method
66.0
41.0
60.0
41.5
90.0
92.1
86.6
28.8
62.0
65.9


Software

Paper
B. Yao*, A. Khosla*, and L. Fei-Fei. "Combining Randomization and Discrimination for Fine-Grained Image Categorization." IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Colorado Springs, CO, USA. June 21-25, 2011. [pdf] [slides] [poster] [bibtex](*-indicates equal contribution)


Contact
Please contact bangpeng@cs.stanford.edu or khosla@mit.edu if you have any questions.