UC3M High Level Feature Extraction at TRECVID 2008
نویسندگان
چکیده
This paper describes experiments carried out by the UC3M team for TRECVID 2008 high-level feature extraction task. Being our first participation in TRECVID, our goal this year has been to develop a modular system to facilitate future developments and incorporation of new functionality (feature extraction and classification modules). We have basically carried out experiments with two different kinds of classification technologies, namely Support Vector Machines (SVMs) and MultiTask Learning (MTL), resulting in the following runs: • SVM baseline: Scheme with late fusion and SVM classifiers trained on the shot level, keyframe level, and region level low-level descriptors. • SVM with bal-SIFT: Similar to our SVM baseline, but incorporating a different kind of region descriptors using a balanced codeword. • Bagging of SVM: Rather than using individual SVM classifiers, multi-net bagging systems are used. • MLP-STL: Scheme with late fusion and Multi Layer Perceptrons (MLP) trained individually for each high-level concept. This run serves as a baseline for the two following Multi-Task systems. • Selective MLP-MTL: Incorporates Multi-Task Learning, assigning all tasks the same importance. • Selective MLP-MTL with priority: Multi-Task Learning scheme, with focus on the individual performance of each category. The six submitted runs have achieved very similar performance in terms of average InfAP, with results which are slightly better or slightly worse than the median of all submitted runs. ∗This work has been supported by Alcatel-Lucent (Spain) and the Spanish Ministry of Science and Innovation under the i3Media CENIT Research Contract.
منابع مشابه
UC3M AT TRECVID 2010 Semantic Indexing Task
This paper describes the experiments carried out by the UC3M team for the TRECVID 2010 high-level feature extraction task. In our previous participations in TRECVID, we have developed a modular system to facilitate the testing of several functionalities. This year we have selected a simple system configuration and we have added some elements expected to provide an additional advantage. For inst...
متن کاملBilkent University Multimedia Database Group at TRECVID 2008
Bilkent University Multimedia Database Group (BILMDG) participated in two tasks at TRECVID 2008: content-based copy detection (CBCD) and high-level feature extraction (FE). Mostly MPEG-7 [1] visual features, which are also used as low-level features in our MPEG-7 compliant video database management system, are extracted for these tasks. This paper discusses our approaches in each task.
متن کاملIRIM at TRECVID 2008: High Level Feature Extraction
The IRIM group is a consortium of French teams working on Multimedia Indexing and Retrieval. This paper describes our participation to the TRECVID 2008 High Level Features detection task. We evaluated several fusion strategies and especially rank fusion. Results show that including as many low-level and intermediate features as possible is the best strategy, that SIFT features are very importan...
متن کاملXJTU at TRECVID2008 High-Level Feature Extraction
In this paper, we present our experiments in TRECVID 2008 about High-Level feature extraction task. This is the first year for our participation in TRECVID, our system adopts some popular approaches that other workgroups proposed before. We proposed 2 advanced low-level features NEW Gabor texture descriptor and the Compact-SIFT Codeword histogram. Our system applied well-known LIBSVM to train t...
متن کامل