To construct optimal training set for video annotation Export

@inproceedings{citeulike:3315675, abstract = {This paper exploits the criteria to optimize the training set construction for video annotation. Most existing learning-based semantic annotation approaches require a large training set to achieve good generalization capacity, in which a considerable amount of labor-intensively manual labeling is desirable. However, it is observed that the generalization capacity of a classifier highly depends on the geometrical distribution rather than the size of the training data. We argue that a training set which includes most temporal and spatial distribution of the whole data will achieve a satisfying performance even in the case of limited size of training set. In order to capture the geometrical distribution characteristics of a given video collection, we propose the following four metrics for constructing an optimal training set, including Salience Time Dispersiveness Spatial Dispersiveness and Diversity . Moreover, based on these metrics, we propose a set of optimization rules to capture the most distribution information of the whole data for a training set with a given size. Experimental results demonstrate that these rules are effective for training set construction for video annotation, and significantly outperform random training set selection as well.}, address = {New York, NY, USA}, author = {Tang, Jinhui and Song, Yan and Hua, Xian S. and Mei, Tao and Wu, Xiuqing}, booktitle = {MULTIMEDIA '06: Proceedings of the 14th annual ACM international conference on Multimedia}, citeulike-article-id = {3315675}, citeulike-linkout-0 = {http://portal.acm.org/citation.cfm?id=1180639.1180667}, citeulike-linkout-1 = {http://dx.doi.org/10.1145/1180639.1180667}, doi = {10.1145/1180639.1180667}, isbn = {1-59593-447-2}, keywords = {casam}, location = {Santa Barbara, CA, USA}, pages = {89--92}, posted-at = {2008-09-22 14:47:23}, priority = {2}, publisher = {ACM}, title = {To construct optimal training set for video annotation}, url = {http://dx.doi.org/10.1145/1180639.1180667}, year = {2006} }

TY - CONF ID - citeulike:3315675 L3 - citeulike-article-id:3315675 N2 - This paper exploits the criteria to optimize the training set construction for video annotation. Most existing learning-based semantic annotation approaches require a large training set to achieve good generalization capacity, in which a considerable amount of labor-intensively manual labeling is desirable. However, it is observed that the generalization capacity of a classifier highly depends on the geometrical distribution rather than the size of the training data. We argue that a training set which includes most temporal and spatial distribution of the whole data will achieve a satisfying performance even in the case of limited size of training set. In order to capture the geometrical distribution characteristics of a given video collection, we propose the following four metrics for constructing an optimal training set, including Salience Time Dispersiveness Spatial Dispersiveness and Diversity . Moreover, based on these metrics, we propose a set of optimization rules to capture the most distribution information of the whole data for a training set with a given size. Experimental results demonstrate that these rules are effective for training set construction for video annotation, and significantly outperform random training set selection as well. CY - Santa Barbara, CA, USA SN - 1-59593-447-2 AD - New York, NY, USA EP - 92 TI - To construct optimal training set for video annotation PB - ACM T2 - MULTIMEDIA '06: Proceedings of the 14th annual ACM international conference on Multimedia SP - 89 KW - casam AU - Tang, Jinhui AU - Song, Yan AU - Hua, Xian AU - Mei, Tao AU - Wu, Xiuqing PY - 2006/// UR - http://dx.doi.org/10.1145/1180639.1180667 ER -