On detecting the playing/non-playing activity of musicians in symphonic music videos
作者：
Highlights：
• We propose a semi-automatic annotation system for large symphonic orchestras videos.
• We leverage video redundancy, image clustering, and human annotation.
• Our method successfully deals with several intra-class variability issues.
• Human annotation effort reduced while maintaining high level of output quality.
• Comprehensive analysis of the impact of different modules on the overall performance.
摘要
•We propose a semi-automatic annotation system for large symphonic orchestras videos.•We leverage video redundancy, image clustering, and human annotation.•Our method successfully deals with several intra-class variability issues.•Human annotation effort reduced while maintaining high level of output quality.•Comprehensive analysis of the impact of different modules on the overall performance.
论文关键词：Cross-modal analysis,Music information retrieval,Human-object interaction,Diarization,Clustering
论文评审过程：Received 20 December 2014, Revised 30 May 2015, Accepted 21 September 2015, Available online 1 April 2016, Version of Record 1 April 2016.
论文官网地址：https://doi.org/10.1016/j.cviu.2015.09.009
