TY - GEN
T1 - Learning to recognize 3D objects with SNoW
AU - Yang, Ming Hsuan
AU - Roth, Dan
AU - Ahuja, Narendra
N1 - Publisher Copyright:
© Springer-Verlag Berlin Heidelberg 2000.
PY - 2000
Y1 - 2000
N2 - This paper describes a novel view-based learning algorithm for 3D object recognition from 2D images using a network of linear units. The SNoW learning architecture is a sparse network of linear functions over a pre-defined or incrementally learned feature space and is specifically tailored for learning in the presence of a very large number of features. We use pixel-based and edge-based representations in large scale object recognition experiments in which the performance of SNoW is compared with that of Support Vector Machines (SVMs) and nearest neighbor using the 100 objects in the Columbia Image Object Database (COIL-100). Experimental results show that the SNoW-based method outperforms the SVM-based system in terms of recognition rate and the computational cost involved in learning. Most importantly, SNoW's performance degrades more gracefully when the training data contains fewer views. The empirical results also provide insight into practical and theoretical considerations on view-based methods for 3D object recognition.
AB - This paper describes a novel view-based learning algorithm for 3D object recognition from 2D images using a network of linear units. The SNoW learning architecture is a sparse network of linear functions over a pre-defined or incrementally learned feature space and is specifically tailored for learning in the presence of a very large number of features. We use pixel-based and edge-based representations in large scale object recognition experiments in which the performance of SNoW is compared with that of Support Vector Machines (SVMs) and nearest neighbor using the 100 objects in the Columbia Image Object Database (COIL-100). Experimental results show that the SNoW-based method outperforms the SVM-based system in terms of recognition rate and the computational cost involved in learning. Most importantly, SNoW's performance degrades more gracefully when the training data contains fewer views. The empirical results also provide insight into practical and theoretical considerations on view-based methods for 3D object recognition.
UR - http://www.scopus.com/inward/record.url?scp=84944184907&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84944184907&partnerID=8YFLogxK
U2 - 10.1007/3-540-45054-8_29
DO - 10.1007/3-540-45054-8_29
M3 - Conference contribution
AN - SCOPUS:84944184907
SN - 3540676856
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 439
EP - 454
BT - Computer Vision - ECCV 2000 - 6th European Conference on Computer Vision, Proceedings
A2 - Vernon, David
PB - Springer
T2 - 6th European Conference on Computer Vision, ECCV 2000
Y2 - 26 June 2000 through 1 July 2000
ER -